Contents
What is GEO (Generative Engine Optimization)?
GEO, or Generative Engine Optimization, covers every action that makes a brand visible in AI engine answers: being mentioned, recommended and cited as a source when someone asks ChatGPT, Gemini, Claude or Google AI Overviews a question.
The term comes from a Princeton research paper published in 2023, which showed that the visibility of content in generative answers can be raised by up to 40% by working on its form and its evidence [9]. You will also come across AEO, LLM SEO or LLMO: they are variations of the same goal. See our GEO definition.
How an AI engine chooses the sources it cites
1
Prompt
The user asks a question in natural language.
2
Fan-out
The engine splits it into several sub-searches.
3
Retrieval
It queries a web index: Google, Bing or its own.
4
Selection
It keeps clear, reliable and up-to-date passages.
5
Answer
It writes a summary and cites a few sources.
Technical gets you indexed and readable (steps 2 and 3). Content makes your passage the one selected (step 4). Authority weighs on the selection through what others say about you (steps 4 and 5). Simplified diagram [24].
GEO vs SEO: what changes and what stays
Criterion | SEO | GEO |
|---|
Where | Google and Bing results pages | ChatGPT, Gemini, AI Overviews, AI Mode, Claude, Copilot, Perplexity answers |
Unit tracked | Keyword | Prompt (a natural-language question, often long) |
Goal | A position, a click | A mention, a recommendation, a citation |
Key authority signal | Backlinks | Brand mentions on trusted sources |
Measurement | Stable from one day to the next | Probabilistic: prompts must be repeated |
Shared foundation | Indexing, useful content, E-E-A-T, structured data | Good SEO remains the fuel of GEO |
Why build a GEO strategy now?
Because search is shifting towards generated answers, in France as elsewhere. And those answers leave less room for links.
900M
weekly ChatGPT users in February 2026
OpenAI via TechCrunch [1]
Jul. 22, 2026
launch of AI Overviews and AI Mode in France
Blog du Modérateur [3]
48%
of French people use generative AI, 73% of them to look for information
Arcep, Digital Barometer 2026 [8]
When an AI summary appears, people click half as often
Click on a source in the summary
1%
Share of Google visits with at least one click on a link. 900 US adults, 68,879 searches, March 2025. Pew Research Center [4]
Ahrefs measures a 58% drop in the click-through rate of position 1 when an AI Overview is shown [5]. On Google's side, AI Mode passed one billion monthly users a year after launch [2].
The good news: being cited changes everything. According to Seer Interactive, a brand cited in an AI Overview gets 120% more clicks per impression than a brand missing from the answer [6]. And this traffic converts: visitors coming from ChatGPT convert at 7% on transactional sites, versus 5% from Google [7].
A nuance: to look for information, French people still use search engines twice as much as generative AI (59% vs 28%) [8]. GEO does not replace SEO, it adds to it.
What a well-run GEO strategy delivers
Three BotRank clients, three different contexts, the same method: define the strategic prompts, produce the missing content, fix the technical side, strengthen external authority.
Smash
45% → 70%
presence in the answers of the main AI engines on priority queries, with work on Reddit and page structure.
Read the case study →IGENSIA Education
25% → 55%
visibility for IGENSIA RH on strategic prompts such as "which HR work-study bachelor programs are the most recognized?".
Read the case study →Conectic+
+70%
traffic from AI engines in 6 months and +120% revenue generated through ChatGPT, Gemini and Claude (GA4 data).
Read the case study →The method: three pillars and continuous measurement
As in SEO, a GEO strategy rests on three pillars. Tackle them in this order, from the fastest to the slowest to pay off. Everything sits on regular measurement.
PILLAR 1
Technical
Make the site accessible, readable and understandable for AI crawlers.
Low effortResults in weeks
PILLAR 2
Content
Create or enrich the pages that answer your customers' prompts.
Medium effortResults in 1 to 3 months
PILLAR 3
External authority
Be mentioned and well perceived on the sources AI engines consult.
High effortResults in 3 to 6 months
Foundation: measurement
Prompt panel · visibility rate · share of voice · cited sources · sentiment
First step: measure your starting AI visibility
A GEO strategy starts with a quantified baseline: which questions your brand shows up on, against which competitors, and which sources AI engines cite instead of you.
Asking a question once is not enough. SparkToro had the same requests run nearly 3,000 times on ChatGPT, Claude and Google: the same list of brands came back less than once in 100, in the same order less than once in 1,000 [26]. Hence a percentage-based indicator, calculated on a panel of repeated prompts.
Building your prompt panel
- Start from your SEO queries that convert and rewrite them as full questions, the way a customer would ask them.
- Cover every intent: discovery ("how to choose…"), comparison ("X or Y?"), recommendation ("what is the best…"), brand ("reviews of X").
- Tag by product range, persona or market to read results by segment.
- Aim for 50 to 300 prompts depending on the size of your offer, tracked daily on the engines that matter to you.
In BotRank: Prompt Studio generates and organizes your prompts, and Visibility Analysis tracks them daily on the real interfaces of AI engines.
Discover Prompt Studio →1
Technical pillar: open the door to AI engines
The fastest building block: everything happens on your site, which you control.
Allow the right AI crawlers
Each AI company runs several crawlers with distinct roles. Blocking a training crawler does not stop you from appearing in answers. Blocking a search crawler does.
Swipe to see the whole table →
Crawler | Company | Role | Recommendation |
|---|
OAI-SearchBot | OpenAI | Appear in ChatGPT search | Allow |
ChatGPT-User | OpenAI | Visits triggered by a user request | Allow |
GPTBot | OpenAI | Model training | Editorial choice |
Claude-SearchBot, Claude-User | Anthropic | Search and user requests | Allow |
ClaudeBot | Anthropic | Training | Editorial choice |
PerplexityBot | Perplexity | Search index | Allow |
Google-Extended | Google | Gemini training (does not affect AI Overviews) | Editorial choice |
Bingbot | Microsoft | Bing index, used by Copilot and ChatGPT | Allow |
Sources: OpenAI documentation [15], Anthropic [16]. How to configure it: our robots.txt guide.
Common trap
Since July 2025, Cloudflare blocks AI crawlers by default on new domains [23]. Check your CDN, firewall and robots.txt: many sites are invisible to AI engines without knowing it.
Serve content that can be read without JavaScript
The crawlers of ChatGPT, Claude and Perplexity do not execute JavaScript: only Google (Gemini) and Apple do [17]. A price, a spec sheet or an FAQ loaded client-side simply does not exist for them. Render the essentials in the initial HTML (server-side rendering or pre-rendering).
Don't neglect Bing
Microsoft and OpenAI AI engines rely on the Bing index. A Seer Interactive analysis shows that 87% of ChatGPT search citations match Bing's top 20 [18]. Register your site in Bing Webmaster Tools, which now reports your citations in Copilot [19], and enable IndexNow so your updates are picked up quickly.
Mark up with structured data
Google states that no special markup is required for its AI features [20]. Microsoft, however, confirms that schema helps its models understand content [21]. Prioritize the types that describe your entity and your offer: Organization, Product and Offer, Article with author, FAQPage, LocalBusiness. The markup must always reflect the visible content. More on structured data.
Take care of SEO fundamentals
AI Overviews and AI Mode rely on the Google index. But with query fan-out (the AI splits a question into dozens of sub-searches) [24], only 38% of AI Overview citations now come from Google's top 10 [13]: well-indexed long-tail pages have a real chance.
What about llms.txt?
Proposed in 2024 to guide LLMs, it is not used by any major AI engine to date. John Mueller (Google) compares it to the keywords meta tag [22]. It costs almost nothing to publish but should never come before the points above. See our llms.txt technical guide.
✓
AI search crawlers allowed in robots.txt, CDN and firewall
✓
Key content (prices, specs, FAQ, reviews) present in the HTML without JavaScript
✓
Site verified in Google Search Console and Bing Webmaster Tools, IndexNow enabled
✓
Organization, Product/Offer, Article and FAQPage schema consistent with visible content
✓
XML sitemap up to date with real modification dates
✓
Good server response time, no key page in error or noindex
2
Content pillar: become the answer
Produce the missing pages and enrich existing ones: articles, guides, product pages.
Start from prompts, not keywords
Questions asked to AI engines are long and full of context. For each prompt where you are absent, identify the page that should answer it. If it does not exist, that is your production priority. AI Mode is making queries longer: the longer the query, the more doors fan-out opens.
Answer first, then expand
- Under each heading, a direct 40 to 60-word answer that makes sense out of context.
- Question-style headings and self-contained sections: AI engines extract passages, not pages.
- Tables for comparisons, lists for steps.
Bring evidence
In the Princeton study, adding source citations raises visibility by 27% and adding statistics by 25%. Keyword stuffing, on the other hand, lowers it [9]. Sourced figures, named expert opinions, proprietary data: this is what AI engines pick up.
Favor formats that get cited
"Best X" style lists account for 43.8% of the pages cited by ChatGPT in the Ahrefs study [14]. Our own data points the same way.
BotRank data
On the GEO tools market in France, 27 of the 30 pages most cited by AI engines over the last 30 days are comparisons or rankings ("the 10 best GEO tools in 2026"…). Vendors' own product pages are rarely cited on these questions.
BotRank data, French prompts on the GEO market, Aug. 25 to Sept. 24, 2026.
Other formats that work: practical guides, definition pages, honest comparison pages (including against your competitors), studies with original data.
Update regularly
Pages cited by AI assistants are on average 25.7% fresher than those in organic results, with ChatGPT being the most sensitive to freshness[12]. Show a real update date and review your strategic pages every quarter.
E-commerce: complete product pages
AI engines recommend products from merchant feeds as much as from the web: ChatGPT relies on product feeds and the Agentic Commerce Protocol [25], Google on its Shopping Graph. Complete attributes (material, dimensions, uses), up-to-date price and stock, visible reviews, a clean Merchant Center feed. Read: ChatGPT shopping favors product feeds and agentic commerce in 2026.
In BotRank: AI Shopping tracks which of your products AI engines recommend, at which merchants and at what price.
Discover AI Shopping →Clarify your entity
A factual About page, the same one-sentence description on your site, LinkedIn and directories, signed expert pages: AI engines must be able to say in one sentence who you are and who you serve. More on entity SEO.
✓
Every priority prompt has a page that answers it directly
✓
A 40 to 60-word answer under each H2, self-contained sections
✓
Statistics and sources cited, author identified
✓
At least one comparison or "how to choose" guide per product range
✓
Visible update date, quarterly review scheduled
✓
Complete product pages and up-to-date merchant feeds
To go further on writing: our 10 GEO writing best practices.
In BotRank: Recommendations and the GEO Roadmap prioritize the pages to create, and Bob, the GEO agent, drafts and optimizes your content.
See Recommendations →3
External authority pillar: what the web says about you
AI engines cross-check their sources. What others say about your brand weighs as much as what you say yourself.
Brand mentions before backlinks
Across 75,000 brands, Ahrefs measures a 0.66 correlation between a brand's web mentions and its visibility in AI Overviews, versus 0.22 for the number of backlinks [10]. A link is still useful, but a brand mention in the right context counts for more. See also our guide on brand authority.
Identify the sources that matter in your sector
There is no universal list. The sources cited change from one sector to another.
Sector tracked by BotRank | Dominant sources in AI answers | Priority action |
|---|
Insurance | Insurers' sites, comparison sites, service-public.fr, local directories | Be present and up to date on comparison sites |
Real estate networks | Network sites, service-public.fr, directories, listing portals | Rich local pages, complete directory listings |
B2B software (GEO tools) | Comparisons and rankings on specialized blogs, YouTube | Get included in the cited comparisons |
BotRank data, last 30 days, anonymized sectors. Read also: how AI engines choose their sources.
In BotRank: Source Analysis lists the domains and exact pages AI engines cite on your prompts.
See Source Analysis →Be present on the platforms AI engines read
- YouTube: the most cited domain in AI Overviews in France in our study (8.4% of citations), and the signal most correlated with AI visibility according to Ahrefs [11].
- Reddit and forums: Reddit is the most cited source in our Top 100 LLM sources. Take part under your real name, without disguised promotion.
- LinkedIn: 5th source in the Top 100, especially for B2B queries.
- Google Business Profile: 2nd source of AI Overviews in France, decisive for local search.
Work on perception: customer reviews and sentiment
Trustpilot is the 4th source in our Top 100 LLM sources. When an AI engine describes your brand, it often summarizes your reviews. Ask for reviews regularly on the platforms of your sector, answer criticism and check that AI engines retain your real strengths.
Press relations and proprietary studies
National and regional press are among the cited sources (Le Figaro is 6th in the Top 100 LLM sources, Ouest-France and Le Parisien are in the Top 10 AI Overviews sources). The best lever to appear there: publish original data that journalists can reuse.
Wikipedia: with caution
Wikipedia is still widely read by models, but its notability rules are strict. Do not write your own page. If your brand is eligible, check that public information is accurate and that sources exist.
✓
Top 20 sources cited on my prompts identified
✓
Presence checked in the most cited comparisons and rankings
✓
YouTube channel with videos that answer customer questions
✓
Active customer review program, replies to negative reviews
✓
One study or proprietary dataset published per semester
Steering: the KPIs of a GEO strategy
Indicator | What it measures | Frequency |
|---|
AI visibility rate | % of answers where your brand appears on the panel | Weekly |
AI share of voice | Your presence compared with your competitors | Weekly |
Average position | Rank of your brand in recommendations | Weekly |
Citation rate | % of answers citing a page of your site | Monthly |
Sentiment | Tone and attributes associated with the brand | Monthly |
AI referral traffic | Sessions and conversions coming from AI engines (GA4) | Monthly |
Beware, GA4 underestimates this traffic: part of the visits coming from AI engines arrive without a referrer. Read: AI Overview traffic is real, but GA4 may be hiding it.
In BotRank: Visibility Analysis tracks these indicators every day, across the AI engines and markets you choose.
See Visibility Analysis →Your 90-day GEO action plan
Day 0 → 15
Measure and audit
- Prompt panel and competitors
- Baseline visibility
- AI technical audit
Day 15 → 45
Fix and produce
- Technical fixes
- 5 to 10 priority pages
- Update key pages
Day 45 → 90
Build authority
- Cited comparisons and media
- Customer reviews, YouTube
- One proprietary study
Ongoing
Steer
- Weekly KPI tracking
- Quarterly content review
- New prompts
Mistakes that hold back a GEO strategy
- Judging on a single answer: AI engines vary, only repeated measurement makes sense.
- Starting with llms.txt instead of opening the site to search crawlers.
- Rewriting the whole site for AI engines: content must stay useful to humans, which is also what AI engines reward.
- Tracking only ChatGPT: sources and results differ widely from one engine to another.
- Buying artificial mentions: fake reviews and undisclosed sponsored content end up hurting brand perception.
Conclusion: the priorities to remember
- Measure your starting visibility on a representative prompt panel.
- Unblock the technical side: search crawlers, readable HTML, Bing, schema.
- Fill content gaps on the prompts where you are absent, with direct and sourced answers.
- Build authority on the sources AI engines cite in your sector.
- Steer every week and adjust.
BotRank.ai
From analysis to action
BotRank.ai is the all-in-one GEO platform that helps marketing and SEO teams measure and improve their brand's AI visibility.
Frequently asked questions about GEO strategy
What does GEO mean?
GEO stands for Generative Engine Optimization. It covers every action that makes a brand visible in the answers of AI engines such as ChatGPT, Gemini, Claude or Google AI Overviews.
What is the difference between SEO and GEO?
SEO aims for a position in a list of links, GEO aims for a mention or a citation in a generated answer. Both share the same foundation (indexing, useful content, authority), but GEO is measured on prompts and gives more weight to brand mentions than to backlinks.
Is SEO still useful with GEO?
Yes. AI engines rely on the Google and Bing indexes to find their sources. A poorly indexed or slow page has little chance of being cited.
How long does it take to see GEO results?
Technical fixes often pay off within a few weeks, content within one to three months, external authority within three to six months. At Smash, visibility went from 45% to 70% in a few months.
Is an llms.txt file essential?
No. No major AI engine has confirmed using it. You can publish one, but opening the site to AI search crawlers and making content readable without JavaScript has far more impact.
How do I measure my visibility in ChatGPT and other AI engines?
By tracking a representative prompt panel, queried regularly on each engine, then calculating a visibility rate, a share of voice against competitors and the list of cited sources. A GEO tool automates this collection.
What are the 3 pillars of a GEO strategy?
Technical (crawler access, readability, structured data), content (pages that answer prompts directly) and external authority (mentions, reviews, presence on the sources AI engines cite).
Do I need an agency or a tool to do GEO?
A tool is enough to measure and prioritize if you have a marketing or SEO team in-house. To build skills, a GEO training course delivered by our expert partners such as Pilot'in is an option.
Sources
[1]
ChatGPT reaches 900M weekly active usersTechCrunch, Feb 27, 2026
[2]
Search updates at I/O 2026Google, May 2026
[3]
Google déploie AI Overviews et AI Mode en FranceBlog du Modérateur, Jul. 2026
[4]
Google users are less likely to click on links when an AI summary appearsPew Research Center, Jul. 22, 2025
[5]
AI Overviews reduce clicks (update)Ahrefs, Feb. 4, 2026
[6]
AIO impact on Google CTR: 2026 updateSeer Interactive, Apr. 24, 2026
[7]
Gen AI statisticsSimilarweb, 2026
[8]
Baromètre du numérique, édition 2026Arcep / CREDOC, Feb. 9, 2026
[9]
GEO: Generative Engine OptimizationAggarwal et al., Princeton, KDD 2024
[10]
AI Overview brand correlationAhrefs, 2025
[11]
YouTube mentions are the strongest signal of AI visibilityAhrefs via Business Wire, May 26, 2026
[12]
Do AI assistants prefer to cite fresh content?Ahrefs, Jul. 28, 2025
[13]
AI Overview citations and the top 10Ahrefs, 2026
[14]
Best lists researchAhrefs, 2025
[15]
Overview of OpenAI crawlersOpenAI, 2026
[16]
Anthropic clarifies Claude botsSearch Engine Land, Feb. 25, 2026
[17]
The rise of the AI crawlerVercel and MERJ, Dec. 17, 2024
[18]
87% of SearchGPT citations match Bing's top resultsSeer Interactive, Feb. 6, 2025
[19]
AI Performance in Bing Webmaster ToolsMicrosoft Bing, Feb. 2026
[20]
AI features and your websiteGoogle Search Central, Dec. 2025
[21]
Microsoft: schema helps Bing and Copilot LLMsSearch Engine Land, Mar. 2025
[22]
Google says llms.txt comparable to keywords meta tagSearch Engine Journal, Apr. 17, 2025
[23]
Cloudflare changes how AI crawlers scrape the internetCloudflare, Jul. 1, 2025
[24]
AI Mode in Google SearchGoogle, May 2025
[25]
Powering product discovery in ChatGPTOpenAI, Mar. 24, 2026
[26]
AIs are highly inconsistent when recommending brandsSparkToro, Jan. 28, 2026
BotRank studies: Top 100 sources cited by LLMs (1.2M answers) and Top 10 Google AI Overviews sources (50,000+ answers in France).
