AI SEO (GEO), in plain terms
When someone asks ChatGPT a question, artificial intelligence no longer just shows ten links: it writes a summary and mentions a few sources. AI SEO means making your website one of those sources. It is built on traditional SEO (being indexed, fast and well structured) and adds one requirement: content that is clear, precise and verifiable.
The difference lies mainly in the goal. SEO aims to rank a URL in a list of results and is measured in positions and clicks; AI SEO aims to have your content used in a written answer and is measured in mentions: another form of visibility. Good SEO content is also the best foundation for AI: there is no separate SEO strategy and AI strategy to run, nor two sites to build, only one to make useful.
Several names are used for the same approach:
- GEO (Generative Engine Optimization): optimising for generative engines, which write an answer instead of listing pages;
- AEO (Answer Engine Optimization): optimising for answer engines;
- LLMO: optimising for large language models (GPT, Gemini, Claude…).
For a small business, the difference between these acronyms does not matter at all. What matters is simple: when a prospect asks a conversational interface such as ChatGPT “who can help me with…”, do your name and your brand appear in the answer? The goal of AI SEO is to make that happen, for the requests that match your trade.
What changed in France with AI Overviews
Since 22 July 2026, Google has shown an AI-generated answer above the results for part of the searches made in France. Websites lose clicks even when their position in the results does not move. According to an Ahrefs study of 963 French domains, the most exposed ones lost 23.1% of their click-through rate within a few days.
The results of this study, published on 17 August 2026, show the scale of the impact:
- 82% of highly exposed domains saw their click-through rate fall;
- the median drop in clicks was 8.8% across the whole panel, and 12.3% for .fr sites;
- almost one site in five lost more than 30% of its daily clicks;
- health was the hardest-hit sector, with click-through rate down 20.4%.
At the same time, traffic sent directly by AI platforms is growing but remains small: 0.26% of the traffic of French websites in early 2026 according to SE Ranking, against more than 50% for classic search. ChatGPT brings more than three quarters of it, with Perplexity and Gemini sharing most of the rest. The traditional search engine is still your first source of visits, but the buying decision is increasingly prepared in a conversation with an AI, before the first click. It is a strong shift in habits, and it affects your marketing too: your customers are picking up new ways of searching.
How an AI engine chooses its sources
An AI, or an agent able to browse the web on its own, does not pick a website at random. It runs several searches in the background, reads the best-ranked pages in an index (Google’s, Bing’s or its own), then keeps the ones that answer the request most directly, in the right context. A page missing from the index, or one that talks around the question without answering it, has no chance of being cited.
Google calls this principle “query fan-out”: for a single request, its system runs several queries on related aspects and cross-checks the results. A page that covers a precise topic in depth, including its follow-up questions, is therefore more likely to be selected than a general page. Three technical conditions must be met:
- Being indexed by the right engines. Google for AI Overviews and AI Mode; Bing for Copilot; OpenAI’s OAI-SearchBot crawler for ChatGPT search. OpenAI states that a site blocking OAI-SearchBot will not appear in ChatGPT search answers.
- Being accessible and readable. The text must be in the HTML (not only displayed by JavaScript), with ordered headings, one clean URL per page and good loading speed.
- Having consistent structured data. Google does not require it for its AI features, but asks that it match the visible text. schema.org markup helps engines identify the entity behind the site: who you are, what you offer, where and at what price.
What gets a page cited by an AI
A language model already knows the generic answers. It cites a page when the page brings something it cannot generate on its own: new information (your data, your experience), a compilation that saves time (a comparison table, a list of dates), or content that is kept up to date and dated. Before publishing, I ask myself a single question: does this page deserve to be cited?
In practice, that gives a few simple writing rules:
- The answer first. Each section opens with a paragraph of two or three sentences that answers the question; the detail comes afterwards. That is the passage the AI uses.
- Sourced, dated facts. “Faster” proves nothing; “page loaded in 1.2 seconds, measured on 3 November” is concrete proof.
- Published prices. An AI asked about a cost readily cites a site that shows a price, rarely one that answers “on quotation”.
- A named author. Name, trade, expertise: Google recommends showing who wrote the content and why they can be trusted. It is a credibility signal for readers and algorithms alike.
- A visible update date, consistent with the one in the structured data.
- An FAQ specific to each page, with your customers’ real questions, not ones already covered elsewhere on the site.
- Explicit vocabulary. Name things (the product, the town, the trade) rather than using vague phrases: the semantic understanding of engines relies on those words.
The content formats that work best
Not all content has the same chance of being picked up. The most useful formats for AI SEO are those that meet a precise need with verifiable elements: a service page with prices, a case study with figures, a step-by-step guide, a comparison, an FAQ. A generic blog post, which the AI could write itself, is worth almost nothing to it.
- The service page: what you do, for whom, at what price, how long it takes. It is the page an AI mentions when a user is looking for a provider.
- The case study: a real project, the problem, the solution, the results in figures. This kind of content is unique by nature, and often the most useful for a prospect.
- The guide or comparison: a complete resource on a topic in your field, with a table when there are options to compare.
- Images and photos of your work, with a clear file name and a descriptive alt text: a well-described image reassures the visitor and gives engines context.
A simple editorial line is enough: fewer pages, but better ones, each focused on one topic. It also makes the site easier to maintain.
Your brand beyond your website
An AI does not only read your site: it cross-checks what is said about you elsewhere. A business mentioned consistently across several reliable sources (Google profile, professional directories, social networks, local media, customer reviews) inspires more confidence in the engines than an isolated site. This is a field observation, not a published rule, but it shows in the answers.
- The same name everywhere, with the same description of your activity and area: that is what lets the AI link these mentions to your brand.
- A complete Google Business Profile, with photos and recent reviews.
- Active social networks on the channels where your customers are (LinkedIn for B2B, for example), with a link to your site.
- Links and mentions from reference sites in your sector or your town: they strengthen your brand’s authority, just as in SEO.
Think also of the less obvious channels: forums in your trade, local groups, partners’ newsletters. Every accurate mention on these channels and networks adds to the signals the algorithms cross-check.
Your communication and marketing therefore also feed your visibility in AI: a brand known across several channels is a brand the engines can identify.
What I did on datapet.fr, and what it produced
I launched the new version of datapet.fr on 29 September 2026, applying these rules from the start. Nine days later, on 8 October, the Lothaire chocolate maker near Reims, at the other end of France, wrote to me through the site’s form: a maintenance plan was signed, then the redesign of their online store, which had been hacked. On 20 October, a visit from chatgpt.com, from a town in Limburg in the Netherlands, landed on my English PrestaShop migration page; forty minutes later, from the same town, the form on that same page brought me a first contact: a Dutch online retailer wanted to know how to keep his PrestaShop store running for five to eight more years. The statistics do not link the two visits, but the sequence speaks for itself. On prix-essence.com, one of my content sites, 17% of visits came from AI assistants three weeks after launch.
What is in place on datapet.fr, and what I reproduce on my clients’ projects:
- a lead paragraph and a “Key points” box at the top of each blog post, summing up the answer in a few lines;
- structured data on every page: business and area served, author, articles, FAQ, prices of the services;
- all prices published, on each service page and on the pricing page;
- a different FAQ on every page, and a visible “updated on” date;
- an llms.txt file summarising the site (see below what it is really worth);
- automatic submission of new URLs to Bing through IndexNow, so that Copilot discovers them quickly;
- an English version, with hreflang tags, for a wider audience.
The server logs also show that AI tools come to read the site: 90 reads by ChatGPT (the ChatGPT-User agent, triggered when a user asks a question) between 25 October and 8 November 2026, six a day, mostly on the home page, the maintenance page and the pricing page. I discarded 63 fake visits pretending to be ChatGPT: only those from the IP addresses published by OpenAI count. In Matomo, 9 visits out of 231 came from an AI assistant between 29 September and 7 November (8 from ChatGPT, 1 from Gemini), almost 4%, against 0.26% on average in France; most land directly on a service page.
These results remain modest: they cover a few weeks and small sites. They show a trend, not a guarantee. The details of the content sites are in the case study AI-driven editorial websites.
What is useless (or nearly)
Many “AI SEO” offers sell recipes that the engines themselves contradict. Google writes it in its Search Central documentation: there are no additional requirements and no special optimisation needed to appear in its AI answers, and there is no need to create AI-specific files or markup.
- The llms.txt file is not used by Google Search. I put one on datapet.fr because it takes ten minutes and may help some AI agents understand a site, but it is not what will get you cited.
- Mass-produced AI content, with no review and no information of its own, is targeted by Google’s spam policy (“scaled content abuse”). Pages churned out in series get neither good rankings nor citations.
- Repeating the keyword or writing for a robot: an AI looks for a relevant answer, not a keyword density.
- Paying to “be in ChatGPT”: no agency and no expert can guarantee a mention, any more than a first place on Google.
Tools to measure your visibility in AI engines
There is no complete dashboard yet, but four free tools give a good idea of your visibility. Search Console counts appearances in AI Overviews together with the rest of your web traffic. Bing Webmaster Tools added an “AI Performance” report in 2026 that shows how often Copilot cites your pages. To go further, paid tools automate citation tracking.
- Google Search Console: impressions and clicks, AI answers and AI Mode included, with no way to isolate them.
- Bing Webmaster Tools, AI Performance report: pages of your site cited in Copilot and Bing’s summaries, and since June 2026, the related intents and topics.
- Your analytics tool (Matomo, Google Analytics): visits from chatgpt.com, perplexity.ai or gemini.google.com. Part of them arrive with no referrer and are counted as direct traffic.
- A manual test every month: ask ChatGPT, Perplexity and Gemini the same ten prospect questions, in a private window, and note whether your site appears. Fifteen minutes are enough, and this tracking gives clear leads on the pages to improve.
- An automatic tracking tool, if you do not want to do it by hand: GreenRed, the French tool I use for datapet.fr, asks your ten prospect questions to ChatGPT, Claude, Gemini, Perplexity and Mistral, counts how often your site is cited, and shows the competitors and sources mentioned instead of you (from €69 excl. VAT per month, 14-day free trial; site in French).
Using AI for your own SEO
AI is not only the engine to convince: it is also a working tool. Used well, it helps analyse what your customers expect, structure an article or write a first draft. Used badly, it produces interchangeable texts that no engine keeps. The rule: the AI prepares, you bring the expertise and the review.
- Finding topics: ask ChatGPT or Claude what a customer wonders before buying your service, then compare it with what you actually hear.
- Analysing the competition: ask the AI engines your prospects’ questions and note which sites come up, and why.
- Writing faster: an outline, a meta description, an FAQ from your notes. The final text must contain your data and your way of saying things.
- Automating the tracking: a workflow (Make, n8n) can collect your positions or your AI visits every month and send you a table, one less task to do by hand.
On my content sites, the AI writes and I approve every text before publication; on a client site, an AI website editor lets you edit your pages by chatting, without anything being published without your approval. To train a team in these uses, I run an AI workshop.
How much does AI SEO cost for a small business?
The essentials are not expensive: it is mostly about quality writing time, on the pages that matter for your business. The return is measured in contact requests, not in number of pages. Before committing a budget, check what an offer brings compared with good SEO: if it only talks about special files or keywords, walk away.
- Doing it yourself: free, thanks to the tools listed above, but it takes time to write and keep pages up to date.
- Having the foundations checked: an express audit at €290 excl. VAT with me, with a prioritised list of fixes.
- Having a text rewritten or optimised: billed by the hour, €75 excl. VAT per hour, from your information.
The right strategic choice for a small business is to adapt first the two or three pages that already bring in contacts, then move forward step by step based on the results.
Getting started as a small business
There is no need to redo everything: start by checking that your site is technically sound, then improve first the pages that answer your customers’ needs. A service page with a price, a case study with figures and an honest FAQ are a more effective lever than twenty generic blog posts, and none of it requires a large budget or a marketing agency.
- Check indexing in Search Console and Bing Webmaster Tools, and make sure your robots.txt file blocks neither OAI-SearchBot nor Bingbot.
- List the ten questions your prospects really ask, and check that a page on your site answers each one clearly.
- Add your data: prices, lead times, results obtained for your clients, with their date.
- Show the author and the date on important pages, and keep the ageing ones up to date.
- Look after your presence beyond the site: Google profile, reviews, social networks, directories.
- Measure every month, then strengthen the pages that never appear.
To find out where your site stands, an express website audit checks these technical foundations and gives you a prioritised list of fixes. If you are starting from scratch, a custom website built with AI includes everything described here from launch. For an online store, the stakes go further: see the article on agentic commerce.
Sources
- Google Search Central, AI features and your website (accessed 06/11/2026)
- Google Search Central, Creating helpful, reliable, people-first content
- Google Search Central, Spam policies (scaled content abuse)
- Ahrefs, one month of AI Overviews in France and already a 23.1% drop in click-through rate (17/08/2026, in French)
- SE Ranking, AI traffic trends in France (21/06/2026, in French)
- OpenAI, Overview of OpenAI Crawlers
- Bing, New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare (June 2026)
- GreenRed, GEO tracking (in French)




