SEO for AI: Why Your Website Is Still the Key to Your Visibility in ChatGPT, Gemini and Google

SEO for AI means making your website understandable, credible and citable by generative engines such as ChatGPT, Gemini, Perplexity and Google AI Overviews. Contrary to a common belief, your website remains the cornerstone of that visibility: across more than 4 million AI citations analyzed by BotRank in one month, nearly one in three points directly to a brand’s own website. More than half point to a website owned by a brand or a retailer.

In other words, AI engines do not just read Wikipedia, Reddit or the press. They go to the source, to your pages. Those pages still need to be accessible, clear, up to date and consistent. This guide covers what the data shows, how AI engines read a website, how to optimize your content and technical foundations, and why your choice of CMS has become a real GEO decision, with Jahia as a detailed example.

Why your website is the first source AI engines rely on

A brand’s website is the only source of information it fully controls: offers, prices, specifications, proof points, authors, terms. Logically, that is where AI engines go to check facts before writing an answer.

What the data says: over 4 million citations analyzed

To measure the real weight of websites in AI answers, we analyzed every source cited in the answers tracked by BotRank between September 20 and October 20, 2026: 445,808 answers generated by ChatGPT, Google AI Overviews, Gemini, Perplexity, Claude, Mistral, Copilot and Google AI Mode, representing more than 4.2 million citations. Each cited domain is classified by site type.

Brand websites come first with 30% of citations, ahead of e-commerce and retailer sites (26%), institutional sites (12.8%) and media (7.7%). Social networks, blogs and forums combined account for less than 8%. In France, the share of brand websites even reaches 33.8%.

This is not a one-off. Over the last four months, the share of brand websites stayed between 26% and 30%, and the share of sites owned by brands or retailers between 55% and 59%, across more than 10 million citations.

AI engine	Share of citations to brand websites	Share to brand or retailer websites
ChatGPT	30.1%	54.4%
Google AI Overviews	29.1%	52.5%
Gemini	31.7%	67.1%
Perplexity	27.1%	63.5%
Mistral	33.5%	77.4%
Copilot	34.5%	66.3%
Google AI Mode	36.9%	57.4%
Claude	57.4%	70.0%

Two lessons stand out. First, no engine ignores brand websites: the lowest share is above 27%. Second, position matters: among sources cited in first position, 62% are brand or retailer websites, compared with 56% for sources cited in positions 4 to 10. Claude, Copilot and AI Mode volumes are smaller in our sample, so their figures should be read as trends.

One last point, essential for what follows: only 7.8% of citations to brand websites target a homepage. More than nine times out of ten, the AI cites a deep page: an article, a product page, a service page, a help page. The quality of each of those pages makes the difference, far more than your homepage.

Brand mentions and page citations: two different goals

Being mentioned (your brand name appears in the answer) and being cited (a URL from your site is used as a source) are two distinct outcomes. An AI engine can recommend your brand based on a third-party comparison, or cite your page without recommending your product. Your website plays on both fronts: it shapes how your brand is understood and it provides the pages the AI can link to. We explain this distinction in our article on AI brand mentions vs citations.

Your website as a source of truth: an asset that can backfire

AI engines cross-check several pages to build an answer. If your site shows two different prices for the same offer, an old terms page that is still indexed or an outdated translated version, the AI may pick up the stale information and present it as reliable. The larger your web estate (multiple sites, multiple languages, subsidiaries), the higher the risk. We documented it in our analysis of conflicting brand information.

How AI engines (really) read your website

To be cited, a piece of content has to clear three successive steps: being crawled, being retrieved when the question is asked, then being selected among candidate sources.

1. Crawling: AI bots must be able to reach your pages

Each engine uses its own bots: GPTBot and OAI-SearchBot for OpenAI, ClaudeBot for Anthropic, PerplexityBot for Perplexity, Googlebot for AI Overviews and AI Mode (the Google-Extended token does not apply to Google Search). An overly restrictive robots.txt, a firewall that blocks unknown bots or content loaded only through JavaScript is enough to make a page invisible. Many AI bots do not execute JavaScript: your main content must be in the initial HTML. To check your pages, follow our method to test whether AI search can retrieve your page.

2. Retrieval: AI looks for passages, not pages

When a question comes in, AI engines run several searches in parallel (query fan-out), retrieve dozens of pages and extract the most relevant passages. A section that clearly answers a specific question is therefore more likely to be used than a long text where the answer is buried. We analyzed this mechanism in what ChatGPT actually reads before it cites your page.

3. Selection: authority, consistency and freshness

Among the retrieved passages, the AI favors those coming from sources considered reliable, consistent with the rest of the web and recent. This is where your brand’s authority (mentions, reviews, links), the presence of proof (figures, sources, authors) and the update date come into play.

Optimizing your content to be cited by AI

Google makes the point in its official guide to generative AI features in Search: the strongest lever remains unique, useful content that satisfies visitors. AI engines reward the same qualities, with an extra demand for clarity and extractability.

Answer first, elaborate second

Put a two or three sentence summary answer at the top of every important page and section, then go into detail. This inverted pyramid serves both busy readers and engines that extract passages.

One page per intent, not per keyword

Questions asked to AI engines are long and specific (“which health insurer covers adult orthodontics best?”). Rather than multiplying near-identical pages around a keyword, build pages that fully cover an intent: definition, options, selection criteria, edge cases, frequently asked questions.

Proof that AI engines can reuse

Dated and sourced figures, proprietary studies, concrete examples, an identified author with credentials, visible publication and update dates: these are what turn a claim into citable information. Our own study of the top 100 sources cited by LLMs shows it: original data attracts citations.

Freshness and consistency across your content estate

A corporate website often has hundreds or thousands of pages. Outdated pages, missing metadata and gaps between language versions are silent traps. Schedule regular audits: pages untouched for more than a year, key information (prices, terms, figures) that differs from one page to another, lagging translations.

Hard to cite	Easy to cite
A three paragraph marketing intro before getting to the point	A direct two sentence answer at the top of the page
“Our innovative solutions support your growth”	“The average reimbursement time is 48 hours after we receive the receipt”
Vague headings (“Our strengths”, “Learn more”)	Headings phrased as questions or answers (“How long does a reimbursement take?”)
Figures with no date or source	Dated, sourced and attributed figures
Key information locked in a PDF or an image	Key information in HTML text, possibly repeated in a table
No author, no date	Identified author, visible publication and update dates

The technical foundations of an AI-readable website

Excellent content is useless if it cannot be reached. Technical SEO fundamentals remain the foundation of AI visibility, as we explain in our technical SEO audit for AI readiness.

Indexability, semantic HTML and performance

Your key pages must be indexable (no accidental noindex tag, a clear canonical URL, an up to date sitemap), rendered server side or statically and structured with semantic HTML (hierarchical headings, lists, tables). Performance also matters: healthy Core Web Vitals make crawling easier and improve the experience of people who click through from an AI answer.

Structured data: useful, not magic

Structured data in JSON-LD (Organization, Article, Product, Offer, BreadcrumbList, FAQPage) helps engines identify the entity behind a page, the nature of the content and its key facts. Google states that no special markup is required for its AI features, but structured data remains useful for rich results and for removing ambiguity. The golden rule: markup must always reflect the visible content of the page.

llms.txt and “AI-specific” markup: myths and facts

Myth: you need an llms.txt file to appear in AI answers. Fact: Google explicitly says its search does not use this file. llms.txt can help some other tools, but it never replaces accessible, well-structured pages.

Myth: you should chop your content into small chunks for AI. Fact: Google advises against rewriting or fragmenting content only for AI. A clear structure designed for readers is enough.

Myth: GEO replaces SEO. Fact: AI engines rely heavily on search indexes. Without solid SEO foundations, there is no lasting AI visibility. For the full method, see our complete GEO guide.

Your CMS has become a GEO decision

If your website is the first source AI engines use, the CMS (Content Management System) behind it becomes a strategic visibility tool. It determines whether your teams can structure content, manage metadata at scale, keep languages consistent and publish fast without breaking governance.

What a CMS should enable in 2026

• Structured content: typed content (product, FAQ, article, service page) rather than large blocks of free text, to feed pages and structured data reliably.
• Control over metadata and schema: titles, descriptions, JSON-LD markup and alt text manageable page by page and in bulk.
• Complete HTML rendering: main content available server side, with no JavaScript dependency for the essentials.
• Solid multilingual management: hreflang, translation tracking, detection of gaps between languages.
• Workflow and governance: approvals, permissions, version history, essential as soon as AI generates or edits content.
• APIs and AI connectors: headless APIs, and now MCP (Model Context Protocol) servers that let an AI agent query or act on content.

No architecture is good or bad by principle. It all depends on the quality of the implementation.

Architecture	What it makes easier for GEO	Watch-outs
Traditional CMS (WordPress, Drupal...)	Fast setup, large SEO plugin ecosystem, native HTML rendering	Often loosely structured content, consistency hard to keep at scale, plugin dependency
Headless CMS (Contentful, Strapi, Sanity...)	Highly structured content models, omnichannel delivery, technical freedom	Server-side rendering must be mastered, SEO has to be rebuilt in the front end
Enterprise CMS or hybrid DXP (Jahia...)	Structured content, governance, multisite and multilingual, both traditional and headless delivery	A more structuring project that requires a clear editorial organization

CMS vendors are moving into GEO

The shift is industry-wide: several vendors (Optimizely, HubSpot, Storyblok, Jahia...) are adding features built for AI search, from metadata generation to content audits. Trade publication CMSWire even asks whether CMS modernization has now become an answer engine visibility decision. To show concretely what this changes, we chose to detail the approach of Jahia, a vendor BotRank regularly talks with.

Spotlight: how Jahia brings AI and GEO into its CMS

Founded in 2002 in Geneva, with offices in France, Canada and the United States, Jahia builds an open source CMS and digital experience platform (DXP) written in Java. Its distinctive feature is a hybrid approach that supports both traditional site management and headless content delivery. Jahia is used in financial services, telecommunications, higher education and the public sector, industries where content governance is critical.

An “Agentic CMS” under human control

Jahia positions its offer as an agentic CMS: AI handles repetitive tasks (content audits, metadata, translations, page drafts, bulk edits) but every action goes through the approval workflow. By default, AI cannot publish without human validation and every operation is logged (author, timestamp, action type). For GEO this is key: producing more content is pointless if quality and consistency do not follow.

Features directly tied to SEO and GEO

According to the vendor, Jahia can notably:

• generate SEO tags, meta descriptions, alt text and Schema.org structured data (Article, FAQPage, BreadcrumbList) at scale;
• run quality checks before publishing: WCAG accessibility, legal compliance, brand consistency;
• audit the content library to spot outdated pages, missing metadata and inconsistencies between language versions;
• apply bulk updates across several sites, always through the approval flow;
• pick the right AI model for each task (ChatGPT, Claude, Gemini, Mistral, Azure OpenAI, OVH AI or a self-hosted model), with on-premise deployment available for organizations that do not want their content to leave their environment.

Jahia also claims content operations up to 10 times faster with its AI agents: a figure provided by the vendor, which every organization should validate on its own use cases.

An MCP server to manage content in natural language

Jahia offers a native MCP server. In practice, an assistant such as ChatGPT, Claude or Gemini can connect to the CMS to list pages, read their metadata, audit content or trigger workflows, under the same governance rules as a human user.

We tested it on a bilingual demo site (a fictional health insurer), in read-only mode. In five requests, an AI agent inventoried the site’s 24 pages, checked that each one had a meta description in French and English, verified the publication status of the whole tree and found that the FAQ page stored its 15 questions and answers as typed content rather than as a single text block. That is exactly the kind of structure that makes FAQPage structured data and passage extraction by AI engines easier.

This capability echoes an idea defended by Julian Maurel, CEO of Jahia, in a column published in Journal du Net (in French): moving from one-off audits to continuous “observability” of the web estate, able to spot outdated or contradictory information before AI engines pick it up.

What this changes for a GEO strategy

A CMS like Jahia works upstream: it helps produce structured, consistent and properly marked-up content. A GEO tool like BotRank works downstream: it measures whether that content is actually used, cited and recommended by AI engines, prompt by prompt and engine by engine. The two complement each other: one makes the site citable, the other checks that it is cited.

Checklist: make your website visible in AI in 10 steps

1. Make sure robots.txt and your firewall do not stop AI bots (OAI-SearchBot, GPTBot, ClaudeBot, PerplexityBot, Googlebot) from reaching your key pages.
2. Ensure your main content is in the initial HTML, without relying on JavaScript.
3. Put a direct two or three sentence answer at the top of every important page.
4. Build one complete page per search intent rather than near-identical pages.
5. Add proof: dated and sourced figures, studies, examples, an identified author.
6. Implement JSON-LD structured data consistent with visible content (Organization, Article, Product, BreadcrumbList).
7. Show publication and update dates, and refresh strategic pages regularly.
8. Audit the consistency of key information across pages, sites and languages.
9. Choose or configure a CMS that handles structured content, metadata at scale and governance.
10. Continuously measure your presence and citations in AI engine answers.

Measuring impact: from crawl to citation

Optimization only has value if you measure it. Three levels of tracking work together:

• Crawling: server logs show which AI bots visit your pages and how often.
• Performance in Google: Search Console now offers a report dedicated to generative AI features.
• Presence in answers: only prompt tracking tells you whether your brand is mentioned, which pages are cited and against which competitors, on each engine.

That is exactly what BotRank does: the Source Analysis feature identifies the pages and domains AI engines cite on your topics, while technical audits check that your pages are accessible and readable by AI bots.

FAQ

What is SEO for AI?

SEO for AI, also called GEO (Generative Engine Optimization), covers the practices that make a website understandable, credible and citable by generative engines such as ChatGPT, Gemini, Perplexity and Google AI Overviews. It builds on SEO fundamentals (accessibility, quality, authority) with an added demand for clarity and proof.

Is traditional SEO enough to appear in ChatGPT?

It is necessary but not sufficient. ChatGPT relies on search indexes to find pages, then selects the clearest and most reliable passages. A well-ranked site with vague or contradictory content will rarely be cited.

Do I need an llms.txt file?

It is not a priority. Google says its search does not use this file. It can help with some tools, but what matters most is having accessible, well-structured and up to date pages.

Which CMS should I choose to be visible in AI engines?

No CMS guarantees a citation. The right choice is one that produces complete HTML, structures content, manages metadata and structured data at scale, handles multiple languages and keeps strong governance. For organizations with many or complex websites, a hybrid enterprise CMS such as Jahia meets those requirements.

Is a headless CMS better for GEO?

Not automatically. A headless CMS offers highly structured content models, but server-side rendering and SEO elements must be carefully rebuilt in the front end. Poorly implemented, it can even hide content from bots that do not execute JavaScript.

How do I know whether my website is cited by AI?

You need to track prompts that represent your customers on each engine and analyze the sources cited in the answers. A GEO tool like BotRank automates this tracking and shows which pages of your site, or your competitors’ sites, AI engines rely on.

Methodology

The data comes from the BotRank database. It covers 445,808 answers generated by the AI engines tracked by BotRank (ChatGPT, Google AI Overviews, Gemini, Perplexity, Claude, Mistral, Copilot and Google AI Mode) between September 20 and October 20, 2026, representing more than 4.2 million source citations. The four-month trend covers July 1 to October 20, 2026 (more than 1.4 million answers and more than 10 million citations). Each cited domain is classified by site type (brand, e-commerce, institutional, media, education, social network, blog, community, etc.). The prompts analyzed relate to brands from a range of industries, with a strong presence of retail and consumer goods, which may overweight retailer websites. Homepages are identified from URL structure.

Sources

• Google Search Central, guide to generative AI features in Google Search
• France Num, generative engine optimization (in French)
• Jahia, AI-powered enterprise CMS
• Wikipedia, Jahia
• Journal du Net, SEO and GEO: AI observability, the next project for large websites (in French)
• CMSWire, Is CMS Modernization Now an Answer Engine Visibility Decision?
• Schema.org

Go further: Jahia and BotRank are hosting a webinar (in French), “Your next visitor is an AI, and it already has an opinion about you”, on Thursday, November 5, 2026 at 1:30 pm Paris time (45 minutes), with Julian Maurel, CEO of Jahia, and David Banget, CEO and co-founder of BotRank. Register for the webinar.