Answerability: does your website answer concrete questions?

What a search result shows is described by Google for its own result descriptions: not the whole page, but an excerpt from it. Snippets, it says, are created automatically from the page content and are meant to emphasise and preview the part that best relates to a user's specific search. That is why, the same page notes, Google Search may show different snippets for different searches.

Google describes the technical side of this briefly. For a page to be eligible to appear as a supporting link in AI Overviews or AI Mode, it must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the search technical requirements; there are, it says, no additional technical requirements. That describes eligibility, not an effect. The page qualifies for consideration — whether it turns up in an answer is not what the sentence says.

Microsoft describes what shifts here in a blog post from its advertising arm. In traditional search, it says, visibility meant appearing in a ranked list of links; in AI search, ranking still happens, but it is less about ordering entire pages and more about which individual pieces of content earn a place in the final answer. This is a blog post and not product documentation, and it speaks of AI search in general, not only of Microsoft's own.

For web search in ChatGPT, OpenAI draws the boundary of the same thought. Results are ranked using multiple factors intended to help users find relevant, reliable information — placement, it states, is not guaranteed.

The question, then, is not whether your page can be read, but whether anyone knows where they stand after reading it. For a business this becomes concrete in very ordinary places. The six cases below are our own grouping and not a provider's list: none of the sources named below carries them as categories relevant to answering.

1. 01 · Price

   Roughly what does this cost?

   “Fair prices” is not an answer to this question. A range, a unit rate or a worked example is one — even when a caveat comes with it.

2. 02 · Service area

   Do you come out to me?

   “Berlin and surrounding area” answers the question for Pankow, not for Bernau. The business knows exactly where the boundary runs — the page often does not say.

3. 03 · Process

   What happens after I get in touch?

   First contact, appointment, quote, execution. Inside the business the process is self-evident, and for exactly that reason it is rarely written down.

4. 04 · Prerequisites

   What do I need to bring or prepare?

   Documents, access to the property, preparatory work, a minimum order. Anyone who only learns this on the phone cannot weigh it up beforehand.

5. 05 · Scope

   Do you do exactly what I need?

   The real question often sits between a service heading and the customer's specific case. A list of umbrella terms does not answer it.

6. 06 · Times

   When can I reach you, and from when do you have capacity?

   Availability by phone and current workload are two different pieces of information. Both help decide whether anyone gets in touch at all.

In all six cases the page is readable. Whether the answer is on it is a second, independent question — and only the second is what this article means by answerability. In this article the term is our own label for that check.

That is how the graphic is built, too. The three conditions it names — explicit, in the right context, verifiably supported — are the structure of the next three sections and our own grouping, not a provider's model.

This section rests on the thinnest evidence in the whole article, and that belongs at the top rather than in a footnote. It lists what the documents named below actually say. A justification for why a written-out detail works better than a scattered hint is not in them — and therefore it is not here either.

Five statements, each tied to its document.

1. 01 · in textual form

   Google lists making important content available in textual form as a practice.

   The point sits under a list heading that places it itself: specific optimisation is not required for AI Overviews and AI Mode, but all existing SEO fundamentals continue to be worthwhile — for example these. It is a recommended practice, not a requirement.

2. 02 · matching what is visible

   You may only mark up what can also be seen on the page.

   In the same list Google gives making sure your structured data matches the visible text on the page as a practice. In the structured data guidelines this becomes a rule: content that is not visible to readers of the page should not be marked up; if the JSON-LD markup describes a performer, the HTML body must describe that same performer.

3. 03 · not hidden

   Microsoft advises against hiding important answers in tabs or expandable menus.

   The reason is in the same sentence: AI systems may not render hidden content, so key details can be skipped. Twice in the possible mood — the source does not say that it happens, but that it can happen. Document type: blog post by Microsoft Advertising.

4. 04 · retrievable without a browser

   Anthropic's web fetch tool does not execute JavaScript.

   The product documentation describes the API retrieving the full text content from the specified URL, and records in the same section that websites rendered dynamically with JavaScript are currently not supported. “Full” is qualified by the same page itself: if the fetched content exceeds an adjustable limit, the tool truncates it, and cached content may not always reflect the latest version available at the URL. For pages that need real browser rendering, Anthropic points directly afterwards to a different tool. This applies to this one tool, not to AI systems in general.

5. 05 · structured

   Clear headings, tables and FAQ sections help with referencing, according to Microsoft.

   The Bing Webmaster blog names them as one of five equally weighted recommendations: they help surface key information and make content easier for AI systems to reference accurately. The verb is “help”, not “cause”, and the post is a blog post. About markup it says nothing at this point, neither for nor against.

For completeness, what structured data is in Google's own description belongs here too: explicit clues about the meaning of a page. Google Search works hard to understand the content of a page, the introduction writes — and you can help by providing such clues on the page. “Can help” is the source's wording. That markup makes a system understand the content is not stated there.

Three sentences are deliberately missing here. That explicit, contextualised details beat scattered hints is said by none of the pages named below; none of them draws that comparison. That sensible HTML structure raises answerability is likewise said by none: Microsoft describes it with “help” and elsewhere frames its structural recommendations explicitly as general practices you may consider. And that visible HTML text is the sole basis of processing is in none of them — they describe text as the reliable part, and not one of them uses the word “only”.

Can a person or a machine answer a real buying or clarification question from this page without guessing? That is the question anyone can answer on any page for themselves, with no provider document at all. Where the answer is no, the next step is not markup but one more sentence in the visible text: under Google's guidelines markup may only describe what is already there anyway.

In order, because here the type of document decides how much weight a statement carries. The Bing Webmaster Guidelines are a guideline text from the provider, and their opening sentence draws the scope itself: the guidelines describe how Bing discovers, crawls, indexes, evaluates and surfaces content across Bing search experiences, Copilot and grounding API results. By Microsoft's own statement, then, what stands there applies not only to traditional search.

The core of the section on content sits in one sentence: grounded answers and citations depend on content that Bing can clearly interpret and verify. A list of shoulds hangs from it, and one of its points is that content should be easy to understand without external context. Elsewhere Microsoft phrases a likelihood and not a condition: URLs are more likely to be selected for grounding queries and citations when content stands on its own. Under the heading “Ensure:” the same section lists, among other things, that key statements do not rely on implied content.

On ambiguity itself Microsoft becomes explicit. Ambiguous references are to be avoided. Clear entity definition improves grounding visibility and citation accuracy. People, organisations, products and locations should be named clearly and consistently, and unrelated concepts should not be mixed on a single URL. How much an unambiguously worded page gains from this Microsoft does not say: the verbs are “improves” and “more likely” — no order of magnitude appears in any of these sentences.

This is also meant to fit together across formats. The Bing Webmaster Tools product documentation carries consistency across formats as a recommendation: text, images and other media should describe the same products, entities and concepts. The guidelines carry the related rule that images and video should reinforce the primary text on the page and should not be the sole source of the information required to understand the topic.

The Google pages named below treat ambiguity as a property of the query. With Google the case therefore lies differently, and that is the most important distinction in this section. The rating guidelines for external evaluators give the example themselves: many queries have more than one meaning — the query “apple” might refer to the computer brand or to the fruit — and Google calls these possible meanings “query interpretations”. That document, however, is a rating guide for human evaluators and not a description of ranking; right at its start it records itself that no single rating can directly influence how a page appears in Google Search.

Microsoft describes ambiguity on the query side as well. Its transparency account states that many queries may have more than one possible intent; Bing tries to provide a comprehensive set of results that reflect all possible intents. And Bing presumes in doing so that the user seeks high-quality, authoritative content — unless the user clearly indicates an intent to research low-quality content. This is a support and transparency article, not a guideline with binding character.

On the ambiguity of page content, the Google pages named below say nothing. What they do say concerns how far content explains itself, and it stands as a question rather than a requirement: whether the content provides a substantial, complete or comprehensive description of the topic — and whether it leaves readers feeling like they need to search again to get better information from other sources.

That content appears as an excerpt away from its own page is something Google describes explicitly for featured snippets: sometimes when you search you find a box with the answer right at the top, and this box shows a little piece of a website that helps answer your question. That help article is aimed at end users and is deliberately kept simple. The developer documentation answers the obvious question from site owners briefly: you can't mark this up. Google's systems determine themselves whether a page would make a good featured snippet for a search request, and if so elevate it. What is denied is positive markup, not every form of influence — the same page does name means of limiting what is shown.

One characteristic from Microsoft's advertising blog belongs here, but only together with where it stands. Under the question of what makes content eligible for featured snippets, the post names self-contained phrasing: sentences that make sense even when pulled out of context. On selection in AI answers the same post says something plainer — there is no secret sauce that guarantees selection in AI answers, and nothing guarantees selection.

Grounding and featured snippets are two different cases. Microsoft's clarity rules consistently name grounding results and citations, never featured snippets. Conversely, for featured snippets no context or ambiguity rule stands in any of the sources named below. Anyone mixing the two builds a rule out of two systems that neither of them has set.

One more observation on the distribution. Of the sources named below, only the Bing Webmaster Guidelines address ambiguity and missing external context explicitly. On the question of how OpenAI, Anthropic or Perplexity handle ambiguous details, none of the sources named below says anything — so nothing stands here on that, neither a statement nor a denial.

In everyday practice this hangs on wordings that sound harmless. None of the sources named below gives examples from the everyday life of small businesses; Microsoft's own examples concern product pages. The following example is therefore ours and explicitly invented. A Berlin trades business writes: “We are here for you when it has to be quick.” To a human being that is a promise. As an answer to the question of who repairs a broken heating system in Kreuzberg on a Saturday, the sentence carries nothing: it names neither the place nor the day nor the service. We are not saying that such a sentence prevents a mention. We are saying that it does not answer the question.

Google's structured data guidelines put it in one memorable sentence: using structured data enables a feature to be present — it does not guarantee that it will be present. The paragraph goes on to say that Google's algorithm tailors search results to create what it thinks is the best search experience for a user, depending on many variables. Beside it stands the warning: Google does not guarantee that your structured data will show up in search results, even if your page is marked up correctly according to the Rich Results Test.

The same guidelines tie markup to the content. Structured data must be a true representation of the page content; as a counterexample Google names a woodworking site labelling instructions as recipes. For time-sensitive content that is no longer relevant, Google states that it will not show a rich result — which conversely does not mean that current content would get one. And if a page breaches the guidelines, that can result in a manual action: the page loses eligibility for rich results, while its ranking in web search is untouched by it.

For the AI features of its own search, Google draws the line more tightly still. You don't need to create new machine readable files, AI text files or markup to appear in them; nor is there special schema.org structured data that you need to add. The guide to generative AI search carries the same thought with an addition: structured data is not required for it, but it remains a good idea to keep using it as part of your overall SEO strategy, because it helps with being eligible for rich results on Google Search.

“No markup you need to add” is not the same as “no such markup exists”. The source denies an obligation, not an existence. That statement also applies to the named Google search features and carries nothing about other systems.

For Bing grounding Microsoft words things cautiously. Structured data may support clearer grounding but does not guarantee visibility or grounding traffic — and markup must accurately reflect visible content. Among the Microsoft sentences in the documents named below, that is the only one with a “must”. It continues that markup which is irrelevant, inaccurate or misleading may be ignored and can affect trust and eligibility for enhanced features.

These two self-descriptions do not add up to a shared provider line, and netting them into one would be the more convenient but wrong route. Google says for its own generative search that no special schema.org markup is needed. Microsoft writes for its own grounding that markup may support, and demands in the same breath that it match the visible content. Two providers, two systems, two sentences — the difference belongs left standing rather than smoothed over.

How little markup carries on its own is shown by a case from Google's own changelog. The FAQ rich result has been withdrawn. The entry from May 2026 announces it in the future tense — the feature will no longer appear in Google Search starting 7 May 2026 — and the entry from June 2026 records the state and gives the reason for a second step: the documentation for the FAQ rich result is being removed because the feature is no longer shown in Google Search results. Anyone reading “since 7 May 2026” is therefore reading both entries together. Historically it was a withdrawal in stages: in September 2023 the same changelog recorded that the FAQ documentation had been updated to state that the feature is only shown for well-known, authoritative government and health websites. That entry dates a documentation change. Today the display is shown for nobody any more.

That an FAQ section in the markup produces an AI answer or a citation is promised by none of the provider pages named below. The withdrawal shows at the same time how quickly a display that once existed can disappear again. The statement concerns the rich result in Google Search — not FAQPage markup as such, and not other search engines or AI systems.

The vocabulary itself is more modest than its reputation suggests. Schema.org explicitly calls itself not a formal standards body but simply a site where the schemas that several major search engines will support are documented. And it records that not every type of information in schema.org will be surfaced in search results; for concrete use it refers to each company's documentation. The text is recognisably older and carries no page-specific modification date.

About the remaining providers this section deliberately says little, because the sources named below give no more. On OpenAI's help page for publishers and developers, structured data, schema and FAQ markup do not appear; what the page names for inclusion in summaries and snippets is access for the OAI-SearchBot. That is an observation about this one page and not a statement about OpenAI's documentation as a whole. For Anthropic, the only page on structured markup as a selection criterion is the web fetch tool page named below.

The following sequence of four questions is our own model. Answerability is the third of them. This sequence of terms appears in none of the provider documents named below; that it exists elsewhere is not thereby ruled out. We use it here because it already appears in the article “How AI reads a website” and both articles should speak the same language.

1. 01 · Discoverability

   Can a system fetch the page at all?

   Access, redirects, status codes, robots.txt. Without an open door nothing further happens.

2. 02 · Extractability

   Can the facts be extracted reliably?

   Service, location, responsibility, contact route — as text a system can assign unambiguously, not merely as an impression.

3. 03 · Answerability

   Does the page answer a concrete question unambiguously?

   That is what this article is about. Readable details do not turn into an adoptable answer by themselves — for that, the question has to be genuinely answered on the page.

4. 04 · Observed AI Visibility

   Does the domain actually appear in AI answers?

   That shows only in a defined, repeatable test — observed, not inferred.

What we look at for stage three is the verifiable state of the page: whether the detail in question stands in the text as delivered; whether it remains understandable without the surrounding context; whether the markup says the same as the visible text; whether the wording for service, place and scope stays the same throughout. These are questions about the page. None of them is a question about somebody else's system, and none of them answers what such a system will do tomorrow.

We do not give a score for this. None of the sources named below names a threshold above which a page counts as answerable — a number at this point would be our invention with the varnish of a measurement. Instead we describe which question the page answers today and which it does not.

Google draws the boundary itself in two places, and both belong named together. The structured data guidelines say that markup enables a feature to be present and does not guarantee it. The page on AI features carries the second half: just because a page meets all requirements, best practices and complies with the policies does not mean that Google will crawl, index or serve its content — indexing and serving are not guaranteed. People-first content, incidentally, Google describes as content created primarily for people and not to manipulate search engine rankings.

For tools that present it differently, Google has a paragraph of its own. Third-party tools do not have access to Google's internal ranking data, they can't guarantee performance, and any predictions are their own — which, like predictions generally, may not happen. That applies to us as much as to anyone else offering such analyses.

Of the sources named below, only Microsoft carries a metric of its own for presence in AI answers — and it is at the same time the best evidence for the restraint of this section. The AI Performance report in Bing Webmaster Tools carries the metric “Total Citations”: by the product documentation, the total number of times your content was visibly referenced or shown as a source in AI-generated answers during the selected date range. Beside it stands “Citation Share”, and Microsoft distinguishes the two explicitly — one says how often you were cited, the other what share a site has of all citations shown across all sites for that same grounding query. That reference figure comes from the announcement post for the public preview; the same post states that the figure indicates neither placement nor presentation within a specific answer.

Microsoft formulates four caveats about this report itself. It 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. Citation Share can help you observe whether your share changes over time after content updates but does not establish a causal link. And the values shown are a sample of overall citation activity, which may be refined as additional data is processed. How large that sample is Microsoft does not say — neither share nor coverage nor margin of error stands there. A zero in the report is therefore no evidence of a zero in reality.

It would be just as wrong to say that Microsoft is silent about optimisation. The report has a recommendation part of its own, which Microsoft itself places: general content best practices you may consider when using this data. Named there are descriptive headings, concise sections, tables and FAQ-style content, so that information is easier to understand and reference in AI answers. Recommendations, then, yes — a promised outcome and a causal inference explicitly no. And what a citation means the same documentation delimits in its FAQs: that your content was visibly referenced or shown in an AI-generated answer. It does not stand for traffic, clicks or user engagement.

A note on the status of these capabilities belongs with it. Microsoft carries several of them as a preview. Whether every account sees them today is stated by none of the pages named below.

What we promise is an assessment you can follow — not a mention, not a citation, not a placement. We tell you which customer question your page answers today, which it does not, and what can be changed about that. What an answering system makes of it is that system's decision. Anyone promising you more at this point is promising something that Google and Microsoft, in the documents named below, explicitly do not promise for their own systems.

Google — AI features and your website

Google names the eligibility condition for supporting links in AI Overviews and AI Mode, the practice notes on textual form and on markup matching the visible text, the statement that no special markup is needed, and the sentence that indexing and serving are not guaranteed.

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

Google — snippets

Google explains that snippets are created automatically from the page content and may differ between searches.

Control your snippets in search results Page version 20 April 2026, retrieved on 4 September 2026

Google — featured snippets

Google describes the excerpt box for end users; the developer documentation answers that site owners cannot mark a page up as a featured snippet.

How Google's featured snippets work · Featured snippets and your website The help article carries no date, the developer documentation page version 10 December 2025. Retrieved on 4 September 2026

Google — structured data

Google describes structured data as explicit clues about the meaning of a page and names the guidelines: markup only for visible content, a true representation, no rich result for outdated time-sensitive content, no display guarantee, and the manual action without any ranking effect.

Introduction to structured data markup in Google Search · General structured data guidelines Page versions 10 December 2025 and 10 July 2026, retrieved on 4 September 2026

Google — generative AI search

Google states that structured data is not required for generative AI search, that there is no special schema.org markup to add, and that using it as part of overall SEO work nevertheless remains worthwhile.

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

Google — changelog

The changelog records the withdrawal of the FAQ rich result in two entries — announcement in May 2026, statement of the condition and removal of the documentation in June 2026 — plus the historical entry from September 2023.

Latest Google Search Documentation Updates Page version 31 August 2026, retrieved on 4 September 2026

Google — people-first content

Google defines “people-first content” and poses the two self-assessment questions about how complete a piece of content is and about readers having to search again.

Creating helpful, reliable, people-first content Page dated 10 December 2025, retrieved on 4 September 2026

Google — third-party tools

Google states that third-party tools have no access to Google's internal ranking data, cannot guarantee performance, and that any predictions are their own.

Google Search's guidance on using third-party SEO tools, services, and advice Page version 5 June 2026, retrieved on 4 September 2026

Google — rating guidelines

The rating guide covers the ambiguity of the query together with “query interpretations” and the self-imposed caveat that no single rating directly influences how a page appears in Google Search. A rating guide for external human evaluators, not product or ranking documentation.

Search Quality Rater Guidelines: General Guidelines (PDF) Header date 11 September 2025, retrieved on 4 September 2026

Microsoft — Bing Webmaster Guidelines

Microsoft explains the scope covering Copilot and grounding results as well, and names understandability without external context, selection stated as a likelihood, avoiding ambiguous references, clear entity naming, one topic per URL, images and video as reinforcement rather than sole source, and the limits of structured data for grounding.

Bing Webmaster Guidelines No date given on the page; the article text only becomes visible through JavaScript. Retrieved on 4 September 2026

Microsoft — Bing Webmaster Tools Help

The Bing Webmaster Tools help lists the metrics of the AI Performance report, the four self-imposed limits, the framing of the content recommendations as general practice, consistency across formats, and the definition of what a citation means.

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

Microsoft — Bing Webmaster blog

The Bing Webmaster blog recommends structure and clarity as well as reducing ambiguity across formats, and names the citation count without any statement about placement or presentation. Blog post and product announcement, not product documentation.

Introducing AI Performance in Bing Webmaster Tools Public Preview Published on 10 February 2026, retrieved on 4 September 2026

Microsoft Advertising — blog

Microsoft Advertising describes how AI search selects individual pieces of content rather than whole pages, advises against hidden answers in tabs and expandable menus, names self-contained phrasing as a characteristic for featured snippets, and issues two refusals of any selection guarantee. Blog post, not product documentation.

Optimizing Your Content for Inclusion in AI Search Answers Published on 8 October 2025, without an update date, retrieved on 4 September 2026

Microsoft — Bing support

The Bing support states that many queries may have more than one possible intent, and presumes that the user seeks high-quality, authoritative content unless they clearly indicate otherwise. Support and transparency article, not a guideline with binding character.

How Bing delivers search results Page version March 2025, retrieved on 4 September 2026

OpenAI Help Center

The OpenAI Help Center describes the ranking of search results by multiple factors together with the exclusion of any placement guarantee; the publisher page names access for the OAI-SearchBot for inclusion in summaries and snippets.

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

Anthropic states that the web fetch tool retrieves the text content of a URL, currently does not support pages built by JavaScript, truncates content at a limit, and may return cached versions.

Web fetch tool No date given on the page; the URL redirects to platform.claude.com. Retrieved on 4 September 2026

Schema.org

Schema.org describes itself as a documentation site rather than a standards body and states that not every type of information in the vocabulary is surfaced in search results.

Schema.org FAQ No page-specific date, site version V30.0 of 19 March 2026, retrieved on 4 September 2026

State of this article: 4 September 2026. Provider statements can change without notice; a measurement stays tied to its date. Two of the pages named here answer non-browser requests with HTTP 403, and two more only show their text after JavaScript has run — anyone wanting to check will need a browser for that.