Guide · SEO & GEO
SEO & GEO: What Makes You Visible in AI Search Interfaces
Google simply calls GEO SEO – with two technical prerequisites
GEO (generative engine optimisation) means optimising for AI search interfaces like AI Overviews, AI Mode and ChatGPT. At page level, Google names two technical prerequisites: the page must be indexed and allowed to serve a snippet. Everything else is solid SEO plus consistent facts across all sources – you don’t need a new discipline for it.
What GEO is – and what Google itself says about it
GEO stands for generative engine optimisation: the work of getting a website to appear in AI-generated answers – in Google’s AI Overviews and AI Mode, in ChatGPT, Claude or Perplexity. The term sounds like a new discipline with its own toolbox. Google disagrees. The official guide to optimising for generative search features states: from the perspective of Google Search, optimising for generative AI features is simply SEO.
That’s not appeasement, that’s architecture. AI Overviews and AI Mode run on the same index and the same ranking systems as classic search. Their prevalence fluctuates strongly, though: the share of US search queries with an AI Overview rose from 6.49 percent in January 2025 to 24.61 percent in July and fell back to 15.69 percent by November 2025 (Semrush).
| Month 2025 | Share of US search queries with an AI Overview |
|---|---|
| January | 6.49 % |
| July | 24.61 % |
| November | 15.69 % |
Source: Semrush, as of 2025.
Two techniques decide which pages end up in an answer: grounding (RAG) backs the generated answer with pages from the search index. Query fan-out generates additional related search queries and fetches further results for them. Whoever wants to be cited in AI answers must first be in the index. There is no side entrance.
Proven, plausible, speculation: the distinction that counts
GEO is a field full of claims. Before you invest money or time, sort every measure into one of three tiers. This sorting is the actual value of this article.
Proven – stated in Google’s documentation:
- At page level, two prerequisites: indexed and snippet permission. Plus the inclusion of the website in the Search Console.
- Unique, experience-based content is the strongest lever. Summaries of what’s already been said are not.
- Structured data is not required for generative search features. A special schema does not exist.
- Google Search does not evaluate llms.txt.
Plausible – a justifiable derivation, without proof of effect:
- Entity consistency: identical company facts across the website, schema, legal notice and business profiles. Grounding systems compare sources; contradictions weaken the assignment.
- A grounding page that bundles your company’s citable core facts. Useful for humans, possibly useful for machines.
- Allow OAI-SearchBot: according to OpenAI, websites that lock out the crawler do not appear in ChatGPT’s search answers. Allowing it is therefore a prerequisite – a citation probability does not follow from it.
Speculation – circulating, but not verifiable:
- llms.txt as a visibility lever.
- “Chunking” content, rewriting “for the AI”, purchased mentions. Google explicitly names all three as unnecessary.
- “AI ranking scores” from third-party tools. Google makes it clear: no external tool has access to internal ranking or AI systems.
A guide that mixes these three levels sells you speculation at the price of facts.
The technical prerequisites in detail
The checklist for Google’s generative search features is short:
- The page is indexed.
- The page is allowed to serve a snippet – no nosnippet, no max-snippet:0, no data-nosnippet over the main content.
- The website is included for generative search features in the Search Console.
- The content is crawlable and available as text, not only in images or behind JavaScript.
- The page experience is right, the main content is clearly distinguishable from the trimmings.
If any of these points fails, the cause lies in technical SEO – and that’s where it gets fixed, not in a GEO tool.
Controlling AI crawlers: GPTBot, ClaudeBot, PerplexityBot, Google-Extended
Every major provider now separates training, search and user requests. The overview:
| Provider | Crawler | Purpose |
|---|---|---|
| OpenAI | GPTBot | Training the models |
| OpenAI | OAI-SearchBot | Search results in ChatGPT |
| OpenAI | ChatGPT-User | Page retrieval on user request |
| Anthropic | ClaudeBot | Training the models |
| Anthropic | Claude-SearchBot | Search quality |
| Anthropic | Claude-User | Page retrieval on user request |
| Perplexity | PerplexityBot | Search results in Perplexity |
| Perplexity | Perplexity-User | Page retrieval on user request |
| Google-Extended | Training and grounding for Gemini and Vertex AI |
The most common misconception first: Google-Extended does not control AI Overviews. The token concerns the training of future Gemini models and grounding in Gemini apps and Vertex AI. Google explicitly writes: the token influences neither inclusion in search nor ranking. AI Overviews run via the normal Googlebot. Whoever wants out of there would have to suppress snippets – and thereby also lose classic search.
Whether you allow the other providers’ crawlers is a business decision, not a technical one. Allowing means: a chance of being named in their answers, often without a click. Blocking means: your content isn’t used there, but you’re invisible. By the way, not every retrieval respects robots.txt. Perplexity itself documents that Perplexity-User generally ignores robots.txt rules for user-initiated retrievals.
llms.txt: the honest state of affairs
llms.txt is a proposal from the developer community: a Markdown file in the domain root that offers AI systems the most important pages and facts in curated form. The state of affairs in one sentence: Google Search does not evaluate the file – it says so in Google’s own guide – and no other major provider documents using it for search answers.
“I manage around 15 client projects in parallel, and in none of them have I seen an effect I could cleanly attribute to llms.txt. I still create it: it costs me half an hour, does no harm, and if a system ever evaluates it, it’s ready.”
— Viktor Pásztor, SEO freelancer
That’s exactly how you should classify the file: a deliberate measure with an open outcome. Whoever sells it to you as a breakthrough hasn’t read the evidence – or is ignoring it.
Entity consistency: the underestimated part
Grounding systems compare statements across multiple sources. If your website, legal notice, Organization schema and Google Business Profile show different company names, addresses or self-descriptions, every contradiction weakens the assignment. The measure is unspectacular: the same facts everywhere. A uniform company name including legal form, identical contact details, the same key figures.
This is no new AI craft, but well-maintained on-page work with clean schema – except that sloppiness here could become more expensive than before. The effect is not proven. But it’s the most plausible unproven measure in the whole field, and it costs almost nothing.
Measurement: one solid report, a lot of noise
For Google there is exactly one solid source: the Search Console report on generative search features. It shows how your content is found via AI features.
For ChatGPT, Claude and Perplexity, no official report exists. What remains: referral traffic in web analytics and manual spot checks. Spot checks fluctuate strongly and are personalised – they serve as an indication, not a metric. Whoever sells you an exact measurement from that is measuring something – just not what they claim.
Case study from a B2B project
A B2B client wanted to know why ChatGPT names the competition for industry questions, but not him. The check revealed: robots.txt blanket-blocked all non-Google crawlers – a leftover from an old security configuration. Plus three different company names in circulation: one in the legal notice, one in the schema, a third in industry directories.
The measures: OAI-SearchBot and PerplexityBot released per the client’s decision, GPTBot blocked at the client’s request, the company name unified everywhere, core facts bundled on a visible, linked page. There was no citation promise – no provider guarantees visibility. But the documented prerequisites are now met, and the decision on every crawler lies with the client in writing.
Common mistakes
Blocking Google-Extended
Whoever blocks Google-Extended to disappear from AI Overviews achieves nothing. The Overviews run via the normal Googlebot.
One page per fan-out question
Creating a separate page for every conceivable follow-up question falls under Google’s spam policy on mass-produced content.
Structured data as an AI lever
Google explicitly says structured data is not required for generative features.
Facts page only for bots
Building a facts page for crawlers and hiding it from users is cloaking and endangers the whole domain.
Numbers without a source
Grounding prefers the verifiable. An unsubstantiated claim is worthless for it.
Believing “internal AI metrics”
Some GEO tools promise internal AI metrics. Such metrics don’t exist for outsiders.
FAQ
Is GEO its own discipline requiring a specialist agency?
No. Google explicitly classifies optimisation for generative search features as SEO. Anything beyond that, any solid SEO provider handles too. You don’t have to pay extra for a separate vocabulary.
How do I get into ChatGPT’s search answers?
Prerequisite according to OpenAI: OAI-SearchBot is allowed to crawl your website. After that, the same applies as everywhere: content that answers a question better than the rest. There is no guarantee.
Does blocking GPTBot hurt my Google ranking?
No. GPTBot is an OpenAI crawler and has nothing to do with Google’s ranking. Blocking Google-Extended also, according to Google, influences neither inclusion in search nor ranking.
Can anyone guarantee me citations in AI answers?
No. Neither Google nor OpenAI nor Anthropic or Perplexity guarantee inclusion or mention, and no third party has access to their internal systems. Whoever sells a citation guarantee is selling something they don’t control.
Viktor Pásztor
Viktor Pásztor is an SEO freelancer in Berlin, has been in digital marketing for over 15 years and manages around 15 client projects in parallel, mostly e-commerce and B2B. He has been working 100 % remotely for more than five years – with WordPress, Shopify, Shopware and TYPO3. More about Viktor Pásztor
Keep reading
Technical SEO
Crawling, indexing and page experience – the basis AI search features also build on.
SEO pricing
What working with me costs.
SEO tools
Which tools I use daily – and what for.
The next step: have your visibility checked
I’ll check where your website stands on the technical prerequisites, crawler control and entity consistency. Reply within 24 hours on business days – an honest assessment instead of a citation guarantee, because nobody can give you that.
→ Request an SEO audit now