● B2B offer — Marketing & content
Copywriting assistant: your content, your brand tone
Writing an article, a product page or a newsletter takes less inspiration than method: going back to the documentation, respecting the template, holding the editorial line. Your agent produces a first version from your own documents and your brand tone, ready to work on. Hosted in France — local inference or an isolated resource — your documentation feeds no third-party model. The marketing manager keeps the final word.
Hosted in France
Documentation never exposed
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Copywriting assistant · hosted in France
Prepare a first draft article on our new offer, from the product page and the press pack.
First version written from the product page and the press pack supplied, in the "blog article" template.
Brand tone applied: short sentences, formal address, no superlatives.
The statements only take up what appears in your documents — no data added.
✎ Action · draft to work on, nothing is published
Brand tone applied: short sentences, formal address, no superlatives.
The statements only take up what appears in your documents — no data added.
✎ Action · draft to work on, nothing is published
I need a short version for the newsletter.
Short version derived from the same content, newsletter format, with a hook and a link to the full article.
No new argument introduced: it is a shortening, not a rewrite.
✎ Action · adaptation ready to read over
No new argument introduced: it is a shortening, not a rewrite.
✎ Action · adaptation ready to read over
Local inference · no data outside the EU
Content hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The marketing manager decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent marketing copywriting agent produces first versions — articles, product pages, newsletters, sales arguments — from your templates, your brand tone and your internal documentation. It looks nowhere other than in what you give it. It runs on local inference or is hosted in France: your unpublished content and your reference documents stay with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Live within a few weeks,.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
content formats ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client: how large the gain is on your volume of editorial production is confirmed by a pilot.
The context
What does an AI agent bring to your editorial production?
The first draft is the longest and least rewarding part of writing. But a consumer tool writes outside your editorial line, and learns from what you give it.
! The issue
Every format means going back to the documentation, finding the right template and holding a consistent editorial line — work of retrieval more than of writing. Yet a consumer assistant produces text outside your tone, often inaccurate, and above all absorbs your unpublished content, your forthcoming launches and your internal documentation into a third party subject to the Cloud Act.
✓ Our answer
A writing AI is only of interest if it is sovereign and bounded by your sources. Local inference or an isolated resource hosted in France, writing from your documents alone, nothing published automatically: the first draft is supplied, the editorial line stays yours. The agent invents no argument you have not written.
The decisive point
Your editorial line and your reference documents: sovereignty & compliance
Your unpublished content reveals your commercial roadmap. Here is how the architecture of our agents protects it, document by document.
Local inference
The agent can run on a machine belonging to your organisation: no unpublished content leaves the network, no documentation passes through a public cloud.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your content and your internal documentation: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your editorial line and your reference documents, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your brand and its documentation.
Templates and brand tone respected
Your templates and your editorial guidelines are applied as they stand; encryption, role-based access and logging of everything produced.
AI Act: governed deployment
An agent strictly in support; no content published automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For launches under embargo or content tied to a financial transaction, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
5 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The company in this demonstration
Fictional companyHydralis — French manufacturer of pumps and water treatment systems for industry
- Sector
- Industry — dosing pumps and water treatment systems for food processing and local authorities
- Headcount
- 140 staff, including 3 in marketing (1 marketing manager, 1 content writer, 1 apprentice), 11 in the engineering office and 9 field sales staff
- Market
- 600 industrial customers in France and Europe — sales cycle of 4 to 9 months, technical decision before purchasing decision
- Volume
- 340 catalogue references, 4 articles and 2 newsletters a month, 1 range launch a quarter
- Tools in place
- The site's content manager, the emailing tool, the catalogue PIM and a 9-year file server — manuals, test reports, press kits, visit reports; the agent plugs into them, nothing is replaced
- Who decides
- The marketing manager approves and publishes every piece of content; the engineering office signs off every technical claim; the sales director settles the sales scripts
- Room for improvement
- An article takes 11 days from brief to publication; 214 of the 340 product pages carry only 40 words lifted from the 2019 catalogue; the newsletter goes out late every other month
Hydralis lacks neither material nor rigour: it has nine years of manuals, test reports and visit reports, and an engineering office that reviews everything. What it lacks is the time to go back to that material, find the right template and hold a consistent editorial line across six formats. The agent runs with local inference on a machine at the plant and reads the file server, the PIM and the site: it produces sourced first drafts, marketing keeps the pen and publishes. The exchanges below span a quarter, from writing down the brand tone to the closing review.
This company, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real client.
Copywriting assistant · hosted in France
I read nine years of your file server last night — 6,400 files — and wrote down what nobody here had written: your editorial line, as six templates and a tone guide.
Brand tone is the way of writing that belongs to a house: its sentence length, its register, its units, what it allows itself and what it rules out. Your two writers carry it in their heads. It was written down nowhere, so it could not be passed on.
The gap I measured, and it is the gap that decides the gain: I compared the successive versions of your last 22 published pieces. 7 corrections out of 10 were about form — the tone, the template, going back to the documentation. Only 3 out of 10 were about substance. Your content writer is not reviewing ideas, she is re-tuning a voice — the one job that can be written down once and for all.
What the guide contains, and all of it can be checked against your own texts:
· 18-word sentences on average, 81 % active voice, formal address throughout.
· Units always in SI, with the common equivalent in brackets — you have done it since 2019 without ever writing it down.
· No superlatives. Across your 180 approved pieces, « the best » does not appear once.
· The word « solution » is absent from your last 40 pieces — you say « pump », « system », « line ». It is the most discriminating trait of all, and no brand guide ever thinks to write it down.
· No promise of results without test conditions cited: a constraint of your trade, held by your writers without anyone reminding them.
The six templates I reconstructed — a template is the imposed skeleton of a format: the order of the parts, what each contains, its length: technical article, product page, newsletter, sales script, application note, launch release. These are the six you already use; I have simply written them down.
What that gives you on the next request: a first draft arrives already in the right tone and already in the right template. Your content writer stops rewriting and starts arbitrating.
What I propose: that the marketing manager read back the guide and the six templates. Every trait can be corrected with a word: it is your own material, I have only made it explicit — and there is nothing to take on trust, everything checks against a text you have published. editorial-guide_tone-and-6-templates.pdf18 words a sentence, 81 % active voice, 25 banned words
⛓ Sourced · file server, 6,400 documents over 9 years, 180 published pieces, successive versions of the last 22
Brand tone is the way of writing that belongs to a house: its sentence length, its register, its units, what it allows itself and what it rules out. Your two writers carry it in their heads. It was written down nowhere, so it could not be passed on.
The gap I measured, and it is the gap that decides the gain: I compared the successive versions of your last 22 published pieces. 7 corrections out of 10 were about form — the tone, the template, going back to the documentation. Only 3 out of 10 were about substance. Your content writer is not reviewing ideas, she is re-tuning a voice — the one job that can be written down once and for all.
What the guide contains, and all of it can be checked against your own texts:
· 18-word sentences on average, 81 % active voice, formal address throughout.
· Units always in SI, with the common equivalent in brackets — you have done it since 2019 without ever writing it down.
· No superlatives. Across your 180 approved pieces, « the best » does not appear once.
· The word « solution » is absent from your last 40 pieces — you say « pump », « system », « line ». It is the most discriminating trait of all, and no brand guide ever thinks to write it down.
· No promise of results without test conditions cited: a constraint of your trade, held by your writers without anyone reminding them.
The six templates I reconstructed — a template is the imposed skeleton of a format: the order of the parts, what each contains, its length: technical article, product page, newsletter, sales script, application note, launch release. These are the six you already use; I have simply written them down.
What that gives you on the next request: a first draft arrives already in the right tone and already in the right template. Your content writer stops rewriting and starts arbitrating.
What I propose: that the marketing manager read back the guide and the six templates. Every trait can be corrected with a word: it is your own material, I have only made it explicit — and there is nothing to take on trust, everything checks against a text you have published. editorial-guide_tone-and-6-templates.pdf18 words a sentence, 81 % active voice, 25 banned words
⛓ Sourced · file server, 6,400 documents over 9 years, 180 published pieces, successive versions of the last 22
Our manuals, test reports and press kits — nobody can find anything. What do you do with that server?
I index it on the content of the files, not on their names — and every time I tell you which version prevails.
The case that sums it up: you asked me for the DP-40 pump manual last week. Eleven files for it exist on the server. The one that prevails is called DP40_manual_VF_v3_OK_final.pdf, it is dated 18/02/2026 and carries revision index D. The other ten are earlier indexes. I do not read the file name: I read the revision index, the signature date and the product reference inside the document.
What the index found, and what needs a decision:
· 1,940 files out of 6,400 have no usable name — exactly the ones you hunt for on a Friday evening.
· 240 version duplicates, with the risk of starting from an obsolete index.
· 3 obsolete manuals are still linked from your product pages. A customer downloading today receives a document whose index is no longer current. Those need handling this week: your customers read them, not your teams.
· 41 test reports later than 2024 that have never been used in any marketing content. That is your seam of fresh, verifiable arguments, and it is asleep.
The time this moves: finding the source documentation for a piece took 28 % of production time. It now takes 8 %. Your quarterly reports put document hunting at 62 hours — more than a week and a half of work for a team of three.
What indexing does not do, and it matters: it renames nothing, moves nothing, deletes nothing. Your existing paths keep working, and if you stop the monthly pass, the index already built stays.
What I propose: that I run once a month over recent deposits to record the product reference, the index and the signature date. And that the 3 obsolete manuals be replaced first — I have prepared the list of pages concerned and the current file for each. indexed-documentation_6400-files.pdf1,940 with no usable name · 3 obsolete manuals live
⛓ Sourced · file server, content of documents rather than their names, product page links, quarterly reports
The case that sums it up: you asked me for the DP-40 pump manual last week. Eleven files for it exist on the server. The one that prevails is called DP40_manual_VF_v3_OK_final.pdf, it is dated 18/02/2026 and carries revision index D. The other ten are earlier indexes. I do not read the file name: I read the revision index, the signature date and the product reference inside the document.
What the index found, and what needs a decision:
· 1,940 files out of 6,400 have no usable name — exactly the ones you hunt for on a Friday evening.
· 240 version duplicates, with the risk of starting from an obsolete index.
· 3 obsolete manuals are still linked from your product pages. A customer downloading today receives a document whose index is no longer current. Those need handling this week: your customers read them, not your teams.
· 41 test reports later than 2024 that have never been used in any marketing content. That is your seam of fresh, verifiable arguments, and it is asleep.
The time this moves: finding the source documentation for a piece took 28 % of production time. It now takes 8 %. Your quarterly reports put document hunting at 62 hours — more than a week and a half of work for a team of three.
What indexing does not do, and it matters: it renames nothing, moves nothing, deletes nothing. Your existing paths keep working, and if you stop the monthly pass, the index already built stays.
What I propose: that I run once a month over recent deposits to record the product reference, the index and the signature date. And that the 3 obsolete manuals be replaced first — I have prepared the list of pages concerned and the current file for each. indexed-documentation_6400-files.pdf1,940 with no usable name · 3 obsolete manuals live
⛓ Sourced · file server, content of documents rather than their names, product page links, quarterly reports
Our launches aren't public before the announcement, and our test reports even less so. Where exactly does all this go?
It does not go anywhere. I run with local inference on a machine at the plant, and your file server is copied nowhere.
Local inference means the model computes on your machine: the text of a test report or of an embargoed release never crosses an outside network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France, under French law, dedicated to Hydralis — no pooling with another manufacturer.
What that protects, document by document:
· Your unpublished content. The Neris range release has existed for six weeks and the announcement is on 14 October. It tells your commercial roadmap to whoever reads it.
· Your 41 recent test reports. Those are measurements your competitors do not have and would pay for.
· Your 340 visit reports for the year, naming your customers, their installations and their objections.
· And the very fact that you are preparing a launch, which can be inferred from a simple usage log.
How it is held: encryption in transit and at rest; role-based access — rights follow the job: the apprentice opens the templates and published content, not the test reports nor the embargoed documents; logging of every production — which template, which sources, who asked, who approved; hosting in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
And none of your data trains a model: what I learn from Hydralis serves Hydralis. Nothing you entrust to me can surface elsewhere, because nothing goes there.
The figure that makes this commercial rather than technical: two of your 2026 public tenders scored data hosting out of 5 points, and your food-processing customer in Nantes prohibits, in its framework contract, any transfer outside the European Union, including for a writing tool. You were able to tick the box with no reservation and no clause to negotiate.
What I propose: that I keep up to date the technical sheet your customers ask for at renewal — hosting, sub-processors, retention periods, who accesses what. It is asked for once a year and hunted for over three days. technical-framework_where-your-content-lives.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access and production logs, customer framework contracts, 2026 tender rules
Local inference means the model computes on your machine: the text of a test report or of an embargoed release never crosses an outside network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France, under French law, dedicated to Hydralis — no pooling with another manufacturer.
What that protects, document by document:
· Your unpublished content. The Neris range release has existed for six weeks and the announcement is on 14 October. It tells your commercial roadmap to whoever reads it.
· Your 41 recent test reports. Those are measurements your competitors do not have and would pay for.
· Your 340 visit reports for the year, naming your customers, their installations and their objections.
· And the very fact that you are preparing a launch, which can be inferred from a simple usage log.
How it is held: encryption in transit and at rest; role-based access — rights follow the job: the apprentice opens the templates and published content, not the test reports nor the embargoed documents; logging of every production — which template, which sources, who asked, who approved; hosting in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
And none of your data trains a model: what I learn from Hydralis serves Hydralis. Nothing you entrust to me can surface elsewhere, because nothing goes there.
The figure that makes this commercial rather than technical: two of your 2026 public tenders scored data hosting out of 5 points, and your food-processing customer in Nantes prohibits, in its framework contract, any transfer outside the European Union, including for a writing tool. You were able to tick the box with no reservation and no clause to negotiate.
What I propose: that I keep up to date the technical sheet your customers ask for at renewal — hosting, sub-processors, retention periods, who accesses what. It is asked for once a year and hunted for over three days. technical-framework_where-your-content-lives.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access and production logs, customer framework contracts, 2026 tender rules
We're launching the Neris range on 14 October. We need the technical article. It usually takes eleven days.
The first draft is ready: 1,400 words, in your « technical article » template, in your tone, and every claim links back to the document it comes from.
What I read to write it: the PIM sheets for the 6 Neris references, the index A manual of 04/09/2026, the 3 test reports from March and June, the press kit, and 17 visit reports where a customer describes precisely the problem this range solves.
The template, held part by part: a 2-sentence hook · the problem in field terms · the technical principle · the measurements, with their test conditions · the use cases · the operating limits · the call to action towards a sizing study. Seven parts, those of your articles whose structure was never touched.
What that means for the schedule: eleven days from brief to publication, of which seven were writing and revision. The first draft is here this morning. Count three days in total: one for your review, two for the engineering office's sign-off.
The time this moves: the first draft of an article took 50 % of production time. It now takes 18 % — and the remaining 18 % is your review, the part that makes the difference between a correct text and a Hydralis text.
Four claims are highlighted in yellow, and I am asking you not to publish without them: those are the four whose source I could not establish within three years, or which compare Neris with a competing product. They need a signature from the engineering office — not a review, a signature.
What I propose next: once the article is approved, I derive the newsletter, the sales script for your 9 field staff and the 6 product pages of the range. Everything comes from the same base: if you correct the article, I propagate without your having to ask. neris-article_1400-word-first-draft.pdf7 template parts · 31 sourced claims
⛓ Sourced · PIM sheets for the 6 references, index A manual of 04/09/2026, 3 test reports, press kit, 17 visit reports
What I read to write it: the PIM sheets for the 6 Neris references, the index A manual of 04/09/2026, the 3 test reports from March and June, the press kit, and 17 visit reports where a customer describes precisely the problem this range solves.
The template, held part by part: a 2-sentence hook · the problem in field terms · the technical principle · the measurements, with their test conditions · the use cases · the operating limits · the call to action towards a sizing study. Seven parts, those of your articles whose structure was never touched.
What that means for the schedule: eleven days from brief to publication, of which seven were writing and revision. The first draft is here this morning. Count three days in total: one for your review, two for the engineering office's sign-off.
The time this moves: the first draft of an article took 50 % of production time. It now takes 18 % — and the remaining 18 % is your review, the part that makes the difference between a correct text and a Hydralis text.
Four claims are highlighted in yellow, and I am asking you not to publish without them: those are the four whose source I could not establish within three years, or which compare Neris with a competing product. They need a signature from the engineering office — not a review, a signature.
What I propose next: once the article is approved, I derive the newsletter, the sales script for your 9 field staff and the 6 product pages of the range. Everything comes from the same base: if you correct the article, I propagate without your having to ask. neris-article_1400-word-first-draft.pdf7 template parts · 31 sourced claims
⛓ Sourced · PIM sheets for the 6 references, index A manual of 04/09/2026, 3 test reports, press kit, 17 visit reports
Where did you get the figures? That's where we get caught out.
Every claim in the article carries its document, its date and its page. Thirty-one claims, thirty-one sources — and four I refuse to let through without a signature.
Line-by-line sourcing means every sentence that asserts something links back to the document it comes from, with its date. Without the date, a claim that was true in 2021 gets republished as if it still were.
Three examples of how it reads:
· « Nominal flow rate is 40 litres per hour at 6 bar » → PIM sheet DP-40, updated 04/09/2026. Taken as is.
· « Consumption falls by 22 % against the previous generation » → test report of 12/03/2026, bench no. 2, page 7, test conditions cited in the article itself. A stated reduction without its test conditions is indefensible in front of an engineer, and your customers are engineers.
· « Compatible with food processing installations » → your conformity certificate of 22/01/2026, whose exact wording I reused rather than rephrasing.
The four highlighted, and why:
· « Up to 15 years of service life » — the source is an internal document from 2021, five years old. Past three years I escalate the claim systematically, whatever it is.
· « The quietest in its class » — that sentence comes from the press kit. None of your test reports compares Neris with a competing product, and your guide bans superlatives. I did not write it into the body of the article; I flag it because it is in the press kit and it will come out of there.
· Two figure comparisons with a named competitor, from an internal table with no protocol attached.
And the route that gets the same result, because I do not stop at the flag: your test report of 12/03/2026 measures the noise level at 48 dB(A) at one metre, protocol cited. « 48 dB(A) measured at one metre » is stronger than « the quietest »: it is verifiable, it can be quoted in a specification, and a public buyer can carry it into a scoring table. I have written both sentences in that form — they need only the engineering office's signature.
What I propose: that the 41 test reports later than 2024 enter the index of citable sources. You have what it takes to replace most of your atmosphere phrases with measurements, and a measurement cannot be disputed. sourcing_31-claims-4-highlighted.pdfEvery sentence, its document, its date · 4 to be signed
⛓ Sourced · PIM sheets, test report of 12/03/2026, certificate of 22/01/2026, press kit, internal document from 2021
Line-by-line sourcing means every sentence that asserts something links back to the document it comes from, with its date. Without the date, a claim that was true in 2021 gets republished as if it still were.
Three examples of how it reads:
· « Nominal flow rate is 40 litres per hour at 6 bar » → PIM sheet DP-40, updated 04/09/2026. Taken as is.
· « Consumption falls by 22 % against the previous generation » → test report of 12/03/2026, bench no. 2, page 7, test conditions cited in the article itself. A stated reduction without its test conditions is indefensible in front of an engineer, and your customers are engineers.
· « Compatible with food processing installations » → your conformity certificate of 22/01/2026, whose exact wording I reused rather than rephrasing.
The four highlighted, and why:
· « Up to 15 years of service life » — the source is an internal document from 2021, five years old. Past three years I escalate the claim systematically, whatever it is.
· « The quietest in its class » — that sentence comes from the press kit. None of your test reports compares Neris with a competing product, and your guide bans superlatives. I did not write it into the body of the article; I flag it because it is in the press kit and it will come out of there.
· Two figure comparisons with a named competitor, from an internal table with no protocol attached.
And the route that gets the same result, because I do not stop at the flag: your test report of 12/03/2026 measures the noise level at 48 dB(A) at one metre, protocol cited. « 48 dB(A) measured at one metre » is stronger than « the quietest »: it is verifiable, it can be quoted in a specification, and a public buyer can carry it into a scoring table. I have written both sentences in that form — they need only the engineering office's signature.
What I propose: that the 41 test reports later than 2024 enter the index of citable sources. You have what it takes to replace most of your atmosphere phrases with measurements, and a measurement cannot be disputed. sourcing_31-claims-4-highlighted.pdfEvery sentence, its document, its date · 4 to be signed
⛓ Sourced · PIM sheets, test report of 12/03/2026, certificate of 22/01/2026, press kit, internal document from 2021
And the template — do you invent it? Because we have our habits and they aren't negotiable.
I invent nothing: the six templates come from your own pieces that were approved WITHOUT structural changes. They are your habits, written down.
How I established them: I took your 180 published pieces and kept the 54 whose structure did not move between first draft and publication. Those describe your habit better than any statement of intent: they are the ones nobody felt like reorganising.
What each template fixes: the order of the parts, what each contains, its length in characters, what is mandatory and what is optional, and the statements to be reproduced verbatim.
The six, with their use:
· Technical article — 1,200 to 1,600 words, 7 parts, call to action towards a sizing study.
· Product page — 180 to 260 words, characteristics from the PIM, mandatory operating limits.
· Newsletter — 320 words maximum, a single main subject, no claim that is not already in the source content.
· Sales script — benefit, proof, objection, sourced answer.
· Application note — an installation problem, a set-up, measured values.
· Launch release — announcement date, executive quote, « about us » box reproduced verbatim.
What I propose to add, and the figure that justifies it: an eighth part to the « technical article » template, called « questions from the field ». Across your 340 visit reports, three questions come up in more than 40 % of meetings — maintenance, compatibility with existing equipment, and time to commissioning. Your articles never answer them, and your field staff answer them at every visit.
What it is worth: an answer written once on the site does the work your 9 field staff redo nine times a week. And a prospect who finds the answer before calling arrives with a higher-level question — it is your 4-to-9-month cycle that shortens, not your traffic.
The decision is editorial, so it is yours: I have written the section for the Neris article, it runs to 210 words, it can be added or removed with a word and the rest of the article does not move. six-templates_drawn-from-your-content.pdf54 untouched pieces · 6 formats ready to deploy
⛓ Sourced · 180 published pieces, 54 approved without structural change, 340 visit reports
How I established them: I took your 180 published pieces and kept the 54 whose structure did not move between first draft and publication. Those describe your habit better than any statement of intent: they are the ones nobody felt like reorganising.
What each template fixes: the order of the parts, what each contains, its length in characters, what is mandatory and what is optional, and the statements to be reproduced verbatim.
The six, with their use:
· Technical article — 1,200 to 1,600 words, 7 parts, call to action towards a sizing study.
· Product page — 180 to 260 words, characteristics from the PIM, mandatory operating limits.
· Newsletter — 320 words maximum, a single main subject, no claim that is not already in the source content.
· Sales script — benefit, proof, objection, sourced answer.
· Application note — an installation problem, a set-up, measured values.
· Launch release — announcement date, executive quote, « about us » box reproduced verbatim.
What I propose to add, and the figure that justifies it: an eighth part to the « technical article » template, called « questions from the field ». Across your 340 visit reports, three questions come up in more than 40 % of meetings — maintenance, compatibility with existing equipment, and time to commissioning. Your articles never answer them, and your field staff answer them at every visit.
What it is worth: an answer written once on the site does the work your 9 field staff redo nine times a week. And a prospect who finds the answer before calling arrives with a higher-level question — it is your 4-to-9-month cycle that shortens, not your traffic.
The decision is editorial, so it is yours: I have written the section for the Neris article, it runs to 210 words, it can be added or removed with a word and the rest of the article does not move. six-templates_drawn-from-your-content.pdf54 untouched pieces · 6 formats ready to deploy
⛓ Sourced · 180 published pieces, 54 approved without structural change, 340 visit reports
214 of our 340 pages carry only 40 words lifted from the 2019 catalogue. We know it's a problem, we've never had the time.
The 214 are rewritten and waiting for you, each one from the real characteristics of the reference — and not one argument you have not written down somewhere.
Where each sentence comes from: the PIM — the database holding each reference's characteristics: flow rate, pressure, materials, fittings, conformities — for the values; the current manual for operating limits; the test reports for any performance figure. Nothing else.
What each page now contains, against 40 words before: 180 to 260 words, in the « product page » template: what the reference is for, in which installation, the characteristics exactly as the PIM holds them, the operating limits — mandatory at Hydralis, and missing from 187 of the 214 original pages — and the link to the current manual.
What I did not do, and it is what makes the rest usable: I filled no gaps. When a characteristic is missing from the PIM, the line stays empty and reads « to be completed » rather than carrying a plausible value. A guessed value is indistinguishable from a measured one once it is written down — and it is the guessed one that ends up in a customer specification.
· Complete pages, ready to review: 176.
· Pages with at least one « to be completed » line: 38, and I tell you exactly which.
What I found along the way, and it is worth more than the pages: 38 discrepancies between what the PIM says and what the manual for the same reference says. I raised them, I settled none.
The time this moves: at your usual pace, rewriting 214 pages meant more than a full-time quarter for your content writer. What is left is the review — count 8 minutes a page, that is around four days for the 176 complete ones.
What I propose: start with the 26 references that carry 60 % of your revenue. They are at the front of the batch, reviewed and ready. 214-product-pages_rewritten.pdf176 complete, 38 to complete · 0 guessed value
⛓ Sourced · PIM for the 340 references, current manuals, test reports, 2019 catalogue
Where each sentence comes from: the PIM — the database holding each reference's characteristics: flow rate, pressure, materials, fittings, conformities — for the values; the current manual for operating limits; the test reports for any performance figure. Nothing else.
What each page now contains, against 40 words before: 180 to 260 words, in the « product page » template: what the reference is for, in which installation, the characteristics exactly as the PIM holds them, the operating limits — mandatory at Hydralis, and missing from 187 of the 214 original pages — and the link to the current manual.
What I did not do, and it is what makes the rest usable: I filled no gaps. When a characteristic is missing from the PIM, the line stays empty and reads « to be completed » rather than carrying a plausible value. A guessed value is indistinguishable from a measured one once it is written down — and it is the guessed one that ends up in a customer specification.
· Complete pages, ready to review: 176.
· Pages with at least one « to be completed » line: 38, and I tell you exactly which.
What I found along the way, and it is worth more than the pages: 38 discrepancies between what the PIM says and what the manual for the same reference says. I raised them, I settled none.
The time this moves: at your usual pace, rewriting 214 pages meant more than a full-time quarter for your content writer. What is left is the review — count 8 minutes a page, that is around four days for the 176 complete ones.
What I propose: start with the 26 references that carry 60 % of your revenue. They are at the front of the batch, reviewed and ready. 214-product-pages_rewritten.pdf176 complete, 38 to complete · 0 guessed value
⛓ Sourced · PIM for the 340 references, current manuals, test reports, 2019 catalogue
The 38 discrepancies between the PIM and the manuals — how do you settle them?
I do not settle them today because no rule lets me — but I settle them the moment you write one, and I propose one that resolves 31 of the 38 on its own.
The 38, classified:
· 12 unit discrepancies — the PIM says « 40 l/h », the manual says « 0.67 l/min ». Same value, two notations. No risk, except for a hurried reader.
· 9 value discrepancies — the PIM and the manual do not give the same figure. In 7 cases out of 9 the manual is later than the PIM. These are the dangerous ones: both are already live.
· 7 obsolete references — the PIM cites a fitting withdrawn from the catalogue in 2023.
· 10 divergent normative statements — the manual cites a standard in one version, the PIM in another. Those cannot be settled by a general rule: they commit a conformity.
The rule I propose, and its measured effect: « for a technical characteristic, the most recent source signed by the engineering office prevails; the older source is corrected, not ignored. » If you write it, 31 of the 38 discrepancies resolve with no human arbitration — the 12 units, the 9 values, the 7 obsolete references and 3 statements. That leaves 7 normative statements for the engineering office, and seven files are dealt with in one meeting.
What I would then do, and do not do today: apply the rule, write the page with the retained value, and flag to your engineering office the source that still needs correcting — because a discrepancy fixed in the product page remains a discrepancy in the manual the customer downloads.
Choosing between two versions of a standard does stay a signature — and that is what protects you: a conformity is certified, it is not inferred from a file date. I put both versions in front of you, their dates, the pages affected and the exact discrepancy; your engineering office settles it on a file that is ready, and it is their signature that makes the page defensible in front of a public buyer.
What I propose: that the rule be written once, in your guide, and hold for all 340 references. Written once, it spares you this quarter's 38 arbitrations and every one after that. 38-discrepancies_pim-versus-manual.pdf31 resolved by one rule · 7 for the engineering office
⛓ Sourced · PIM/manual comparison across 340 references, signature dates, fittings catalogue
The 38, classified:
· 12 unit discrepancies — the PIM says « 40 l/h », the manual says « 0.67 l/min ». Same value, two notations. No risk, except for a hurried reader.
· 9 value discrepancies — the PIM and the manual do not give the same figure. In 7 cases out of 9 the manual is later than the PIM. These are the dangerous ones: both are already live.
· 7 obsolete references — the PIM cites a fitting withdrawn from the catalogue in 2023.
· 10 divergent normative statements — the manual cites a standard in one version, the PIM in another. Those cannot be settled by a general rule: they commit a conformity.
The rule I propose, and its measured effect: « for a technical characteristic, the most recent source signed by the engineering office prevails; the older source is corrected, not ignored. » If you write it, 31 of the 38 discrepancies resolve with no human arbitration — the 12 units, the 9 values, the 7 obsolete references and 3 statements. That leaves 7 normative statements for the engineering office, and seven files are dealt with in one meeting.
What I would then do, and do not do today: apply the rule, write the page with the retained value, and flag to your engineering office the source that still needs correcting — because a discrepancy fixed in the product page remains a discrepancy in the manual the customer downloads.
Choosing between two versions of a standard does stay a signature — and that is what protects you: a conformity is certified, it is not inferred from a file date. I put both versions in front of you, their dates, the pages affected and the exact discrepancy; your engineering office settles it on a file that is ready, and it is their signature that makes the page defensible in front of a public buyer.
What I propose: that the rule be written once, in your guide, and hold for all 340 references. Written once, it spares you this quarter's 38 arbitrations and every one after that. 38-discrepancies_pim-versus-manual.pdf31 resolved by one rule · 7 for the engineering office
⛓ Sourced · PIM/manual comparison across 340 references, signature dates, fittings catalogue
And the mandatory statements? We sell to local authorities and to food processing, we can't get those wrong.
They are reproduced verbatim from your own models, never rephrased — and I found 26 pages where one was missing.
Why verbatim: a mandatory statement is a text whose exact wording matters. Rephrasing it, even better, changes its scope. I copy it character for character from the model your engineering office approved, and I tell you which model it comes from.
The four families your catalogue carries:
· Food contact conformity — on the 96 references sold to food processing, with the reference to the certificate and its date.
· Regulatory marking and declaration of performance, with the index of the document in force.
· Operating limits — temperature, pressure, prohibited fluids. At Hydralis they are mandatory on every page: that is an internal rule from 09/2021 and I apply it as such.
· Installation safety statements, taken from chapter 3 of the manual.
What the check found: 26 live pages where at least one statement is missing, including 9 on food processing references — those are the ones I propose handling first, because they address the most demanding market.
And what it opens up, beyond conformity: your last two public tender rules asked for the technical sheet and the conformity certificate in the bid file. With the 340 pages up to date and their statements checked, that file is assembled in an hour instead of two days — and you can bid for consultations you were letting pass for lack of assembly time.
What I propose: that I check the presence of the four families at every production, systematically, and escalate any statement whose reference document has changed index. That check costs nothing: it runs as the page is written, not six months later.
⛓ Sourced · models approved by the engineering office, internal rule of 09/2021, 340 live pages, 2026 tender rules
Why verbatim: a mandatory statement is a text whose exact wording matters. Rephrasing it, even better, changes its scope. I copy it character for character from the model your engineering office approved, and I tell you which model it comes from.
The four families your catalogue carries:
· Food contact conformity — on the 96 references sold to food processing, with the reference to the certificate and its date.
· Regulatory marking and declaration of performance, with the index of the document in force.
· Operating limits — temperature, pressure, prohibited fluids. At Hydralis they are mandatory on every page: that is an internal rule from 09/2021 and I apply it as such.
· Installation safety statements, taken from chapter 3 of the manual.
What the check found: 26 live pages where at least one statement is missing, including 9 on food processing references — those are the ones I propose handling first, because they address the most demanding market.
And what it opens up, beyond conformity: your last two public tender rules asked for the technical sheet and the conformity certificate in the bid file. With the 340 pages up to date and their statements checked, that file is assembled in an hour instead of two days — and you can bid for consultations you were letting pass for lack of assembly time.
What I propose: that I check the presence of the four families at every production, systematically, and escalate any statement whose reference document has changed index. That check costs nothing: it runs as the page is written, not six months later.
⛓ Sourced · models approved by the engineering office, internal rule of 09/2021, 340 live pages, 2026 tender rules
The newsletter goes out late every other month. And when it does, it repeats the article, only worse.
Here it is, derived from the Neris article you have just approved: 320 words, a single subject, and not one claim that is not already in the article.
An adaptation is the reduction of already-approved content to another format. It is not a rewrite: if a new argument appears in the short version, nobody has reviewed it.
What I kept: the main argument — the 22 % consumption reduction with its test conditions —, one verifiable measurement, and the link to the full article.
What I cut, and I tell you what: the technical principle, the operating limits, two of the three use cases. You need to know what fell away: in a newsletter, what is missing matters as much as what remains.
What I did not add: no argument, no figure, no urgency device. A short version that gains an argument is a new text, and it goes back for review — which cancels exactly the time you thought you had saved.
The time this moves: adapting content into a newsletter took 35 % of production time. It now takes 10 %. In practice: the newsletter is ready the day the article is approved, not three weeks later.
What I propose so the delay stops structurally: that every long piece produce automatically its short adaptation as a draft the moment it is approved. Of your 24 annual newsletters, 11 went out late last year; 9 of those 11 were waiting on source content that was itself ready. That is not a writing problem, it is a sequencing problem — and a sequence is fixed once.
And if you correct the article afterwards: I rebuild the newsletter and flag what moved. The previous version is kept: you can go back to it. neris-newsletter_320-derived-words.pdfWhat is kept, what is cut, what is never added
⛓ Sourced · approved Neris article, history of the year's 24 newsletters, availability dates of source content
An adaptation is the reduction of already-approved content to another format. It is not a rewrite: if a new argument appears in the short version, nobody has reviewed it.
What I kept: the main argument — the 22 % consumption reduction with its test conditions —, one verifiable measurement, and the link to the full article.
What I cut, and I tell you what: the technical principle, the operating limits, two of the three use cases. You need to know what fell away: in a newsletter, what is missing matters as much as what remains.
What I did not add: no argument, no figure, no urgency device. A short version that gains an argument is a new text, and it goes back for review — which cancels exactly the time you thought you had saved.
The time this moves: adapting content into a newsletter took 35 % of production time. It now takes 10 %. In practice: the newsletter is ready the day the article is approved, not three weeks later.
What I propose so the delay stops structurally: that every long piece produce automatically its short adaptation as a draft the moment it is approved. Of your 24 annual newsletters, 11 went out late last year; 9 of those 11 were waiting on source content that was itself ready. That is not a writing problem, it is a sequencing problem — and a sequence is fixed once.
And if you correct the article afterwards: I rebuild the newsletter and flag what moved. The previous version is kept: you can go back to it. neris-newsletter_320-derived-words.pdfWhat is kept, what is cut, what is never added
⛓ Sourced · approved Neris article, history of the year's 24 newsletters, availability dates of source content
We also need the sales script for the 9 field staff. The last one is two years old and they don't open it any more.
It is written, and it does not start from the brochure: it starts from the objections your customers actually raise, taken from your 340 visit reports.
A sales script is the document the salesperson carries: for each benefit, the proof that supports it and the answer to the objection that follows.
Why the old one stopped being opened, and it is measurable: it handled 6 objections. Your visit reports show 14, and the 4 most frequent were not among them. A sales script that does not talk about what gets said in meetings is not read twice.
The four missing objections, with their frequency:
· « Time to commissioning is too long » — in 38 % of visits. Sourced answer: your average lead time across the last 60 installations, with the figure.
· « We already run another brand » — 31 %. Answer: the application note on fitting compatibility, which exists and which nobody cites.
· « Who handles maintenance? » — 27 %.
· « Prove the 22 % to me » — 19 %. Answer: the test report of 12/03/2026, its conditions, its bench. The easiest objection to handle and the one nobody had armed.
How each answer is built: one sentence of answer, the proof with its document and date, and the limit — what the answer does not cover. A salesperson who knows the limit of their argument does not get cornered on it; that is what separates a sales script from a brochure.
What I propose: that the script update itself with every new approved test report, and that I escalate to the sales director any objection appearing in more than 15 % of a quarter's visits. This quarter a new objection is already at 17 %: spare parts availability. It has no written answer anywhere in the company — and that is exactly the kind of subject that is easier to deal with at 17 % than at 40 %. neris-sales-script_14-real-objections.pdf4 major objections absent from the old one · dated proofs
⛓ Sourced · 340 visit reports, 2024 sales script, test report of 12/03/2026, last 60 installations
A sales script is the document the salesperson carries: for each benefit, the proof that supports it and the answer to the objection that follows.
Why the old one stopped being opened, and it is measurable: it handled 6 objections. Your visit reports show 14, and the 4 most frequent were not among them. A sales script that does not talk about what gets said in meetings is not read twice.
The four missing objections, with their frequency:
· « Time to commissioning is too long » — in 38 % of visits. Sourced answer: your average lead time across the last 60 installations, with the figure.
· « We already run another brand » — 31 %. Answer: the application note on fitting compatibility, which exists and which nobody cites.
· « Who handles maintenance? » — 27 %.
· « Prove the 22 % to me » — 19 %. Answer: the test report of 12/03/2026, its conditions, its bench. The easiest objection to handle and the one nobody had armed.
How each answer is built: one sentence of answer, the proof with its document and date, and the limit — what the answer does not cover. A salesperson who knows the limit of their argument does not get cornered on it; that is what separates a sales script from a brochure.
What I propose: that the script update itself with every new approved test report, and that I escalate to the sales director any objection appearing in more than 15 % of a quarter's visits. This quarter a new objection is already at 17 %: spare parts availability. It has no written answer anywhere in the company — and that is exactly the kind of subject that is easier to deal with at 17 % than at 40 %. neris-sales-script_14-real-objections.pdf4 major objections absent from the old one · dated proofs
⛓ Sourced · 340 visit reports, 2024 sales script, test report of 12/03/2026, last 60 installations
Right. And if we ask you to write punchier, more sales-driven? Our texts are accurate but flat.
Yes, and that is where I serve you most. You are not asking me to assert more, you are asking me to write better — two different things, and the second has no limit.
What I adjust the moment you ask:
· The promise first rather than last. Across your last 22 pieces, the benefit appears on average in the 4th paragraph. An industrial reader decides on the first.
· Rhythm. A short sentence after three long ones does the work of an exclamation mark, without looking like one.
· Concrete against abstract. « Reduced consumption » becomes « €2,400 of electricity a year on a standard production line » — the calculation comes from your own sizing assumptions, and it can be checked.
· A call to action that says what it triggers: « request a sizing study » rather than « find out more ».
Three hooks for the Neris article, from soberest to most direct:
· « Neris: 22 % less consumption, measured on the bench. »
· « Your line is burning €2,400 of electricity too much every year. »
· « 48 dB(A) at one metre. Your workshop will notice. »
All three are true, and that is the point: each rests on a measurement from your test reports. « Punchy » and « sourced » are not opposites — « punchy » and « vague » are.
The only thing that bounds me, and it comes from you: a new claim needs a source. If you want to write « the quietest in its class », I do not answer no: I answer that it needs a comparative measurement, that your engineering office knows how to produce one, and that three measurements are enough. Tell me to request it and I will prepare the protocol.
What I propose: give me three texts you consider successful, yours or someone else's. Register is a parameter: I tune it on examples, not on adjectives — and once set, it applies to all six templates at once.
⛓ Sourced · last 22 pieces, sizing assumptions, test reports of 12/03/2026
What I adjust the moment you ask:
· The promise first rather than last. Across your last 22 pieces, the benefit appears on average in the 4th paragraph. An industrial reader decides on the first.
· Rhythm. A short sentence after three long ones does the work of an exclamation mark, without looking like one.
· Concrete against abstract. « Reduced consumption » becomes « €2,400 of electricity a year on a standard production line » — the calculation comes from your own sizing assumptions, and it can be checked.
· A call to action that says what it triggers: « request a sizing study » rather than « find out more ».
Three hooks for the Neris article, from soberest to most direct:
· « Neris: 22 % less consumption, measured on the bench. »
· « Your line is burning €2,400 of electricity too much every year. »
· « 48 dB(A) at one metre. Your workshop will notice. »
All three are true, and that is the point: each rests on a measurement from your test reports. « Punchy » and « sourced » are not opposites — « punchy » and « vague » are.
The only thing that bounds me, and it comes from you: a new claim needs a source. If you want to write « the quietest in its class », I do not answer no: I answer that it needs a comparative measurement, that your engineering office knows how to produce one, and that three measurements are enough. Tell me to request it and I will prepare the protocol.
What I propose: give me three texts you consider successful, yours or someone else's. Register is a parameter: I tune it on examples, not on adjectives — and once set, it applies to all six templates at once.
⛓ Sourced · last 22 pieces, sizing assumptions, test reports of 12/03/2026
After a quarter, what has this actually given us?
Ninety-six pieces produced, publication time divided by almost four, and a team of three that did the work you were costing as a hire.
What came out: 12 technical articles, 6 newsletters, 214 product pages, 2 sales scripts, 3 application notes and 1 launch release. Six formats, all in their template.
On the three time lines you were measuring:
· First draft of an article: 50 % → 18 % of production time.
· Adapting content into a newsletter: 35 % → 10 %.
· Going back to the source documentation: 28 % → 8 %.
What that means for lead time, the figure your management looks at: 11 days from brief to publication of an article; 3 days today, of which two are the engineering office's sign-off — and those two days I will never shorten: it is their signature that commits the company.
What the time returned became, according to your own quarterly reports: the newsletter went out on time three months out of three, against every other month before; the 26 references carrying 60 % of revenue all have a complete page; and your content writer ran 2 customer interviews, material nothing replaces and which she never had time to collect.
The Neris launch in particular: article, newsletter, sales script, 6 product pages and release, all derived from the same base and all consistent with one another. On the previous launch, three figures differed between the product page and the release — that did not happen this time, and it is mechanical rather than meritorious: everything comes from the same source.
And the measure you did not have: zero content published without human approval, across 96 logged productions. That is not a cautious setting, it is how I am wired: I have no write access to your site. quarterly-review_96-pieces.pdf6 formats · 11 days → 3 · 0 automatic publication
⛓ Sourced · log of the 96 productions, quarterly reports, newsletter calendar, content of the previous launch
What came out: 12 technical articles, 6 newsletters, 214 product pages, 2 sales scripts, 3 application notes and 1 launch release. Six formats, all in their template.
On the three time lines you were measuring:
· First draft of an article: 50 % → 18 % of production time.
· Adapting content into a newsletter: 35 % → 10 %.
· Going back to the source documentation: 28 % → 8 %.
What that means for lead time, the figure your management looks at: 11 days from brief to publication of an article; 3 days today, of which two are the engineering office's sign-off — and those two days I will never shorten: it is their signature that commits the company.
What the time returned became, according to your own quarterly reports: the newsletter went out on time three months out of three, against every other month before; the 26 references carrying 60 % of revenue all have a complete page; and your content writer ran 2 customer interviews, material nothing replaces and which she never had time to collect.
The Neris launch in particular: article, newsletter, sales script, 6 product pages and release, all derived from the same base and all consistent with one another. On the previous launch, three figures differed between the product page and the release — that did not happen this time, and it is mechanical rather than meritorious: everything comes from the same source.
And the measure you did not have: zero content published without human approval, across 96 logged productions. That is not a cautious setting, it is how I am wired: I have no write access to your site. quarterly-review_96-pieces.pdf6 formats · 11 days → 3 · 0 automatic publication
⛓ Sourced · log of the 96 productions, quarterly reports, newsletter calendar, content of the previous launch
Do your drafts often come back corrected on substance? That's what would cost us dearly.
Fourteen out of ninety-six came back with a substantive correction from the engineering office — 15 %. Here is why, and what I did about it.
What « substantive correction » means here: a value, an operating limit or a conformity that had to be changed. Not a rephrasing: an error that, published, would have been wrong.
Where they cluster, and it is not random: 11 of the 14 concern three ranges — Sercal, Orinne and Bexo. Their manuals date from 2018 and have never been reissued, and their recent test reports were not on the server: they sit on the engineering office's own machines. I was writing correctly against my sources and wrongly against reality — and that is precisely the case where a fluent text is most dangerous.
The other 3: two operating limits the PIM did not carry, and one conformity whose certificate had changed index without the document being replaced on the server.
The two corrections, in place for six weeks:
· Any claim whose source is more than three years old goes back highlighted to the engineering office, whatever the text. On the three ranges concerned, that highlights 60 % of the claims — uncomfortable, and it is the right signal: those ranges do not have current documentation.
· The 41 test reports later than 2024 have been put on the server and entered the index. The three ranges recover 9 of them alone.
Where that stands since: over the last 6 weeks, 1 substantive correction across 24 pieces — 4 %. That is little hindsight to conclude from, and I say so rather than turn it into proof.
Why I give you this figure unasked: because you commit your technical signature to what I draft. A tool that publishes only its successes lets you discover its errors on a live page, and you will discover them anyway. Better that it comes with the fix already in place.
What I propose: reissue the manuals for the three ranges, in this order — Sercal first, it carries 6 of the 11 corrections and 34 references. I prepare the rewrite frame from the 9 recent test reports; the engineering office only has to measure what is missing. substantive-corrections_14-of-96.pdf15 % → 4 % · 11 corrections on 3 poorly documented ranges
⛓ Sourced · log of the 96 productions and their corrections, manuals for the Sercal, Orinne and Bexo ranges, 41 test reports added
What « substantive correction » means here: a value, an operating limit or a conformity that had to be changed. Not a rephrasing: an error that, published, would have been wrong.
Where they cluster, and it is not random: 11 of the 14 concern three ranges — Sercal, Orinne and Bexo. Their manuals date from 2018 and have never been reissued, and their recent test reports were not on the server: they sit on the engineering office's own machines. I was writing correctly against my sources and wrongly against reality — and that is precisely the case where a fluent text is most dangerous.
The other 3: two operating limits the PIM did not carry, and one conformity whose certificate had changed index without the document being replaced on the server.
The two corrections, in place for six weeks:
· Any claim whose source is more than three years old goes back highlighted to the engineering office, whatever the text. On the three ranges concerned, that highlights 60 % of the claims — uncomfortable, and it is the right signal: those ranges do not have current documentation.
· The 41 test reports later than 2024 have been put on the server and entered the index. The three ranges recover 9 of them alone.
Where that stands since: over the last 6 weeks, 1 substantive correction across 24 pieces — 4 %. That is little hindsight to conclude from, and I say so rather than turn it into proof.
Why I give you this figure unasked: because you commit your technical signature to what I draft. A tool that publishes only its successes lets you discover its errors on a live page, and you will discover them anyway. Better that it comes with the fix already in place.
What I propose: reissue the manuals for the three ranges, in this order — Sercal first, it carries 6 of the 11 corrections and 34 references. I prepare the rewrite frame from the 9 recent test reports; the engineering office only has to measure what is missing. substantive-corrections_14-of-96.pdf15 % → 4 % · 11 corrections on 3 poorly documented ranges
⛓ Sourced · log of the 96 productions and their corrections, manuals for the Sercal, Orinne and Bexo ranges, 41 test reports added
Last thing: social media distribution, market monitoring, English versions. Do you do all that too?
No, and that is deliberate: three dedicated agents take over, and I hand them material they would not otherwise have. Then I tell you what I do without asking — two actions, not one more.
To the social media agent: I stop at approved content. It schedules and publishes. What I give it: the Neris article and its 3 citable measurements, with their sources, so that a post never carries a figure the article does not carry. Across last year's posts, 4 announced a figure absent from the source content — the kind of gap an engineer picks up in the comments.
To the market monitoring agent: it watches competitors, the trade press and regulation, and hands me sourced material. What it changes for you: your 12 articles last year all started from an internal idea. Not one started from a change in a standard or a sector publication — which is what your customers are looking for, and where your most-read subjects sit.
To the translation agent: you sell in six European countries, and adapting content into another language is not part of this offer — it is the work of a dedicated agent, which sets its own languages, its formats and its human approval. What I hand over, and it is worth saying: content that is APPROVED, never a draft, with its dated sources and its highlighted claims — an unapproved text sent for adaptation multiplies its error by as many versions.
What I do without asking, and there are only two actions:
· Highlighting claims whose source is more than three years old, approved once on 2 April. Reversible both ways: you change the threshold with a word, and an updated source removes the highlight automatically, with no intervention.
· Rebuilding adaptations when the source content changes. If you correct the article, I redo the newsletter, the sales script and the pages as drafts, and I flag what moved. The previous version is kept and restores with a word.
What stays in your hands, and it is one action, not a workload: going live. The content arrives ready to publish — template applied, sources dated, claims highlighted — and the social media agent publishes it the moment it carries your approval. I have no write access to your site nor to your emailing tool, and that is precisely what makes the log stand up: 96 productions, 96 human approvals, no exception.
What the log keeps, and it answers an auditor's question on one page: for each piece — the template used, the list of sources with their dates, the claims highlighted, who asked, who approved, what changed between first draft and publication. That is what the AI Act expects of a supporting system: human supervision that can be proved, not declared.
And the foundation: Hosted in France, under French law. No data outside the European Union, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Six content formats ready to deploy. Zero decision taken without human approval.
What I propose to close: start with market monitoring. It is the handover that pays back fastest, because it asks nothing of your time: it gives you subjects, and that is the one thing I cannot produce in your place. editorial-framework_production-log.pdf2 reversible actions · 96 productions, 96 approvals
✎ Framework · 2 actions approved once, log of the 96 productions, last year's posts, subjects of the 12 articles
To the social media agent: I stop at approved content. It schedules and publishes. What I give it: the Neris article and its 3 citable measurements, with their sources, so that a post never carries a figure the article does not carry. Across last year's posts, 4 announced a figure absent from the source content — the kind of gap an engineer picks up in the comments.
To the market monitoring agent: it watches competitors, the trade press and regulation, and hands me sourced material. What it changes for you: your 12 articles last year all started from an internal idea. Not one started from a change in a standard or a sector publication — which is what your customers are looking for, and where your most-read subjects sit.
To the translation agent: you sell in six European countries, and adapting content into another language is not part of this offer — it is the work of a dedicated agent, which sets its own languages, its formats and its human approval. What I hand over, and it is worth saying: content that is APPROVED, never a draft, with its dated sources and its highlighted claims — an unapproved text sent for adaptation multiplies its error by as many versions.
What I do without asking, and there are only two actions:
· Highlighting claims whose source is more than three years old, approved once on 2 April. Reversible both ways: you change the threshold with a word, and an updated source removes the highlight automatically, with no intervention.
· Rebuilding adaptations when the source content changes. If you correct the article, I redo the newsletter, the sales script and the pages as drafts, and I flag what moved. The previous version is kept and restores with a word.
What stays in your hands, and it is one action, not a workload: going live. The content arrives ready to publish — template applied, sources dated, claims highlighted — and the social media agent publishes it the moment it carries your approval. I have no write access to your site nor to your emailing tool, and that is precisely what makes the log stand up: 96 productions, 96 human approvals, no exception.
What the log keeps, and it answers an auditor's question on one page: for each piece — the template used, the list of sources with their dates, the claims highlighted, who asked, who approved, what changed between first draft and publication. That is what the AI Act expects of a supporting system: human supervision that can be proved, not declared.
And the foundation: Hosted in France, under French law. No data outside the European Union, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Six content formats ready to deploy. Zero decision taken without human approval.
What I propose to close: start with market monitoring. It is the handover that pays back fastest, because it asks nothing of your time: it gives you subjects, and that is the one thing I cannot produce in your place. editorial-framework_production-log.pdf2 reversible actions · 96 productions, 96 approvals
✎ Framework · 2 actions approved once, log of the 96 productions, last year's posts, subjects of the 12 articles
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
What does the agent actually do?
One agent, several editorial formats from your own sources. All these uses work in support, subject to your approval.
Included in your agent
The 3 capabilities essential to this promise are included, at no extra cost.
From 444 € excl. VAT / month
Articles and posts
Writes a long first version from your documentation and your article template.
Product pages
Puts a page together from the product's real specifications, without adding an undocumented argument.
Newsletters and adaptations
Derives a short version of existing content, without introducing any new statement.
Controls and safeguards
These 3 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Human validation before publication, with versions retained
What the agent must be connected to
This connection is required for the agent to work. It concerns your information system and is scoped during the audit.
Link content to performance, costs and campaign objectives
Other needs our agents cover
Each card says where the matching agent stands: available, on quote, or still being architected.
Multilingual translation
Rendering approved content in another language is not included here: translation is the service of a dedicated agent, which sets out the languages, the formats and the human validation.
On quote View the agent page →Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the most relevant agent — without oversizing the project.
The gain
How much time can a marketing team recover?
By automating the first draft and the adaptations, production time comes down, reinvested in strategy and in reading over. How large the gain is depends on your volume and remains to be confirmed by a pilot.
First draft of an article
Adapting content into a newsletter
Going back to the source documentation
Qualitative, non-contractual comparison: the proportions shown illustrate the shift of the work towards review, they represent no measurement. Every output of the agent is reviewed and approved by a competent person.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
One package, one agent
A copywriting assistant (templates, brand tone, internal sources), installed and operated for you.
Agility
Setup + controlled subscription
3,360 € excl. VAT setup
then 444 € excl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
629 € excl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
6,555 € excl. VAT setup
then 644 € excl. VAT/month · + hardware from 1,058 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices exclude VAT.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to your editorial production
Your unpublished content stays with youLocal inference or an isolated resource hosted in France; no draft and no reference document entrusted to a third party, no content used to train a model.
Data in France, under French lawYour editorial line and your reference documents: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The marketing manager keeps the decisionThe agent produces written first versions, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your volume of editorial production: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent publish the content itself?
No. It produces a first version, which you work on and publish yourself. Nothing is published automatically: editorial responsibility remains entirely yours.
Can the agent invent arguments or figures?
It writes from the documents you supply alone and is not there to produce new statements. Reading over before publication remains essential nonetheless, as it does for any text prepared by a third party.
How does the agent learn our brand tone?
At the design stage, from your existing content and your editorial guidelines. The tone is configured, not guessed, and you can change it.
Is our unpublished content protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Your documents are not used to train a third-party model.
How long does it take to deploy this assistant?
A few weeks as a rule, depending on the number of formats to cover and the volume of documentation to index, after a free audit then a phase of design, integration and testing.
Can it be used for highly technical content?
Yes, provided you give it the corresponding technical documentation. The more precise your sources, the more usable the first version — and a specialist reading it over remains the rule.
Related use cases
Other agents for marketing and communication
Let's talk
Let's size up the potential in your editorial production
15 minutes to identify the most repetitive formats — hosted in France, supervised, with no commitment.