Making your company readable by AI. Outside, and in. Questions to ask before signing with an AI search provider
AI search optimisation (GEO, also called AEO) is a young market. Promises are easy to make and evidence is scarce. Here are the thirteen questions we would ask any provider if we were in your position, along with our own answers. To map the players on the market, see also how to choose your provider.
Show me a client, even anonymised, with the exact queries and before-and-after measurements.
As of August 2026, our three public proofs have three different origins. realfrench.co is the residential immersion school founded in 2008 by Bernard Henusse, Théo’s father, who is its sole owner and runs it. It has more than 400 former students from over 30 countries since 2008, according to figures published by the school; Théo is its co-director, and the school is a HENUSSE client. nicolehenusse.be is the studio of Nicole Henusse, HENUSSE’s partner and Théo’s wife. english-performance.com is the executive English coaching business created and run by Théo, now moving towards a separate entity with its own coaches; it is not a client, but a demonstration of our work on our own business. We state all three origins so you can judge each piece of evidence knowing where it comes from. Our rule remains the same: we freeze the questions and publish the baseline before starting, so the before-and-after comparison cannot be rewritten afterwards.
What proportion of your work takes place outside my website?
A meaningful part. Answer engines cross-check your website against what third-party sources say: press, professional directories, partners and reviews. We work on both: your website, so it answers your buyers’ questions precisely, and your presence in the sources AI systems consult. A provider who talks only about your own website is doing only half the job.
How do you handle robots.txt, AI engine crawlers and llms.txt?
Access comes down to robots.txt and to allowing each engine’s crawlers, and not every crawler plays the same role. Three families matter. Search crawlers decide whether you can appear as a source: OAI-SearchBot for ChatGPT, Claude-SearchBot for Claude, Googlebot for Google. User-triggered fetchers read a page live when someone asks the model a question: ChatGPT-User, Claude-User. Training crawlers have no effect on whether you get cited: GPTBot, ClaudeBot, Google-Extended. We verify this access with real read tests, because a hosting firewall can block these crawlers without anyone noticing. The llms.txt file is a complementary layer: we maintain it, but it authorizes nothing by itself, and Google said in 2025 that it does not use it. Be wary of an offer that gives it the leading role, or that treats every crawler as one and the same.
After six or twelve months, what would make you say we should stop?
If, after full deployment, the share of AI answers that cite or recommend your company for the questions fixed at the outset has not increased from the baseline, and visits from answer engines have not moved either, we will tell you to stop. Readings are weekly. The report is quarterly for the Foundation tier and monthly for the Authority and Sovereignty tiers. The stop rule is judged on quarterly steps built from those readings, so that no conclusion rests on the noise of a single week: two consecutive quarters after full deployment with no progress on either that share or those visits, and we recommend in writing that you stop. And so that this clock does not depend on anyone’s judgement, “full deployment” is a written date: the go-live of the last Foundation deliverable, recorded in the quotation, at the latest 90 days after signature unless stated otherwise. The two quarters run from that date.
Do you measure the real ChatGPT interface, and can I see the raw answers?
Our readings go through the vendors’ own official tools: the command-line interfaces from OpenAI, Anthropic and Google. They are the same engines, through each vendor’s official web search tool, and every reading names the exact model that answered. This surface has a measurement virtue: the session is neutral, with no history and no personalisation, and anyone can rerun the test and find our numbers again, whereas an answer from the application depends on the account asking. That gap, the model served to an average user by default, has now been measured: at the reading of the night of 14 to 15 August 2026, the ChatGPT application and our instrument returned the same presence verdict on all 30 frozen questions, 30 out of 30. The correlation study is published, method, limits and evidence included, and will be rerun quarterly.
And yes: the raw answers behind every reading are kept, dated, and handed over on request with the report. You can recount behind us.
You are a small house: who does the work, and what if you are gone?
Théo does the work himself: the research, the writing, the technical layer and the measurement. Nicole handles positioning and visual identity. Nothing is passed to a content mill, and we deliberately limit the number of clients we follow at once: availability can be checked from the first call. What we do take on, we run with our own instruments: the engine that publishes, measures and monitors this site is the one that works for our clients.
If the house stopped tomorrow, you would carry on without us: everything is published on your site, under your name, and documented so any serious provider can take over.
Who writes the code and handles incidents? Théo Henusse writes and maintains the code and steps in when something fails. We are a small firm: the same person is responsible from quotation to incident response, with no hidden subcontracting. The trade-off is written down: everything is documented so another provider can take over without us, and that documentation is part of the deliverable, not an option.
What do we keep if we leave?
Everything. The website, the content, the structured data, the source corpus, the measurement readings and the documentation are yours, published on your site, under your name. The publishing plugin can be removed in one click and the site keeps working without it. Leaving costs nothing beyond the end of the programme: it is written into the contract, and you can verify it before signing.
Can we exit before the end of the 12 months?
Yes, and without a negotiation. If our own readings trigger the public stopping rule, two consecutive quarters without progress after full deployment, we recommend stopping in writing, and that recommendation opens an exit at thirty days’ notice, before the end of the term. The exact duration and exit terms are written into the quotation, before you sign. You are never locked into a programme our own numbers judge useless.
Where do your tools run, and where does my data go?
For an integration, the tools are installed on your machines or on a server you control, never on ours. An AI search audit does not deploy any tool in your environment. In both cases, the deliverable or scoping document states which third-party models are called and what they receive, where logs are stored, who has which access rights and how long each item of data is kept. If a point is not in the document, it has not been decided, and we do not put anything into production that has not been decided.
There are two of you. Who steps in if Théo is ill, injured or overloaded?
Théo answers and handles the request himself. On average, he responds and deals with it within two to three working days. If illness, workload or another unexpected event slows him down, he tells you straight away and gives an estimated date. Unless the request falls outside the agreed scope, the intervention is completed within ten working days. A large team does not guarantee a quick response: waits of several weeks are common in this market. We would rather publish our commitment and be judged against it. And the commitment has a written consequence: beyond ten working days of unavailability on our side, monthly billing is suspended pro rata. Continuity is not only about response time: the work is documented as it progresses, the quotation and our terms set out your exit rights, and the handover file allows another provider to take over.
Why does HENUSSE not yet appear in its own barometer?
Because the barometer tells the truth and our brand is young: henusse.com has only been indexable since 6 August 2026, and being cited by the engines is built over time. That is exactly why we publish the zero, dated, rather than hiding it: the barometer is the instrument that will show the progression, reading after reading, in the same place and with the same method.
A provider who only publishes good numbers is not showing you an instrument, they are showing you an advertisement. What we can prove is the method. The result depends on algorithms nobody controls: that is why we measure it instead of promising it.
What happens if Théo is unavailable?
The question is worth asking of any two-person consultancy, and our answer does not rest on a promise but on how the work is built. Everything is documented as we go, not at the end: the scoping phase produces a document that belongs to you and that another provider can execute, the systems installed in your environment are documented and reversible, and the content and data produced are your property.
So an interruption pauses the programme; it breaks nothing. Nothing is published without your approval, so nothing goes out on its own; your website, your access and your knowledge base stay with you; and when work resumes, the barometer reports what happened during the pause, as it always does. The same rule protects you if you leave: what we leave behind must be something anyone can pick up, us or someone else.
Why don’t you go through subsidised support schemes?
Because a subsidy reduces the invoice, not the total cost. A subsidised path adds a file to assemble, an approval to wait for, a standardised format calibrated for the programme rather than for your problem, and a provider picked from a list for its compliance with the file, not for its craft. That friction is paid in weeks and in scope, and it never shows on a quote.
With us, work starts when you decide, the scope fits your problem, and we answer to you alone. Our prices are public, the same for everyone, and every euro is written on the quote. If public support exists for your project, nothing stops you from claiming it on your side: we simply do not build our offer around it.
These questions apply to us as much as to anyone else. A serious provider answers them with figures and limits, not promises. What each player on the market actually publishes is recorded in the market register.