GEO: optimising your website for AI engines (ChatGPT, Perplexity, Google AI overviews)
A growing share of searches no longer goes through a list of ten blue links but through a written answer: an AI overview at the top of Google, a ChatGPT or Perplexity reply with a few cited sources. For a business, being one of those sources is becoming as important as ranking on page one. That is the purpose of GEO, Generative Engine Optimization. This guide explains what these engines actually look at, what needs to change on a website (often less than you would think) and how to measure progress. It draws on what we apply to our own website and to our clients' sites.
What GEO is, and what it is not
GEO covers the practices that increase the likelihood that a generative answer engine uses and cites your website. It concerns Google's AI overviews and AI Mode, ChatGPT with web search, Perplexity, Microsoft Copilot and the assistants built into browsers. It is not a discipline separate from SEO: these engines rely on classic search indexes (Google's, Bing's) to find their sources, then select the clearest and most reliable pages to write from. A website invisible on Google and Bing is invisible to AI. Conversely, a well-ranked site with confusing writing will be read but not cited.
The difference with classic SEO
SEO targets a position in a list. GEO targets a citation in an answer. In a list, the title and meta description do the work; in an answer, the precise passage that answers the question is what matters. Generative engines split your pages into fragments and keep those that answer directly, with facts, figures and definitions. The work therefore bears on content structure and precision, more than on tags.
Why it matters now
Commercial-intent searches ("which agency to build a Shopify store in Aix-en-Provence", "how much does an SEO audit cost") are precisely those where assistants recommend providers by name. Businesses cited today build a lead others will have to catch up on, because engines readily reuse sources they have already judged reliable.
How answer engines choose their sources
Three families of criteria emerge from observations and from documentation published by the vendors. Relevance: the fragment answers the exact question, using the same terms as the user. Reliability: the website has a clear identity (who writes, which company, which address), sources, dates, consistency with what other sites say. Accessibility: the content is available as raw HTML, without a login wall, without crawler blocking, without depending on JavaScript rendering. Add to this authority in the classic sense: a site cited elsewhere, with quality inbound links, is preferred to an isolated one.
Fragments rather than pages
When an assistant cites a page, it has extracted a passage of a few sentences. The pages that win are those where each section can be read on its own: a heading that poses the question, a direct answer in two or three sentences, then the nuances. A page that buries information in a long preamble is rarely selected, even if it is thorough.
Consistency across sources
Engines cross-check. If your website states a price, an address or a founding date different from your Google listing, LinkedIn or a directory, trust drops. GEO starts with a consistency check of your business information across the web, which overlaps with local citation work.
Step 1: make the website accessible to AI crawlers
Many websites block AI crawlers without knowing it, through a CDN setting or an overly restrictive robots.txt file. Before any editorial work, check this point.
The robots.txt file
The main crawlers to allow are GPTBot and OAI-SearchBot (OpenAI), PerplexityBot, ClaudeBot (Anthropic), Google-Extended (use by Gemini) and Bingbot, on which ChatGPT and Copilot depend. A website that allows everyone has nothing to change. A site that has added blocking rules must decide knowingly: blocking training is a legitimate choice, blocking search means giving up citations.
CDN-level blocking
Cloudflare and other CDNs offer one-click AI crawler blocking, sometimes enabled by default on new accounts. We found this on our own website: correct robots.txt, but AI crawlers blocked at CDN level. Check your CDN's "AI bots" section and disable blocking if you want to be cited.
The llms.txt file
The llms.txt file, placed at the root of the site, is a Markdown page that presents the business and lists its important pages with a one-sentence description each. Adoption by engines is still partial, but it costs nothing, it helps assistants understand the site's structure and it forces you to state clearly what you do. You can use ours, at ecommercedynamics.fr/llms.txt, as a template.
Content in HTML, not only in JavaScript
AI crawlers read HTML as served. A website whose content only appears after JavaScript execution (some single-page applications, some site builders) is largely unreadable to them. Server-side rendering or static generation solves the problem; it is one of the reasons we rebuilt our own website in server-rendered PHP.
Step 2: structure content to be cited
This is the core of GEO and the part that requires the most editorial work. The rules are simple to state and demanding to apply across a whole site.
A direct answer under every heading
Every section heading should match a question a customer could ask, and the first two or three sentences should answer it without detour. Nuances, examples and special cases come next. Re-read each section asking: if an assistant cited only these three sentences, would they be accurate and complete?
Dated, sourced facts
Engines favour content that gives precise figures, dates, tool names and verifiable sources. "A fast website converts better" is an opinion; "beyond three seconds of load time, conversion drops, according to measurements published by Google" is a citable fact. Show the update date of your pages: assistants prefer recent information.
FAQs, with real questions
A frequently asked questions section is the most easily reused format, provided the questions are those customers actually ask (in their words) and the answers are self-contained. Six to eight questions per service page, with three- to five-sentence answers, is a good balance. Avoid rhetorical questions that only serve to place a keyword.
Definitions and comparisons
Assistants love definitions ("Link building consists of…"), structured comparisons (Shopify or WooCommerce, manual publication or synchronisation) and step lists. These formats match the most frequent questions and lend themselves to clean extraction. Our Shopify vs WooCommerce comparison is built on this principle.
An identified author
Sign your content with a name, a role and a presentation page. Generative engines, like Google, give more credit to a text whose author and credentials are known. On a business website, a complete about page and a biography of the founder or team are reliability signals that weigh in source selection.
Step 3: structured data and entity signals
Structured data (schema.org, in JSON-LD format) describes your business, pages and content in a language machines read without ambiguity. It does not guarantee a citation, but it removes doubts: who the organisation is, where it is, who the author is, what the page is about. The most useful types for a service business: Organization or LocalBusiness with address, phone, service area and links to official profiles (sameAs), Person for the author, Article for editorial content with publication and update dates, FAQPage for FAQs, Service with price for offers. Consistency between this data, the visible content and your external profiles (Google listing, LinkedIn, directories) forms what is called a strong entity signal. A complete SEO audit checks these points and their technical validity.
Step 4: measure your visibility in AI engines
There is no Search Console equivalent for generative engines yet. So you measure with indirect means, which are enough to follow a trend.
Manual tests
Build a list of ten to twenty questions your customers ask (with local variants if you are a local business) and query ChatGPT, Perplexity and Google AI Mode every month. Note whether you are cited, in which position, and which pages are used. A simple table is enough; the trend matters more than the absolute level.
Referral traffic from assistants
In Google Analytics 4, filter sessions whose source contains chatgpt.com, perplexity.ai, copilot.microsoft.com or gemini.google.com. These visits remain few but they are highly qualified: the user arrives with a recommendation. Create a segment or a dedicated exploration to track them over time.
Server logs
AI crawlers identify themselves in the server access logs (GPTBot, PerplexityBot, ClaudeBot, etc.). Their visits prove the site is accessible and show which pages they read. Bing Webmaster Tools also provides useful data, since ChatGPT and Copilot rely on Bing's index.
What GEO does not change
GEO replaces neither useful content, nor authority, nor local SEO. A website cited by assistants is first of all a well-ranked website, with pages that answer real questions, quality inbound links earned through clean link building, visible customer reviews and an up-to-date Google listing. Techniques that promise to "manipulate" AI (hidden text, instructions injected into pages, mass-generated content) are detected and harm the site. The right approach is the same as in SEO: genuinely be the best source on your topic, and make it easy to see. That is the spirit of our SEO and GEO optimisation offer, which handles both together.
Key takeaways
- AI engines draw from Google's and Bing's indexes: GEO starts with good SEO
- Check that AI crawlers are blocked neither by robots.txt nor by your CDN
- Every section must answer a question directly in two or three sentences
- Dated facts, sourced figures, identified author: these are the reliability signals that get selected
- Structured data and consistent business information form your entity signal
- Measure with monthly question tests and assistant referral traffic
Checklist
- robots.txt: GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended allowed
- AI crawler blocking disabled at CDN level
- llms.txt file at the root, up to date
- Content readable in HTML without JavaScript
- Direct answer under every section heading
- Real FAQs with self-contained answers
- Publication and update dates displayed
- Identified author with presentation page
- Valid Organization, Person, Article, FAQPage, Service structured data
- Consistent business details on the site, Google listing, LinkedIn and directories
- Test question list and monthly citation tracking
- GA4 segment for AI assistant referral traffic
Frequently asked questions
Does GEO replace SEO?
Should I allow AI crawlers to read my website?
Is the llms.txt file mandatory?
How long before ChatGPT or Perplexity cites my site?
How do I know whether my website is cited by AI?
Can a small local business get cited?
Is your website readable by AI engines?
We check AI crawler accessibility, structured data and content clarity, then fix what blocks. SEO and GEO are handled together.
Discover SEO and GEO optimisation