What is GEO and how do AIs decide to recommend my business?
The essentials: GEO (Generative Engine Optimization) is the method for being cited by generative AI engines like ChatGPT, Google AI Mode, or Perplexity. Where SEO ranked you in a list of results, GEO makes you appear in a synthesized response — and with 88% of AI citations not coming from the Google top 10, your current ranking is no longer enough.
What you will learn:
- How ChatGPT and Google AI Mode choose which businesses to cite — the query fan-out mechanism that determines your visibility
- The 6 concrete levers to make your business appear in AI responses (without rebuilding everything)
- What does NOT work in GEO (and why counting citations is a vanity metric)
- The weak signals that reveal if an AI is hesitating to recommend you
Before continuing: This article is for SMB directors who already have a website — visible or not on Google — and want to understand how to exist in AI responses too. If you do not have a website at all, start there — GEO does not replace a healthy technical foundation.
Published on June 27, 2026
When Jade looks for a plumber in Avignon in ChatGPT, the response the AI gives her is not a list of 10 blue links — it is a synthesized sentence that cites 2 or 3 business names. The others, even if they rank well on Google, are not mentioned.
With a billion active monthly users on Google AI Mode and queries doubling every quarter, this snapshot is becoming the new norm. The problem: your site can be on Google page 1 for years without ever being cited by an AI.
The problem: The rules of the game are changing. 77% of Google searches no longer generate any clicks to websites. Users stay on Google, read the AI summary, and leave. If your business is not cited in this summary, it is invisible — even on the first page.
The solution: GEO (Generative Engine Optimization) is the method for appearing in these responses. It is not a new SEO. It is its natural extension — and this article gives you the keys to get started.
The proof: Google itself confirmed that AI Overviews use the same ranking systems as traditional search. But the data is clear: only 12% of citations in Google AI Mode match URLs from the organic top 10 (Semrush, analysis of 40,000 queries). Good SEO is necessary, but not sufficient.
How does an AI choose which businesses to recommend?
Let’s start with the mechanism. Unlike Google, which displays a list of results ranked by relevance, a generative AI like ChatGPT or Google AI Mode constructs a response. And to do that, it follows a three-step process.
When a user asks a question (“Which plumber do you recommend in Avignon?”), the AI does not look for a pre-written answer. It breaks down the question into several internal sub-questions: “plumber in Avignon”, “urgent plumber”, “cheap plumber”, “Google reviews plumber Avignon”. This is called query fan-out — a mechanism validated by the Ahrefs study of 1.4 million ChatGPT prompts.
The AI then retrieves web pages via Google’s traditional ranking systems (crawling, indexing, ranking signals). And this is where the logic changes: it doesn’t keep the top 10 results. For each sub-question generated, it compares the title and snippet of each page to its question. Only the pages with the best matching title are retained.
“Cited pages have a cosine similarity of 0.656 between their title and the AI’s sub-question — compared to 0.484 for uncited pages.” — Ahrefs, study on 1.4 million ChatGPT 5.2 prompts (May 2026)
What this means in practice: your page title is what determines whether the AI reads you or ignores you. A title like “Plumbing Services — Avignon” has little chance of matching a sub-question. “Which plumber to call for an emergency in Avignon?” statistically has a much higher chance.
The Ahrefs study confirms this: natural language titles achieve an 89.78% citation rate compared to 81.11% for classic SEO titles. And about 50% of the URLs retrieved by the AI are cited — the remaining 50% are discarded solely based on the title and snippet.
What I learned in the field: In talking with SMBs that already appear in ChatGPT (without knowing it), I noticed a common point: their pages have section titles that answer specific questions. Not generic “Our Services” — but “How much does a leak repair cost?”. The AI looks for answers, not menus.
Why does being first on Google no longer guarantee being found?
This is a question I receive almost every week. A business owner tells me: “I am number 1 on Google for my city and my job. Yet, when people look for me on ChatGPT, I do not appear.”
The figures speak for themselves. Semrush analyzed 40,000 queries on Google AI Mode: only 12% of citations match URLs that are in the organic top 10. It is a massive gap.
Why this gap?
The logic of ranking (SEO) and the logic of sourcing (GEO) are different. SEO ranks pages by relevance for a query. GEO chooses reliable sources to build an answer. An answer engine does not ask “what is the best page on this topic?” but “which source do I trust to answer THIS specific question?”
The AI looks for content that directly answers a specific question — not a service page that skims the topic. If your plumbing page talks about everything (emergency repairs, quotes, installation, maintenance), the AI will have difficulty identifying it as a source for the question “How much does an emergency repair cost?”. Specificity becomes more important than general authority.
What this changes in practice
| What worked in SEO | What matters in GEO |
|---|---|
| Generalist page well ranked | Direct answer from the 1st paragraph |
| Title optimized for keyword | Title in the form of a natural question |
| Long content covering the whole topic | Specific depth > wide coverage |
| Backlinks = domain authority | Topical authority + applicable answer |
A restaurant with 1,200 reviews, 4.8 stars, and a perfectly optimized Google profile can see ChatGPT recommend its competitors instead. The divide is not technical — it is semantic.
What I learned in the field: A landscaper client had a technical score of 85/100 and ranked 3rd on Google for “landscaper Toulouse”. In ChatGPT, he was never cited. The reason? His homepage talked about “custom garden creation” — not a direct answer to the question people ask (“How much does a landscaper cost in Toulouse?”). We added an immediate answer block at the top of the page. In 3 weeks, he appeared in ChatGPT Search suggestions.
How does query fan-out work, the invisible engine that decides your visibility?
Query fan-out is the most important mechanism to understand in GEO — yet the least known. It is what determines, technically, whether your content will be cited or ignored.
When a user asks an AI a question, the AI:
- Breaks down the main query into 3 to 10 internal sub-questions (the fanout queries)
- Searches for web sources for each sub-question
- Compares the title and snippet of each source to its sub-question
- Selects the sources whose titles match best
- Synthesizes the final answer from the selected sources
The Ahrefs study mapped this mechanism on ChatGPT 5.2: for 1.4 million prompts analyzed, the AI generates an average of 33 fanout queries per user request. Only ~16.5 URLs are retained — the other ~16.5 are discarded based solely on the title and snippet.
The 3 factors that determine if your page passes the filter
- Title/fanout query cosine similarity: the more your title resembles the question the AI is asking itself, the more likely you are to be retained. A score of 0.656 for cited pages versus 0.484 for uncited pages.
- Natural language of the title: natural language titles (questions, complete sentences) are cited at 89.78% compared to 81.11% for classic SEO titles (keywords only).
- Page age: the median age of cited pages is 500 days. Freshness is not enough — semantic relevance takes priority.
“A page can be on Google page 1 for years without ever being cited by an AI. The difference lies in the title and the content format, not in the ranking.” — EarlySEO, optimization method for AI Overviews (May 2026)
What this means for your site
Each page of your site must anticipate the sub-questions the AI will ask about your trade. Specifically:
- Your “Plumbing Avignon” page must answer “How much does a plumber cost in Avignon?”, “How to find a reliable plumber?”, “Cheap plumber Avignon”
- Your headings must be questions — not “Our Services” but “What are our plumbing rates?”
- The first paragraph must give the answer — not set the context
| Classic Service Page ⌠| GEO Optimized Page ✅ |
|---|---|
| Title: “Plumbing Avignon — Repairs and Installation” | Title: “How much does a plumber cost in Avignon?” |
| Intro: “We are a plumbing company based in Avignon…” | Intro: “Plumbing repairs in Avignon cost between €80 and €150 depending on the urgency…” |
What are the 6 levers to appear in AI responses?
GEO levers do not replace SEO — they complement it. Google itself published its official optimization guide for generative AI in May 2026, confirming that AI Overviews use the same ranking systems as traditional search. The 6 levers below are therefore to be applied in addition to good SEO practices.
1. Direct answer in the first paragraph
The AI does not read your entire page before deciding whether to cite it. It scans the beginning looking for an answer to the question it is asking. The first paragraph of each page must contain the answer, not an introduction.
“Google has confirmed that content that answers directly in the first paragraph is favored in AI Overviews citations.” — Google Search Central, official optimization guide for generative AI (May 2026)
Concrete example:
- ⌠“Welcome to Dupont Plumbing. We have been operating throughout Vaucluse for 20 years. Trust us for your plumbing work…”
- ✅ “Plumbing repairs in Avignon cost between €80 and €150 excluding materials. Our team responds within 2 hours for emergencies.”
2. Headings in the form of a natural question
Query fan-out matches your title with the sub-questions generated by the AI. The more your title looks like a question a human would ask ChatGPT, the more likely you are to be retained.
Application on your site:
- The homepage → secondary heading “What is the actual cost of a landscaper in Toulouse?” (even if the H1 remains your business name)
- Service pages → “What rates for a leak repair in Avignon?”
- Pricing pages → “How much does a boiler installation cost?”
The Google Search Central guide for generative AI confirms that question-formatted headings improve cosine similarity with fanout queries.
3. Structured and machine-readable content
AIs analyze HTML, not your design. I already talked about this in my article on the reasons why a slow site loses clients. Google has rejected the idea of using markdown for AI content — HTML is critical for crawlers. Your pages must use:
- Clear and descriptive
<h1>,<h2>,<h3>tags - Lists (
<ul>,<ol>) for enumerations - Tables for comparative data
- Structured JSON-LD (FAQ, LocalBusiness, Product)
The BrightEdge study (June 2026) shows that AI engines do not use the same sources in the same way depending on their format. Structured content takes priority in synthesized responses.
What I learned in the field: The official Google guide says there is no need to “chunk” content for AI — but it must be structured. A well-written H2 is worth 20 meta descriptions. AIs read headings as entry points to the content.
4. Topical authority via backlinks
Backlinks remain the number 1 correlation factor for Google ranking — and they are also becoming a signal for LLMs. The Semrush study on 1,000 domains confirms: backlinks help traditional SEO AND become an authority signal for AI responses.
Specifically: a citation in a specialized article or a local news outlet carries more weight than dozens of generic backlinks. AIs value sources that have domain authority AND topical authority.
Lily Ray’s (Amsive) rule is crystal clear: AI Overviews cite self-promotional listicles but recommend competitors 69% of the time. A strong link profile protects against competitor recommendations.
5. Reinforced local presence
This is probably the most underestimated lever. Google AI Mode cites heavily via Google Maps and Google Business Profile (GBP) listings. The SE Ranking study of 10,000 AI Mode keywords confirms this: Google.com is the most cited domain in AI Mode responses, primarily via Google Maps and GBP profiles.
Why it matters: a local business with a complete GBP listing (categories, hours, reviews, photos, services) can appear in AI Mode responses without even having an optimized website. The Google profile becomes a direct entry point to AI citations.
What to check on your Google listing:
- Primary and secondary categories correctly filled out
- Recent reviews with replies (the AI scans the exchanges, not just the stars)
- Identified photos (descriptive file name, not
IMG_20240301.jpg) - Services listed with indicative prices
6. Watch out for “weak signals” in AI responses
GEO is not just about counting citations. As Emeric Guisset (Nodiris, Journal du Net) points out: AI share of voice is a vanity metric if it does not measure the quality of the citation.
Weak signals to watch for:
- The AI says of you “this is an interesting option” -> reserve (deficit of perceived authority)
- The AI cites you AFTER your competitors -> you are an alternative, not a reference
- The AI uses modal verbs (“can be relevant if”, “more oriented toward”) -> the AI is not sure about recommending you
- Your business is mentioned only when its name is in the question -> conditional visibility, not organic
These signals are the best indicator of your true level of AI credibility. Counting them is easy. Decoding them requires a qualitative analysis that SaaS tools do not do yet.
What should you absolutely not waste your time on in GEO?
Not everything that glitters is GEO. Some practices are either ineffective or downright dangerous.
Black-hat GEO already exists — and it is risky
Microsoft detected more than 50 Recommendation Poisoning attempts in 60 days on its AI platforms. The same manipulation techniques that plagued SEO (spammy backlinks, mass-generated content) are already emerging in GEO. Google confirmed that its spam policies apply to AI Overviews and AI Mode — manipulating AI responses is spam.
“GEO already has its black-hat. AI visibility manipulation techniques are emerging as fast as historic black-hat SEO techniques.” — Search Engine Journal, “The Grounding Wars” (June 2026)
Things that do not work according to Google itself
Google published an official guide in May 2026 debunking several GEO beliefs:
| Myth | Reality |
|---|---|
| You need an llms.txt file | Purely speculative. Google recommends WebMCP as a standard for the agentic web. |
| You need to “chunk” content | Useless. AIs know how to extract sections from a page. |
| You need to write specifically for AI | Useless. Well-structured HTML is enough. |
| Special GEO schema markup is critical | Ahrefs study (1,885 pages): JSON-LD does not significantly increase AI citations. LLMs use visible HTML. |
| Inauthentic mentions boost visibility | Counter-productive. Google detects and penalizes. |
What is really worth your time
If you had to remember three priorities:
- Structure your pages to answer a specific question in the first paragraph, with a natural language heading
- Optimize your Google Business Profile — this is the main entry point for local AI citations
- Develop topical authority (targeted backlinks, in-depth content) rather than volume
How to check if your business already appears in AI responses?
Before starting, check where you stand. Several tools allow you to measure your current AI visibility.
Free tools
- Google Search Console: AI Overviews are tracked in GSC since mid-2026 (Brodie Clark confirmed the roll-out). Check your impressions.
- Bing Webmaster Tools: A dedicated AI Performance dashboard shows which pages are cited in AI responses, with Citation Share, Intents, and Topics metrics.
- GA4: Google Analytics has added a dedicated “AI Assistant” channel in its referral traffic reports.
Specialized tools
- Semrush AI Visibility Toolkit (formerly Copilot, now Adobe): the most complete, with prompt tracking analysis, AI Share of Voice, and competitor benchmarking
- Profound / Otterly / Limy: AI visibility tracking tools, from cheapest to most advanced
- Manual method: type your trade + city in ChatGPT Search, Perplexity, and Google AI Mode, and note if you appear. Do it every month.
How to establish a GEO action plan for my business?
Here are the concrete actions I recommend to the SMBs I support.
What you need to do:
- Identify your 5 most visited pages on Google Analytics. Are they optimized to answer a specific question in the first paragraph?
- Verify your page headings: are they in natural language? (e.g. “Plumbing Avignon” -> “How much does a plumber cost in Avignon?”)
- Audit your Google Business Profile: categories, reviews, replies, named photos
- Test your appearance in ChatGPT: type your trade + city in ChatGPT Search
- Test your appearance in Google AI Mode: activate it, and search for your trade
- Verify weak signals: if the AI cites you, with what modal verbs? “Interesting option” = needs reinforcement
- Fix the answer block: each service page must have a complete answer within the first 100 words
Summary — GEO checklist:
| # | Action | Done? |
|---|---|---|
| 1 | 5 main pages: heading in natural question | ☐ |
| 2 | First paragraph = direct answer to the question | ☐ |
| 3 | Complete Google Business Profile (categories, reviews, services) | ☐ |
| 4 | ChatGPT Search test: business cited? | ☐ |
| 5 | Google AI Mode test: business cited? | ☐ |
| 6 | Weak signals analyzed: modal verbs, position in response | ☐ |
| 7 | Incoming links from topical authority sources | ☐ |
Interpret your score:
- 0-2: You are starting in GEO. Priority: structure your pages and your Google profile. The rest will follow.
- 3-5: You have the basics. Work on topical authority and weak signals.
- 6-7: You are on the right track. Monitor evolution — GEO moves fast.
Key Takeaways
- GEO does not replace SEO — Google has confirmed it. Both share the same technical foundations. But 88% of AI citations do not come from Google’s top 10.
- Your page heading is your best GEO asset — query fan-out matches the heading with the AI’s sub-questions. A natural language heading doubles your chances of being cited.
- Your Google Business Profile is an AI entry point — AI Mode cites heavily via Google Maps and GBP profiles. Optimize it.
- Do not fall into black-hat GEO traps — 50 poisoning attempts detected in 60 days by Microsoft. The same risks as black-hat SEO apply.
- Weak signals say everything — do not just count your citations. Analyze how the AI talks about you. “Interesting option” ≠reference.
GEO is still young. Few businesses have integrated it. Those that do now will take a lead that their competitors will take months to catch up with.
If you want to check where you stand, I offer a free audit of your AI and Google visibility. We look at your score out of 100, your blocking points, and the 3 actions that will have the most impact. No obligation — we show before we sell.
To go deeper: If you are interested in the topic of Google queries that no longer generate clicks, I wrote a detailed article on how to know if my website is making me money. You will find figures and a method to concretely measure your return on investment.
To go further
- Complete GEO Guide by Backlinko — the reference guide in 7 steps
- Google’s official guide for generative AI — published in May 2026, a must-read
- How to optimize your Google Business Profile for SEO and GEO — complete Oplia guide on the Google profile in 2026
In a world where AI decides who is visible, the best strategy is not to trick it — it is to give it exactly what it is looking for: a clear, structured, and reliable answer.
Need local support? Discover our service areas across France.