Making your company readable by AI. Outside, and in. Primary practice · AI search optimisation
Be cited, not just ranked.
Outside: what AI systems say about you. The other direction of the same work is AI inside the business, on the inside.
When your customers ask ChatGPT, Claude or Gemini a question, the answer names a few companies. Rarely yours. HENUSSE makes your company citable, then shows you in black and white where you appear and where you do not. The decision to cite rests with the answer engines. The work that makes you eligible is ours.
- Initial call with no commitment
- No access or installation
What the engines answer
Reading recorded on 12 August 2026
“One-on-one French immersion Brittany”
realfrench.cocited 7 times out of 8
“French immersion for diplomats”
realfrench.cocited 4 times out of 8
“Why can’t I understand spoken French?”
realfrench.cocited 0 times out of 8
A real reading from realfrench.co, a website we built and operate, owned by the Hénusse family. We disclose the tie so you can judge the proof knowing it. Here, “cited” means the answer uses the site as a source; a name mentioned without a link is a separate measure. An AI system does not always answer the same question identically: we therefore ask it several times, using ChatGPT, Claude and Gemini, and publish the number of answers that cite the site out of the number of answers returned. The first two are buyer questions. The third comes much earlier: it is the work in progress. Its zero result is published.
The terrain
In 2026, search has moved to new terrain.
For twenty years, being visible meant ranking well in a list of websites on Google. Today, your buyers ask ChatGPT, Claude or Gemini their question directly. The answer names a few companies. The others remain invisible. Why these three and not Perplexity, Mistral or Grok: the panel’s scope, with dated figures.
Traditional search engine optimisation remains the base. A second terrain now sits alongside it: the answer itself. Both are won in the same place, your website and what it says about your company.
Anatomy of an article
One page. Two readings.
What your reader sees, then what the machine can connect. Follow the path of a real question as AI searches for sources.
Your customer asks AI
AI opens your article. It does not read it like a human.
The real page, as published
Real, published article · And it answers the question precisely.
Why Can I Read French But Not Understand It Spoken?
You can read French because print is stable. Spoken French fails when your ear has not learned the vowel categories native speakers use without effort every day.1
Why can you read French but not understand it spoken?
Sources: 1 Linguistic Experience Alters Phonetic Perception in Infants by 6 Months of Age (1992) · 2 Early language acquisition: cracking the speech code (2004)
FAQ · Frequently asked questions
A real article published on realfrench.co, a website designed and published by HENUSSE
Under the bonnetAn invisible layer placed in the page's code
The same page, restated in JSON: what you read on the left, written separately for machines.
Markup clarifies. It does not promise a citation.
An editorial engineering method
The answer, author, sources, FAQ and structure. Each piece in its place, so your content can be used as a reference.
The answer comes back with a citation
“You can read French because print is stable. Spoken French fails when your ear has not learned the vowel categories native speakers use without effort every day.”
answerauthorsource ↳ source: realfrench.coAn illustration of the mechanism. No citation is promised.
You do not have to master all of this.
We put each piece in its place. You approve it; we carry the editorial and technical work.
“These are not buyer questions.” No. They decide which name is cited when the buyer’s question arrives.
The measurement that confirms itYou do not have to master all of this.
We put each piece in its place. You approve it; we carry the editorial and technical work.
“These are not buyer questions.” No. They decide which name is cited when the buyer’s question arrives.
The measurement that confirms itThe public measurement
The same article, in Google’s public tool.
Google’s public tool displays the structured data detected on this same article.
9 valid items detectedGoogle’s tool confirms that it correctly understood what this article is about, who wrote it and which company it belongs to. Google also reports non-blocking remarks about optional fields we have not filled in. We leave the notice visible; it is what you would see if you ran the test again.
These scores measure how the website is built, not the engines’ decisions. This is the part of the work for which we are accountable.
The free-access line in this result simply declares that the article can be read without a paywall. Run again regularly. We publish the latest measurement, not the best.
And the page that carries it
The website that hosts this article, measured on the same day.
A well-declared article is not enough: the page must load quickly and remain readable, for your readers and for agents. Here is the measurement of this website’s homepage.
- 96Performance
- 100Accessibility
- 100Best Practices
- 100SEO
- 3/3Readable by AI agents
Run again regularly. We publish the latest measurement, not the best.
The objection
“You don’t know my field.”
That is true. And we do not make it up either. What we bring is not your expertise: it is the mechanism that takes it out of your head, organises it, checks it and publishes it in a form the engines can use.
Everything you have already published goes in first. Your website, your documents, your fact sheets, your answers to customers. This is the raw material, and it already exists. We turn it into a library that belongs to you: your facts, your references, your entities and your jargon. It grows with every article and never starts again from zero.
Your tone is measured against this material. Sentence length, vocabulary, phrasing and the way you give an example: a fingerprint calculated from your texts, not imitated by ear. An article that does not match it is not released.
We look for questions worth answering. Not keywords: the real questions in your market for which a solid answer can make you citable, and those that no one has yet addressed properly. Each one comes with sources to verify.
And we monitor what changes. A radar cross-references French regulatory sources with your sector. When a rule changes, a question emerges the same day, and it often takes weeks before anyone answers it properly. The radar flags it, a human decides, and nothing is published on its own.
You keep the final say. Any consequential claim carries a reference you can open; anything that cannot be verified is not published. Nothing goes out without your agreement, with a deadline set out in the written quotation for you to say no.
The tools that do this work are ours, built for this purpose and no one else. They are not for sale: they ensure that the article bears your name without you having to write it.
The second objection
“These are not buyer questions.”
That is true, and that is the point. Your buyers ask two families of questions at two different times, and they are not won in the same way.
The first family comes early. The buyer is trying to understand a problem, not yet choose a provider. “Why do the people I speak to struggle to understand my English?” No one signs after this answer. Yet it is useful as soon as it is published, and it may contribute to what a future model learns, though that effect cannot be measured.
The second comes when it is time to choose. “Which provider should I choose?” Here, the answer consists of three or four names. This is not the day when you build your place; it is the day when you find out whether you have one.
We work on both and measure them separately: the dated reading at the top of this page distinguishes these two families precisely.
Two families, so two horizons. Web search changes in weeks: a well-built article can be found and cited quickly. Models’ memory changes with each generation: that effect is possible, but no one can measure or promise it, and we do not sell it.
The work
The work, piece by piece.
One job goes by several names: AI search optimisation, AEO for answer engine optimisation, GEO for generative engine optimisation and sometimes LLMO for large language model optimisation. We use the first because a business owner is not looking for an acronym; the other terms of the trade are defined in the glossary. They all mean the same thing: being the name the machine cites when your buyer asks a question.
And this is not classic search work repainted for AI. The discipline is strict: original content that answers the real questions of your field, carried by an identifiable author, connected to proofs and dated sources, on a technically sound site that machines can read as text. Nothing is manufactured “for the models”: this is the ground the engines themselves document, Google for its AI features and OpenAI for its crawler access. We do not bet against the rules of the terrain; we apply them, piece by piece, as follows.
- A voice analysis based on your existing texts.
- References checked against the original source before publication.
- Author pages that name the people responsible for the content.
- Structured data added to the pages and validated in the public tool.
- A dated measurement of observed citations, including zero results.
We commit only to what can be proved. Neither citations nor rankings: the decision belongs to the engines. Our commitment is to what we build and what we measure.
We also work on what others say about you
AI search does not stop at your own website. Answer engines also read what others publish about you: press, professional directories, partners, reviews. We find the sources AI systems cite for your questions, and those that carry authority in your field, then strengthen your presence where they look. A perfect website absent from third-party sources leaves that part of the reading empty. The decision to cite stays with the engine: our readings record it rather than promise it.
An AI answer does not always reveal all its sources: we can observe the ones it cites, not know everything that influenced it. So we work on verifiable sources, without pretending to read inside a black box.
This off-site work is a separate service: its foundation costs €2,500 excl. VAT, once, and it is included at the Sovereignty tier, refreshed each quarter. At the Essential and Authority tiers it is added to the monthly programme.
Honest measurement
Three verifiable commitments.
The Honest Measurement Charter publishes the zeros. The test is reproducible under the conditions it describes. Access is returned. These three commitments appear in the charter and the written quotation. The measuring instrument is itself checked against the ChatGPT application: at the published reading, same verdict, 30 questions out of 30. That figure measures agreement between two instruments, not a citation rate.
Read the Honest Measurement Charter · See the dated Barometer
Before choosing a provider, read Questions to ask before signing, the map of the market’s players, the register of what each one publishes, quoted and dated, and the speed of their websites, measured.
To see how the engagement begins, follow your first 90 days with HENUSSE.
The price and its logic
How much it costs, and why
The AEO Foundation first brings your website up to standard, then the monthly programme publishes. Both have public prices.
This is not a three-month return on investment. It is your presence in the way your buyers search today: the immediate answer that AI gives from memory, and web search when it retrieves its sources. Training future models remains possible; it cannot be measured and is not sold.
Before the monthly programme, the website is brought up to standard: a website build, an existing-site rebuild or AEO setup only, depending on its condition; the AI Citation Diagnostic determines which. Public prices. See website build and existing-site rebuild.
Then comes the monthly programme: Essential at €1,500 excl. VAT, Authority at €3,500 excl. VAT and Sovereignty at €10,000 excl. VAT. The same price for everyone. See the full pricing table and what it includes.
The AI Citation Diagnostic costs €1,500 excl. VAT and is deducted from the programme or AEO Foundation that follows, once, within twelve months. Its verdict may be that you should buy nothing from us. The AI Citation Diagnostic in detail.
The deliverable
A signed evidence pack, not a dashboard.
You do not receive figures to interpret on your own. You receive a dated evidence pack and a decision note, signed by Théo Hénusse for HENUSSE.
Every claim in the pack is dated and sourced. What we have not measured is stated as such. These packs accumulate in your workspace, alongside the articles awaiting your veto: you read the history of your measurements, not a gauge.
Access and governance
Your website remains yours, keys included.
Temporary administrator access, publication as a draft, a right of veto over every article, and no access retained after delivery unless you separately entrust us with webmastering. You remain the sole publisher, free to unpublish anything at any time.
The person in charge
Théo Hénusse, founder of HENUSSE.
HENUSSE is an independent business, not an invented team. Théo Hénusse is responsible for the method, the reporting and the commercial relationship. Every evidence pack and every decision note bears his signature.
The first step
Thirty minutes with the person who will do the work.
Bring a question your customers ask. Together, we look at what ChatGPT, Claude and Gemini answer today and what that says about your market. No charge, no commitment, no access to your systems.