Generative engine optimization (GEO): a starter guide for marketers

Generative engine optimization (GEO) is the work of making your content more likely to be retrieved, cited and linked when an AI system writes an answer. That covers chat assistants with web search, such as ChatGPT, Claude, Perplexity and Microsoft Copilot, and AI features inside search results, such as Google’s AI Overviews and AI Mode.

Start with what is actually known. As of November 2026, the platforms document only a few things: which crawlers to allow, the technical conditions for eligibility, and how to measure some results. Most other GEO advice is inference from tests and observation. Nothing in this guide guarantees a citation. The seven steps below improve your eligibility and your odds.

What GEO means and how AI answers get their sources

The term comes from a 2023 research paper on generative engine optimization (opens in a new tab), later accepted at the KDD 2024 conference, which tested ways to raise a source’s visibility in AI-generated answers. Google frames it differently. Its guide to optimizing for generative AI features (opens in a new tab), last updated in July 2026, says that from Google Search’s point of view this work is still SEO (search engine optimization). The reason it gives is that AI Overviews and AI Mode are built on its core ranking and quality systems.

AI systems use web content in two ways:

• Training. Models learn from crawled pages. This does not produce a citation, and you cannot measure it directly.
• Retrieval. At answer time the system runs searches, reads the pages it finds and writes a response with links. Google calls this retrieval-augmented generation (RAG), or grounding. It also describes query fan-out, where one question triggers several related searches at once. OpenAI’s help page on ChatGPT search (opens in a new tab) describes a similar process, in which a prompt is rewritten into one or more targeted queries that may go to partner search providers.

Citations come from retrieval, so that is where GEO effort goes. Each platform splits these jobs across different bots.

Platform	Search bot	Training control	User-triggered fetcher
OpenAI (ChatGPT)	OAI-SearchBot	GPTBot	ChatGPT-User
Anthropic (Claude)	Claude-SearchBot	ClaudeBot	Claude-User
Perplexity	PerplexityBot	None listed	Perplexity-User
Google Search AI features	Googlebot	Google-Extended	Not covered here

The names come from OpenAI’s crawler overview (opens in a new tab), Anthropic’s crawler help page (opens in a new tab), Perplexity’s crawler documentation (opens in a new tab) and Google’s list of common crawlers (opens in a new tab).

Documented facts versus industry inference

Keep this table in mind when you read GEO advice, including the advice below.

Claim	Status as of November 2026
Blocking OAI-SearchBot keeps a site out of ChatGPT search answers	Documented by OpenAI, which adds that the site can still appear as a navigational link
A page must be indexed and eligible for a snippet to appear in AI Overviews or AI Mode	Documented by Google
Clear headings, tables and FAQ sections make content easier to reference	Suggested by Microsoft; Google says you need not write or chunk content specially for AI
Adding cited sources and statistics raises visibility	Found in one academic benchmark; Microsoft advises backing claims with evidence but gives no measured effect
Brand mentions across the web shape AI recommendations	Industry inference; Google says chasing inauthentic mentions is less helpful than it seems
An llms.txt file improves citations	Not confirmed; Google says Search ignores the file

Step 1: Let the right bots in

Open your robots.txt file and look for rules that block the search bots in the table above. OpenAI says blocking OAI-SearchBot keeps a site out of the answers ChatGPT search gives. Anthropic says disabling Claude-SearchBot or Claude-User may reduce your visibility in Claude’s search results. Each bot is controlled separately, so you can allow search bots and still opt out of training. Crawlers are allowed by default, so the Allow groups below only record your intent.

Be careful when you add a named group. As Google’s robots.txt guide (opens in a new tab) explains for its own crawlers, a crawler follows only the most specific group that names it. Assume other bots behave the same way, and repeat inside each named group any general Disallow rules you still want.

# AI search crawlers: allowed
User-agent: OAI-SearchBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

# Optional: opt out of AI model training
User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: Google-Extended
Disallow: /

Then check the details that robots.txt does not cover:

• Firewalls and content delivery networks (CDNs). Google’s AI features documentation (opens in a new tab) lists crawl access through robots.txt and through any CDN or hosting layer among the basics. OpenAI and Perplexity publish IP ranges you can allow.
• Timing. OpenAI and Perplexity both say a robots.txt change can take about 24 hours to take effect.
• User-triggered fetchers. OpenAI says robots.txt rules may not apply to ChatGPT-User, and Perplexity says Perplexity-User generally ignores them, because a person asked for the page.
• Google. Google’s AI features use the regular Search index, which Googlebot crawls. A page must be indexed and eligible to show with a snippet. Search Console also has a Search generative AI control (opens in a new tab), available to all sites as of August 31, 2026. Inclusion is the default, so confirm nobody has switched your property to exclude.
• Microsoft. Microsoft reports Copilot and Bing AI citations in Bing Webmaster Tools and says Bing respects the preferences you set in robots.txt, so verify your site there too.

Google-Extended is different. Google says it governs model training and grounding for the Gemini apps and Vertex AI, and has no effect on Search inclusion or ranking.

Step 2: Answer the question directly

Open each page, and each major section, with a one- or two-sentence answer in plain words. Then add the detail. Name the subject in the sentence instead of using “it” or “this”, so the sentence still makes sense when it is lifted out of the page.

Our read: retrieval systems work with passages, and a self-contained answer is easier to use and to attribute. This is inference, not a documented rule. Google’s guide says you do not need to rewrite content for AI systems or split it into small chunks, and warns that mass-producing pages for every query variation can break its spam policies. Write this way because readers benefit.

Step 3: Structure pages clearly

• Give each page one topic, with headings that describe each section.
• Use tables for comparisons and numbered lists for steps.
• Put key facts in text, not only in images or video. Google lists text availability among its basics.
• Keep structured data consistent with the visible page. Google says it is not required for AI features and that no special schema exists for them, but it still supports rich results.

Microsoft’s AI Performance announcement for Bing Webmaster Tools (opens in a new tab) suggests clear headings, tables and FAQ sections to help AI systems reference content accurately.

Step 4: Show sourcing and expertise

Link factual claims to primary sources. Name the author and reviewer, and link to pages that explain who they are. Describe how you tested or measured anything you report. Google’s guidance on helpful, reliable content (opens in a new tab) asks whether a page shows clear sourcing, evidence of expertise and background on the author or site. Its AI guide adds that first-hand, non-commodity content stands out against summaries of what others have said.

Research points the same way, with limits. The GEO paper found that adding citations, quotations from relevant sources and statistics significantly improved visibility in its benchmark, and that the effect varied by subject area. That was a controlled academic test, not a platform commitment. Treat it as a hypothesis, and never invent a statistic to fill a gap.

Step 5: Keep facts fresh

List the pages that carry prices, dates, specifications or policy details, and put them on a review schedule. Update the facts first, then the visible date. Google’s content guidance warns against changing dates when nothing substantial has changed.

Microsoft says regular updates help AI systems reference the current version of a page. It points to IndexNow, a protocol that notifies participating search engines when pages are added, updated or removed. If you run a local or ecommerce business, keep your Google Business Profile, Merchant Center and Bing Places details current too.

Step 6: Build genuine brand mentions

This step is mostly inference. When someone asks an assistant for the best option in a category, the answer appears to draw on reviews, comparisons, forums and news, not only on your own site. Google confirms part of this: its guide says AI features can reflect what is said about products and services across the web. The same guide says chasing inauthentic mentions is less helpful than it might seem. Google’s documentation changelog (opens in a new tab) also records a May 2026 clarification that its spam policies apply to generative AI responses.

So earn mentions the slow way:

• Publish original data, tools or research that others have a reason to cite.
• Pitch real stories to publications your customers read.
• Ask customers for honest reviews on the platforms that matter in your category.
• Describe your product the same way everywhere, so the facts agree.

Do not pay for undisclosed mentions, post fake reviews or seed forums under false identities.

Step 7: Measure AI visibility and referral traffic

Tool	What it reports	Limits
Search Console Generative AI performance report	Impressions of your links in AI Overviews and AI Mode, by page, country, date and device	Google Search only
Bing Webmaster Tools AI Performance	Total citations, cited pages and sample grounding queries for Copilot and Bing AI summaries	Launched as a public preview in February 2026; no ranking or placement data
Google Analytics AI Assistant channel	Visits from assistants such as ChatGPT, Gemini and Copilot	Excludes AI Overviews and AI Mode, which count as Organic Search
Manual prompt checks	Whether assistants cite you for a fixed list of questions	Answers vary from run to run

Google’s Generative AI performance report documentation (opens in a new tab) says the report reached all sites as of August 31, 2026. The channel definitions come from Google Analytics’ default channel group documentation (opens in a new tab).

Record a baseline now. Each month, compare impressions and citations with referral visits and conversions, and rerun the same prompts. Google’s guide cautions that no third-party tool has access to its internal ranking or AI systems, so read vendor visibility scores as estimates.

Frequently asked questions

Is GEO different from SEO?

Mostly no. Google says optimizing for its AI features is still SEO, and the access and indexing basics are the same. Our read is that GEO adds three things: access for AI search crawlers, content that is easy to cite, and AI-specific measurement.

Do you need an llms.txt file?

No. The llms.txt proposal (opens in a new tab), first published in 2024 and revised in August 2026, is a community proposal for a Markdown file that helps AI agents use a site. Google says Search ignores such files, so one will neither help nor harm your visibility there. Chrome’s Lighthouse llms.txt audit (opens in a new tab) treats the file as optional, and none of the crawler documentation cited in this guide names it as a search or citation factor.

Should you block AI training crawlers?

That is a business decision, not a visibility tactic. OpenAI and Anthropic document separate bots for training and for search, so blocking GPTBot or ClaudeBot does not block their search bots. Controls are still changing: Anthropic says that as of October 30, 2026 it has begun recognizing a machine-readable signal for opting out of AI training, and Google says it intends to support a similar signal in early 2027.

Can anyone guarantee citations in AI answers?

No. Answers vary with the wording of the prompt, the user’s context and the sources retrieved at that moment, and none of the platform documentation cited here promises a citation to any site. Be skeptical of any vendor that does. The work in this guide makes your pages eligible and easier to cite, which is the part you control.