Invented specifications
A model fills gaps. A wrong hardness rating or a maintenance instruction that does not hold ends up under your name, not the tool's.
Article engine
Every week, an article that only says what you supplied
An article on your site that claims nothing beyond your own documentation. You read it in your portal, you approve it, and only then does it go live.
01 – Why not just AI
A text generator creates three problems that become yours
Copied supplier text
Your suppliers' documentation is protected by copyright. Lifting it word for word is the risk that gets underestimated most often.
Google penalises scaled content
Publishing a lot without adding value falls under the scaled content abuse policy. That hits your whole domain, not one page.
02 – How it works
Four locks on the door
Your sources, and nothing else
Product sheets, maintenance instructions, warranty terms, the explanations you now email one by one. Anything outside that does not count.
The gate on the way in
A piece with a claim that has no source never arrives. It is refused at the door, so it does not reach you and it does not reach your site.
A source behind every claim
Behind each claim sits the paragraph it rests on, and which document it came from. Click it and you read that paragraph, with the address attached.
Your approval
You read, edit, or send it back with a reason. Who decided what and when is recorded. Without that click, nothing happens.
03 – What does not get through
Four checks, not a second opinion
The gate is arithmetic, not a second AI judging the first one's work. If a piece fails, it never even reaches you.
invented sources
Every reference has to point at a paragraph that genuinely sits in your documentation.
figures without a source
A paragraph with a number in it and no reference below does not get in. That is exactly where a text generator goes wrong.
pieces without references
An article that points nowhere is a guess, however well it reads.
human who approves
You. Which is also your cover under the EU AI Act, article 50.
A treated worktop keeps stains out for as long as the sealant holds.Reference 1, pointing at the source below.
Worktop maintenance sheet, paragraph 3
The sealant forms a shield against stains from oil, grease, wine and coffee. Reapply it as soon as water stops beading on the surface.
This is what your portal shows: click the number and you read the paragraph from your own document. The example itself is made up; in your articles it is your documentation sitting underneath.
04 – Your portal
Everything asked of you sits in one place
You log in with your own email address. It shows what is waiting for you, what it rests on, and where the project stands. Twenty minutes per piece, no more than that.
- Approve, edit, or send back with a reason.
- Manage your sources: paste them in or pick a file.
- Adjust the brief: what about, for whom, what to avoid.
- Your documents and the numbers from your tools, in the same place.
05 – Where it lands
On your site, whatever it runs on
The moment you approve, the piece goes to your own site. Which route that takes we settle in the first week, and it changes nothing about the price.
- WordPress, through an application password. Straight online or as a draft in your editor.
- A static site through GitHub: a file in your repo, your build does the rest.
- Your own system or a middleman, through an address with a secret in the header.
- If your platform cannot take it, we put the blog beside it on your own domain, in a subfolder and never on a subdomain.
06 – What we need from you
Three things, all three necessary
The source set
The thicker the pack, the better the articles. We test this in week one, not in month four.
One hour with your sales desk
The questions customers ask daily live in their heads and nowhere else.
One fixed approver
Someone who can judge the content and press approve. That is your legal protection too.
07 – Law and liability
Sorted before you ask
- Article 50 of the EU AI Act has applied since 2 August 2026. Where a human checks the copy before it appears, the labelling duty falls away, and that check is exactly what you do in the portal, with a log to show for it.
- You warrant that you may use the documentation you supply. We claim no copyright on AI output; in the EU it probably does not exist, and we would rather say so.
- Sensitive subjects are excluded, medical or occupational safety advice for instance. That sits in the settings as a forbidden subject, not as a good intention.
08 – Questions
What people ask before they start
What does an article engine cost?
A fixed price for your situation, and you hear it after one conversation. What sets it is how much material you have and how often you want to publish. No hourly billing, and you are not tied to anything afterwards.
How do I know nothing gets invented?
Because a piece with a claim that has no source behind it does not get in. That is a check at the door, not a promise: every reference has to point at a paragraph that really exists in your own documentation, and a paragraph with a number in it and no reference is refused. Behind every claim it says which document it came from, and you can read that paragraph yourself.
What if I have no documentation?
Then we say so, at the end of the first week. We start with a source check and not with writing, because without material it becomes a text generator and that is not what we sell. It is often better than people think: product sheets, maintenance instructions and the explanations your sales desk emails out every day already add up to a set.
Do I have to label the text as AI-generated?
AI Act article 50 has applied since 2 August 2026. Editorial review by a person removes that labelling duty, and that review is exactly what you do in the portal: you read, you approve, and who decided what and when is recorded. That log is your cover.
Does it work with my website?
Yes. WordPress through an application password, a static site through your own repo, or your own system through an endpoint with a secret in the header. If your platform cannot accept anything, we put the blog next to it on your own domain, in a subfolder and never on a subdomain. We decide the route in the first week, and it does not change the price.
How much of my time does it take?
About twenty minutes per piece: reading, and then approving, editing or sending it back with a reason. Plus one hour-long conversation with your sales desk at the start, because the questions customers ask every day live in their heads and nowhere else.
Who owns the text that comes out?
You do. We claim no copyright on AI output; in the EU it probably does not exist, and we would rather say that plainly than bury it in small print. You do warrant that you are allowed to use the documentation you supply.
Shall we start with a source check?
By the end of the first week we both know whether there is enough material. If there is not, we will say so.
Book a call