SEO, GEO, AI Visibility: three terms, three questions

• SEO

  Does a search engine understand the content, and do people find the page through it?

  In its own introduction Google describes SEO as the work of helping search engines understand your content and helping people find the site and decide whether to visit it. The same page states explicitly that there are no secrets here that will automatically rank a website first in Google.

• GEO

  What is meant when people talk about optimising for generative answers?

  The term is known to Google. Google's guide to generative AI features spells out the abbreviations itself — AEO as “answer engine optimization”, GEO as “generative engine optimization” — and frames both as terms you may come across online. Google does not adopt them as a category of its own: from Google Search's perspective, the guide writes, optimising for generative AI search is optimising for the search experience and therefore still SEO.

• AI Visibility

  Does the website actually appear in the answer of a named system?

  This is the only one of the three questions that can be answered by observation alone. This article sets the expression itself: on Google's pages named below it does not appear as an established term — there the wording is about improving visibility in AI search experiences. Microsoft uses it as a section heading in the Bing Webmaster Tools help, without defining it there.

That these questions stay separate can be read off OpenAI and Anthropic themselves: both separate their own requests. OpenAI describes OAI-SearchBot as the access that concerns appearing in ChatGPT search results, and GPTBot as the one through which content may feed into training; the settings for these are, by OpenAI's own account, independent of one another. Anthropic speaks of a variety of robots and lists three of them individually — one for model development, one for searching the web, and one for retrieving web content at users' direction.

Classic search is not a single operation either. Google describes it as three stages — crawling, indexing and serving search results — and notes in the same sentence that not every page makes it through each stage.

No sequence is implied by any of this. That a system is allowed to fetch a page does not prove that it appears in an answer; and being mentioned in an answer does not make classic discoverability superfluous. The three questions are connected, but they do not redeem one another — and that is how the diagram is built: three equally weighted lines of sight onto the same website, with no chain of arrows between them.

For its own search, Google names technical minimum requirements for a page to be eligible for indexing at all. There are three: Googlebot is not blocked, the page responds with the status code HTTP 200, and the page has indexable content. The caveat follows immediately — that a page meets these requirements does not mean it will be indexed; indexing is not guaranteed.

For the AI features of its own search, Google describes the same foundation. For a page to be able to appear as a supporting link in AI Overviews or AI Mode, it must be indexed, eligible to be shown with a snippet, and fulfil the technical requirements of Search. There are no additional technical requirements. Here too the source places the caveat right behind it: indexing and serving are not guaranteed.

Nor does Google require any special markup for these features. By its own account, no new machine readable files, no AI text files and no special schema.org markup are needed to appear in them. Files such as llms.txt or Markdown versions of a page are not used by Google Search itself — not for its generative features either. That is a statement about Google Search and about nothing else.

It is easy to turn this into a shared provider line that does not exist in the documents named below. Google says, for its own generative search, that no special schema.org markup is needed. Microsoft, in the same period, recommends schema markup in its webmaster help; unlike everything else here, that recommendation has no source of its own below. Anyone who pulls both into a single rule is stacking two self-descriptions of two different systems on top of each other.

What remains after that is unspectacular and the same for all three questions. We group it into four building blocks; the requirements behind them come from the documents named below.

• reachable

  Can a system get to the content at all?

  Access, status code, robots.txt. That is precisely what Google lists as minimum requirements: an unblocked Googlebot, a page that responds with HTTP 200, and indexable content.

• named

  Are name, service, location and contact route on the page as text?

  Google lists as a practice making sure that important content is available in textual form. That clear or consistent company details secure rankings, mentions or inclusion in a knowledge graph is stated in none of the documents named below — this building block concerns whether a detail can be found, not what it achieves.

• answered

  Is the answer to the question at hand written out on the page?

  What a system is meant to take over has to be on the page in the first place. How strongly a written-out detail works compared with scattered hints is stated by none of the provider documents named below — this building block describes what is present, not what it brings about.

• evidenced

  Does the markup match what a human reads on the page?

  Google explicitly lists as a practice making sure your structured data matches the visible text on the page. No special markup needed therefore does not mean that markup is irrelevant — it means that Google requires no additional markup for these features.

One misunderstanding belongs cleared up at this point. The robots.txt file is a crawl control, not a lock and not a delete button. Google describes it as a file that tells search engine crawlers which URLs they may access on the site, and states in the same paragraph: it is not a mechanism for keeping a web page out of Google. Anyone who wants that should block indexing with noindex or password-protect the page.

The same page names the limits itself. A page disallowed in robots.txt can still be indexed if other sites link to it — Google phrases this as a possibility, not as the rule. Its URL may then appear in the search results, though without a description. And the instructions in the file cannot enforce crawler behaviour: whether they are obeyed is up to the crawler in question, and the rules may not be supported by all search engines. As the route intended for properly preventing a URL from appearing in Google Search results, Google names password protection on the server, the noindex meta tag or the corresponding response header, or removing the page entirely.

The robots standard itself says the same in one sentence: the rules in a robots.txt file are not a form of access authorisation. Whether the file steers AI systems, incidentally, is not stated on Google's introduction page for robots.txt — it speaks of search engine crawlers and of Googlebot.

With OpenAI a second condition is added to permission. So that a website appears in search results, OpenAI recommends allowing OAI-SearchBot in the site's robots.txt file and allowing requests from its published IP ranges. On Anthropic's help page named below there is no comparable condition. About further providers these two pages say nothing.

Overarching all of this is one sentence Google itself wrote into the Search Essentials: even if a page meets all of the requirements and best practices there, that does not mean Google will crawl, index or serve its content.

The first of these building blocks — reachable — depends technically on what an address delivers in the first place. Which paths exist for that, and why a framework name says nothing about it, is covered in the deep dive How modern websites are built and delivered.

Measurement and optimisation are not set against each other by any of the provider documents named below — Google's guide is itself an optimisation guide, and Microsoft's help article has a section of its own with content recommendations. The difference lies elsewhere: in that help article Microsoft gives recommendations, but promises no outcome and draws no causal inference.

• Google

  Search Console: impressions in generative AI features

  For the question of how content performs in the generative AI features, Google points to a report of its own in Search Console. What is counted there are impressions: how many times links to your site were shown to a user in a generative AI feature on Google Search. This can be grouped by pages, countries, dates and devices; a dimension for individual search queries is not listed in the help article.

• Microsoft

  Bing Webmaster Tools: citations as the unit of count

  The AI Performance report in Bing Webmaster Tools carries the metric Total Citations. Below it, the help article describes it as the total number of times content from your own site was visibly referenced or shown as a source in AI-generated answers during the selected date range. Next to it stands Citation Share: according to Microsoft's announcement, the percentage of citations attributed to your own site out of all citations shown across all sites for that same grounding query. Microsoft distinguishes the two explicitly — one says how often something was cited, the other what share of the citation space a site receives for a specific grounding query.

• OpenAI

  A parameter that counts clicks

  With OpenAI the measurable thing is a referral parameter: to referral URLs from ChatGPT search, ChatGPT appends utm_source=chatgpt.com, which allows clear analysis of inbound traffic. The precondition, according to the same source, is that OAI-SearchBot may access the content. What is measured with it are clicks, not mentions.

Since June 2026 Microsoft has added four capabilities to this report: Intents, Topics, Citation Share and Compare. The announcement post carries them as a preview — “in preview” — and states in the availability section that they are beginning to roll out globally. Whether every account sees them today is stated by none of the documents named below. Intents classifies the grounding queries in the AI Performance report into broader categories such as Informational, Commercial, Navigational, Learn and Solve, Research, Creation, Local and more; the post itself treats this classification as provisional.

The most remarkable sentence about these figures comes from Microsoft itself. Citation Share is, by its own account, designed as an observational metric — not a ranking system and not a competitive scoreboard; the metric shows neither competitor domains nor traffic share nor quality scores. That is a statement about the intent behind the metric, not a finding confirmed from outside.

On the same matter Google warns about tools that promise ranking success or claim to use internal Google metrics: no third-party tool has access to Google's internal ranking or AI systems. That is not a blanket rejection — in the same paragraph Google writes that you should feel free to use such tools if they help your workflow, but measure their advice against the official documentation.

What a measurement yields can depend on the circumstances — Google and OpenAI each describe this for their own systems themselves. Google names time, location, language and personalisation as reasons why search results may differ between people, and states on the same page that you may also get the same or similar results as someone else. For ChatGPT, OpenAI describes that a VPN or the network location may affect the approximate location inferred from the IP address, and that saved location information may also influence search results. And for search results it can, according to OpenAI, take around 24 hours for a change to robots.txt to be reflected in their systems.

A method follows from this — ours, not a provider's. Pixelkiez observes AI Visibility against a set of questions fixed in advance, in a named system, at a recorded point in time and in a recorded language and country version. None of the provider documents named below names this evaluation, recommends it or requires it. It makes an observation traceable — a promise about future answers does not follow from it.

Google's guide asks whether SEO is still relevant for generative AI search and answers yes. The established SEO best practices remain relevant because the generative AI features of Google Search are rooted in its core Search ranking and quality systems. Elsewhere the same page describes these systems more precisely — the core ranking systems focus on high-quality content, other systems block spam, and the generative AI features depend on both groups of systems. The bullet point this stands under warns against seeking inauthentic mentions; it notes in the same breath that the generative features can show what is being said about products and services across the web.

How often the AI answer surface appears at all is described by Google with restraint: AI Overviews are only shown when Google's systems determine that they are additive to classic Search, and accordingly often do not trigger. That is not a criterion site owners can check from outside — it is a decision made by the systems. For control, Google points to the robots.txt directives for Googlebot; they manage how pages are crawled for Search. To limit what is shown, the same section names nosnippet, data-nosnippet, max-snippet and noindex. And for training and grounding in some of Google's other systems it points to Google-Extended.

Microsoft argues in a post on its own Bing blog about the role of the index. The classic model of crawling, quality evaluation and ranking has worked extraordinarily well — and still does. In a section of its own about two systems with two responsibilities it says that classic search and grounding systems share the same foundation of crawling, understanding and ranking the web, but are optimised for fundamentally different outcomes. A further section explicitly clears up a common misconception: grounding does not replace search. It builds on the same foundational infrastructure — the same crawlers, the same quality signals, the same deep understanding of the web — and adds a new optimisation layer on top. Summed up there in two sentences: the infrastructure is shared. The purpose is different.

One qualification: the post is a position piece by four Microsoft authors on the company blog. It describes and justifies an index strategy; it is not product documentation and not a specification for site owners.

About its own search OpenAI says a third, quite different sentence: ChatGPT search sometimes partners with other search providers and then typically rewrites the question into one or more targeted queries sent to those providers. Which providers those are is not named in the body text of this page — and sometimes means sometimes.

One side finding concerns counting: sites appearing in Google's AI features are, according to Google, included in the overall search traffic in Search Console just like the rest of the search results page. That describes how Search Console counts — not how the systems relate to one another.

What follows from this is an interpretation, not a market finding. Pixelkiez reads the three self-descriptions as coexistence: classic discoverability and presence in answer systems exist side by side, and neither of the two is settled by the other. None of the provider documents named below formulates this thesis — each speaks only about its own systems. This section is about Google, Microsoft and OpenAI; about Anthropic, Perplexity or other providers it says nothing, not even a denial. And there is deliberately no figure in this section: on market share, usage share or lost clicks, no value is available from the documents named below.

• observable

  What can be measured and repeated

  The state of a page: status codes, redirects, robots.txt, and whether the load-bearing details are in the delivered document. Alongside that, the referral traffic ChatGPT marks with its parameter — clicks, not mentions. And whatever a provider counts itself: impressions in Search Console, citations in the AI Performance report of Bing Webmaster Tools.

• inferred

  What can be derived with reasons

  Microsoft writes that the values shown represent a sample of overall citation activity and may be refined as additional data is processed. It follows that a zero in the report is not evidence of a zero in reality. How large the gap is does not follow from it: Microsoft names neither share nor coverage nor margin of error — an estimate would be invented at this point.

• not verifiable

  What stays open from outside

  Whether an AI system names a particular page in a particular answer cannot be established beforehand and afterwards only for the answers actually observed. Equally open: what happens to user-triggered requests once they have been made. In the documents named below from OpenAI, Anthropic and Perplexity this is not documented.

At this point Google is clearer than the market that talks about such systems. Google states that it does not guarantee it will crawl, index or serve a page — not even if the page follows the Google Search Essentials. For the AI features the same sentence appears again: even if a page meets all requirements and best practices and complies with the policies, that does not mean Google will crawl, index or serve its content. Indexing and serving are not guaranteed. The same holds for structured data — correct markup does not force a display. And on ranking promises Google is categorical: no one, it writes, can guarantee a first place in Google. The same page explicitly warns about providers who guarantee rankings, allege a special relationship with Google or advertise priority submission.

Microsoft draws the limits for its own report just as clearly. The AI Performance report does not measure rankings, authority, performance or importance; it simply shows which content was cited. Improving content quality, structure and freshness can help strengthen content, but does not guarantee a specific citation outcome. And the causal inference is explicitly ruled out: Citation Share can help you observe whether your own share changes over time after content updates, but it does not establish a causal link.

For ChatGPT, OpenAI describes what search results are ordered by, and states in the same section: placement is not guaranteed. The same page names two conditions for a website to be eligible for inclusion at all — OAI-Searchbot may crawl the site, and the host or content delivery network allows traffic from the published searchbot IP addresses. In the other direction: if OpenAI obtains the address of a disallowed page from another source and has signals that the page is relevant to the user's query, just the link with the page title may surface in ChatGPT Atlas. The source names the remedy right along with it — the noindex meta tag.

On user-triggered requests the three documents named below diverge. Perplexity writes about its own user-triggered fetcher that it generally ignores robots.txt because a user requested the fetch. OpenAI writes for the corresponding requests that robots.txt rules may not apply because the action is initiated by a user. Anthropic commits its bots across the board to the industry standard directives in robots.txt, without making a separate promise for the user-triggered fetcher. That holds for exactly these three pages.

The robots standard says what is needed here: it is not a substitute for valid content security measures, and listing paths makes them publicly discoverable. And Google names one gap in the figures itself — data from experiments in Search Labs is not included in Search Console, because those experiments are still in active development.

Nothing follows in the opposite direction from an exclusion of guarantees. That no mention is promised in the documents named below does not mean a mention is unlikely; it means the question can only be answered by observation. Anyone who assures you of a mention in AI answers or a place in search is promising something that Google, Microsoft and OpenAI, in the documents named below, explicitly do not promise for their own systems.

Google — SEO and Search

Google describes SEO and the three stages of Google Search and states that Google does not guarantee crawling, indexing or serving even when the Search Essentials are met.

SEO Starter Guide: The Basics · In-depth guide to how Google Search works Pages dated 10 and 18 December 2025, retrieved on 4 September 2026

Google — generative AI features

Google spells out GEO and AEO, places both as SEO, and names the statement on llms.txt, the pointer to the Search Console report, the warning about third-party tools, and the rooting of the AI features in the core systems of Search.

Optimizing your website for generative AI features on Google Search Page dated 10 July 2026, retrieved on 4 September 2026

Google — AI features and your website

Google describes the shared technical foundation, the practice notes on textual form and markup, the trigger condition for AI Overviews, the controls, and how Search Console counts.

AI features and your website Page dated 10 December 2025, retrieved on 4 September 2026

Google — requirements

Google names the three technical minimum requirements and states that even all requirements and best practices met are not a promise.

Google Search Essentials · Google Search technical requirements Pages dated 10 and 18 December 2025, retrieved on 4 September 2026

Google — robots.txt

Google explains what a robots.txt file governs and what it does not, its limits, and the means intended for preventing a page from appearing in the search results.

Introduction to robots.txt Page dated 10 December 2025, retrieved on 4 September 2026

Google — exclusions of guarantee

Google states that correct markup does not force a display, and refuses any ranking guarantee.

General structured data guidelines · Do you need an SEO? Pages dated 10 July 2026 and 5 June 2026, retrieved on 4 September 2026

Google Search Console Help

The Search Console help defines impressions and names the dimensions of the report, and the exclusion of data from Search Labs.

Generative AI performance report (Search) No update date on the page, retrieved on 4 September 2026

Google Search Help

The Google Search help explains that time, location, language and personalisation can lead to differing search results.

Why your Google Search results differ from others No update date on the page, retrieved on 4 September 2026

Microsoft — Bing Webmaster Tools

Microsoft names the Total Citations metric, its distinction from Citation Share, the use of “AI Visibility” as a section heading, and the non-measurement of ranking and importance, the recommendations without a promised outcome, the ruled-out causal inference and the sample character of the data.

AI Performance in Bing Webmaster Tools No date or version given on the page; the article text only becomes visible through JavaScript. Retrieved on 4 September 2026

Microsoft — AI Visibility Insights

Microsoft gives the formula for Citation Share, lists the four capabilities in preview and the Intents categories, and places Citation Share as an observational metric. Company blog, not product documentation.

New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare Published on 16 June 2026, retrieved on 4 September 2026

Microsoft — Role of the index

Microsoft describes the shared foundation of classic search and grounding. A position piece on the company blog, not a specification for site owners.

Evolving role of the index: From ranking pages to supporting answers Published on 6 May 2026, retrieved on 4 September 2026

OpenAI

OpenAI separates OAI-SearchBot and GPTBot and names the recommendation on permission and reachability, the statement of around 24 hours for changes to robots.txt, and the handling of user-triggered requests.

Overview of OpenAI Crawlers No date given on the page, retrieved on 4 September 2026

OpenAI Help Center

The OpenAI Help Center describes the eligibility conditions and the exclusion of a placement guarantee, the influence of network location and saved location details, the partnering with other search providers, the referral parameter, and the surfacing of disallowed pages in ChatGPT Atlas.

Searching the web with ChatGPT · Publishers and Developers — FAQ Only a relative date on the pages; without a browser they respond with HTTP 403. Retrieved on 4 September 2026

Anthropic Help Center

The Anthropic Help Center describes three robots individually with three separate purposes and commits across the board to the directives in robots.txt.

Does Anthropic crawl data from the web, and how can site owners block the crawler? Page dated 7 April 2026, retrieved on 4 September 2026

Perplexity

Perplexity states that the user-triggered fetcher generally ignores robots.txt.

Perplexity Crawlers No date given on the page, retrieved on 4 September 2026

IETF — RFC Editor

The RFC states that the rules in a robots.txt file are not a form of access authorisation and are not a substitute for content security measures.

RFC 9309: Robots Exclusion Protocol Published in September 2022, retrieved on 4 September 2026

Status of this article: 4 September 2026. Provider statements can change without notice; a measurement stays bound to its date. Two of the pages named here respond to requests without a browser with HTTP 403, and one more shows its text only after JavaScript has run — anyone wanting to check will need a browser for that.