Make the page eligible for search
Keep it indexable, avoid accidental noindex or nosnippet settings, and make sure robots.txt, CDN rules, and WAF rules allow the bots you want.
Generative Engine Optimization
Generative Engine Optimization, or GEO, is the practice of making a page easy for AI search systems and assistants to discover, understand, quote, and cite accurately. In practice, that means strong SEO fundamentals, clearer entities, answer-first content, and an explicit policy for AI crawlers.
Author: Piotr ZientaraLast updated: Canonical: https://www.alfaumi.pro/generative-engine-optimization
What changed
GEO is less about inventing new markup and more about serving modern retrieval systems: AI Overviews, AI Mode, ChatGPT Search, Claude Search, and Perplexity. Since August 31, 2026 every site can see its AI Overviews and AI Mode impressions in Search Console and opt out of those features without leaving Google Search.
What did not change
High-quality technical SEO still matters. Google says optimizing for generative AI search is still SEO, with no special files, markup, or schema required.
What to do now
Build pages around real intents, keep important content in HTML, decide the search and training questions separately, and check that your CDN is not blocking the bots you want.
Current checklist
Answer the query in the first 150 words
Lead with a direct, definition-style answer in the opening lines. AI retrieval pipelines weight the first ~150 tokens heavily during summarization, so a buried answer is easy to miss.
Back claims with citations and concrete data
Include named sources, statistics, dates, and first-party numbers inline at the claim. Pages with verifiable, cross-checkable facts are cited noticeably more often than vague brand copy.
Keep time-sensitive pages fresh
Citation priority decays fast: most pages cited in AI answers are under 30 days old. Revisit comparisons, pricing, and how-to content on a regular cycle and show a real update date.
Expose facts in crawlable text
Products, policies, pricing, comparisons, and explanations should be visible in the rendered HTML, not only inside client-side widgets or PDFs.
Use schema that reflects the page
Add Organization, SoftwareApplication, Article, Product, BreadcrumbList, or other relevant schema only when readers can see matching content. Google says structured data is not required for its AI features but keeps pages eligible for rich results, and FAQ rich results stopped appearing on May 7, 2026.
Show who is responsible
Include authors, organization details, and dates so the page has stronger trust and provenance signals.
Create pages for real intents
Exact-match pages such as product comparisons, definition pages, pricing pages, and policy pages are easier for search systems and answer engines to retrieve and cite.
Separate discovery from training
Search bots and user fetchers put you inside an answer with a link back; training crawlers decide whether the model recalls you at all. They are separate controls, so decide each one on purpose. Do not count on robots.txt for user-triggered fetchers: Perplexity says Perplexity-User generally ignores it.
Use Google's own switch for AI features
Google-Extended covers Gemini training and grounding in Gemini Apps and Vertex AI, not Google Search. To leave AI Overviews, AI Mode, and AI features in Discover while staying in regular results, use the Search generative AI control, which reached every site in Search Console on August 31, 2026.
Check the edge, not only robots.txt
CDN and WAF rules can refuse AI bots before robots.txt is ever read. Cloudflare now sorts AI traffic into Search, Agent, and Training on every plan, and from September 15, 2026 new domains block Agent and Training traffic by default on pages that show ads. The same day it added a Disallow AI Training setting, so a site can refuse training while staying indexed for search. Confirm the bots you want can actually fetch your pages.
Measure AI visibility where platforms report it
Search Console now reports impressions, pages, and countries for AI Overviews and AI Mode, and Bing Webmaster Tools previews Citation Share for grounding queries. Check those reports after a change instead of guessing.
Build visibility across multiple AI engines
ChatGPT Search, Google AI Mode, Perplexity, and Copilot each retrieve differently. Brands cited across several engines see far higher overall citation rates than single-engine presence.
Treat llms.txt as optional, not foundational
llms.txt is cheap to publish and some IDE agents and MCP tools read it, but Google says Google Search ignores it, so it will neither harm nor help your visibility there. Do not let it displace good HTML, internal links, or schema.
A practical robots.txt stance
Search bots and training bots are separate controls, and they cost you different things. Search bots and user-directed fetchers are what put you inside an answer, with a link back — block those and you disappear from AI search. Training bots decide whether the model recalls you at all when nobody searches; block those and you keep your content out of the corpus, but you also give up that recall.
Which trade you want depends on what your pages are. A publisher protecting original reporting has a real reason to refuse training access. A product site whose pages exist to be found usually does not — which is why this site allows both.
User-agent: *
Allow: /
Disallow: /api/
User-agent: OAI-SearchBot
Allow: /
Disallow: /api/
User-agent: ChatGPT-User
Allow: /
Disallow: /api/
User-agent: Claude-SearchBot
Allow: /
Disallow: /api/
User-agent: Claude-User
Allow: /
Disallow: /api/
User-agent: PerplexityBot
Allow: /
Disallow: /api/
# Training crawlers are a separate control. Allow them if you want the model
# to recall you without searching; swap Allow for Disallow if you would rather
# your content stayed out of training corpora. Either way, the block above
# keeps you in the answers themselves.
User-agent: GPTBot
Allow: /
Disallow: /api/
User-agent: ClaudeBot
Allow: /
Disallow: /api/
User-agent: Google-Extended
Allow: /
Disallow: /api/Google blurs the line: Google-Extended covers Gemini grounding as well as training, and it is separate from Google Search. Refusing it therefore costs more than refusing a pure training crawler, which is why it deserves a deliberate choice rather than a copied default. It is also not the way out of AI Overviews: since August 31, 2026 that is the Search generative AI control in Search Console, which Google says does not act as a ranking or inclusion signal for the rest of Search. And robots.txt is only half the policy, because a CDN or WAF can refuse these bots before they ever read it.
Where llms.txt fits
llms.txt is a helper file for inference-time agents and developer tooling. Google now states that Google Search ignores it, so publishing one will neither harm nor help your visibility there, and 2026 adoption studies show other major AI search crawlers rarely fetch it either. It should not replace good HTML pages, internal linking, or structured data. Its real value today is for IDE agents and MCP tooling that read a curated shortlist of your most important pages.
AlfaUMi now publishes both /llms.txt and /llms-full.txt so agents can get either a short or expanded view of the product.
References
- Google AI Features and Your Website
Google says there are no extra AI-only requirements for AI Overviews or AI Mode.
- Google AI optimization guide
Optimizing for generative AI search is still SEO: no special schema is required, and Google Search ignores llms.txt.
- Search generative AI control
Leave AI Overviews, AI Mode, and AI features in Discover without affecting the rest of Search. Live for every site since August 31, 2026.
- Google crawler reference
Google-Extended controls Gemini training and grounding and does not affect inclusion in Google Search.
- OpenAI crawler controls
OpenAI separates OAI-SearchBot, ChatGPT-User, and GPTBot.
- Anthropic crawler controls
Anthropic separates Claude-SearchBot, Claude-User, and ClaudeBot.
- Perplexity crawler docs
Perplexity recommends allowing PerplexityBot and, where relevant, its published IPs.
- llms.txt proposal
A proposal for an AI-readable site overview. Useful for IDE agents, but Google Search ignores it and other AI search crawlers rarely fetch it.
- Cloudflare AI traffic controls
Search, Agent, and Training controls on every plan. From September 15, 2026 new domains block Agent and Training on ad pages by default, and a Disallow AI Training setting refuses training without costing search indexing.
- Bing AI Performance insights
June 2026: Citation Share, Intents, Topics, and period comparison for grounding queries, in preview.
- Google AI Overviews & AI Mode links update
May 2026: Google expanded inline links, hover previews, creator attribution, and article suggestions inside AI answers.
FAQ
Is GEO different from SEO?
It builds on SEO rather than replacing it. Google's AI optimization guide puts it plainly: optimizing for generative AI search is still SEO. SEO gets pages discovered and indexed; GEO makes the same pages easier for AI systems to extract, summarize, and cite correctly.
Does schema guarantee AI citations?
No. Google says structured data is not required for its generative AI features. Schema helps machines understand page structure and entities, but citations still depend on the relevance, clarity, originality, and usefulness of the visible content.
Does FAQPage schema still help?
Not in Google results: FAQ rich results stopped appearing in Google Search on May 7, 2026. FAQPage is still valid schema.org markup, so keep it where the page really has questions and answers, but put the answers in visible text, because that is what AI systems read and cite.
Should I publish llms.txt?
It is low-effort and some IDE agents and MCP tools use it, so it is fine to publish. Google says Google Search ignores it, and 2026 adoption studies show other major AI search crawlers rarely fetch it, so it should never replace strong HTML pages, internal linking, or structured data.
Can I block training bots and still show up in AI search?
Often yes. OpenAI, Anthropic, and Perplexity document separate controls for search or user-requested fetches versus model-training crawlers. At Google, Google-Extended governs Gemini training and grounding, while AI Overviews and AI Mode have their own opt-out in Search Console.
Can I opt out of AI Overviews without leaving Google Search?
Yes. The Search generative AI control in Search Console, rolled out to all sites by August 31, 2026, removes a site from AI Overviews, AI Mode, and generative AI features in Discover. Google says it is not used as a ranking or inclusion signal for the rest of Search, and exclusion usually takes a few days. Blocking Google-Extended does not do this.
How fresh does my content need to be for AI search?
Fresher than most teams expect. For time-sensitive topics, citations decay within weeks and the majority of pages cited in AI answers are under a month old. Keep comparisons, pricing, and how-to pages on a regular update cycle and surface a genuine update date.
What content gets cited most often?
Pages that answer a narrow question in the first 150 words, include concrete data and named citations, use consistent entity names, stay recently updated, and make supporting details easy to verify are cited most often.
Can a GEO tool guarantee AI citations?
No, and be wary of any that promise it. Google's June 2026 guidance on third-party SEO tools notes that they do not have access to its internal ranking data, and flags vendors promising gains in AI experiences, including 'AEO' and 'GEO' tools. AlfaUMi scores are estimates of readiness against public guidance, not a view into any engine's ranking.
Did Google change anything for AI search in 2026?
Yes. In May 2026 Google stopped showing FAQ rich results, published an AI optimization guide, and expanded links inside AI Overviews and AI Mode. In June it clarified that Google Search ignores llms.txt. By August 31 every site could see generative AI impressions in Search Console and opt out of AI features without leaving Search. The core position is unchanged: optimizing for generative AI search is still SEO.
Audit your site against the checklist
AlfaUMi checks your site against these requirements and scores structured data, crawlability, AI bot policy and content clarity.