Most AI-search disagreements are not about strategy. They are about vocabulary. A screenshot of a generated answer, a count of links, a rank position and a conversion report are four different measurements taken from four different systems, and a proposal that treats them as one chain will be impossible to audit.

The four questions to keep separate

Before any work begins, agree on which of these four an engagement is actually meant to move. They differ in who can observe them, how they are produced, and what evidence exists that they happened.

The four-way distinction, as described by the vendors whose systems are involved
QuestionWhat it asksWhat a vendor documentsWhat that does not establish
AppearanceDid content from your site contribute to a generated answer, whether or not a link was shown?Google describes AI Overviews and AI Mode as retrieval-based features that use the Search index and may run multiple related queries (“query fan-out”) to assemble a response and supporting links (s1, s2).That your page was the source of a specific sentence. Generated answers vary between features and over time; Google states AI Overviews and AI Mode may use different models and techniques.
CitationWas your page displayed as a linked supporting source for that answer?Google says a page must be indexed and eligible to be shown with a snippet in Google Search to be eligible as a supporting link, with no additional technical requirements (s1). Microsoft states that Copilot Search cites sources with inline links to passages and provides access to cited sources (s8).That a citation recurs, that it is stable, or that it says anything about how much traffic it sends.
IndexingCan the page be retrieved and served at all?Google states that meeting requirements, best practices and policies does not mean Google will crawl, index or serve the content — indexing and serving are not guaranteed (s2).That a robots.txt rule is the right blocker if you actually want a page excluded; Google notes disallowed URLs can still be indexed when linked elsewhere and points to noindex or password protection instead (s11).
Traffic and outcomeDid a person arrive, and did they do something you value?Google says sites appearing in AI features are included in overall Search traffic and reports AI-feature performance in Search Console (s1).That a citation produced a click. Citation is a citation; only your own analytics can tell you about visits, but its attribution will be imperfect and will not be fixed by vendor reporting.

Why the vendors cannot settle this for you

Google is unusually direct about the limits of third-party measurement: it advises being wary of tools that promise ranking success or claim to use “internal” Google metrics, and states that no third-party tool has access to its internal ranking or AI systems (s2). That is the single most useful sentence to keep in front of you when a supplier presents a visibility score.

The vendor-documented reporting that does exist is explicitly observational. Microsoft describes Bing Webmaster Tools AI visibility features — Intents, Topics, Citation Share and Compare — as preview capabilities, notes that labels may be broad for niche domains, and says Citation Share does not expose competitor domains, does not represent traffic share, and does not assign quality scores (s9). A citation rate is therefore not a revenue forecast, and should never be presented as one.

Access controls are not rankings

A frequent confusion in proposals: “making the site crawlable” is described as though it were an optimisation that improves citation share. In practice it is a permission gate, and gates are documented per crawler rather than per outcome.

OpenAI publishes separate, independent controls. OAI-SearchBot is used to surface sites in ChatGPT search features; sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links. GPTBot governs crawl content that may be used in training foundation models, and is a separate setting. ChatGPT-User covers user-initiated actions, is not used for automatic crawling, and is not the control for search opt-outs (s4). OpenAI also notes it can take roughly 24 hours from a robots.txt change for its systems to adjust.

And robots.txt is only one gate. OpenAI’s advertiser guidance warns that a robots.txt rule alone may be insufficient because firewalls, CDNs, bot-protection, login walls, CAPTCHAs or browser challenges can separately block crawler access, producing 403, 404, 429 or blank and challenge pages — and that fixing access does not guarantee approval (s6). That article concerns ad landing pages, not organic inclusion. We adapt its diagnostic idea — check what the crawler actually received, not what your browser rendered — as our own checklist rather than as OpenAI guidance on organic AI-search inclusion.

What the primary sources say actually matters

Google states there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations are necessary, and that you do not need new machine-readable files, AI text files or special markup — Google Search ignores LLMS.txt (s1, s2). Its own guide says creating unique, non-commodity, people-first content with a distinct point of view will likely matter more for presence in generative AI search in the long run than any other suggestion in that guide, and it explicitly recommends against “chunking” content, rewriting for AI systems, keyword-variation padding and seeking inauthentic “mentions” (s2).

Those are qualitative statements about Google Search only, with no effect size attached, and they say nothing about other engines. Treat them as design constraints, not as a scoring rubric.

Wooden study desks inside the Ewing Branch of the Mercer County Library System, Ewing Township, New Jersey.
Study desks at the Ewing Branch of the Mercer County Library System, Ewing Township, New Jersey, May 2025. Photograph by Famartin; CC BY-SA 4.0. Cropped, resized and converted to WebP. Famartin · CC BY-SA 4.0 · source

Put the distinction into the brief

The practical output of this article is a short set of clauses. Add them to your written scope before any supplier starts, so that a future disagreement is about delivery rather than about definitions.

  1. Name the target. State whether the engagement is meant to change appearance, citation, indexing eligibility or traffic — and which of those will be reported.
  2. Name the surface. Specify the product and feature, since the vendors document their own systems only. Google’s guidance covers Google Search; OpenAI’s covers OpenAI products.
  3. Name the evidence type. Distinguish between a vendor-documented report, your own analytics, and a manually observed example. Say who is allowed to produce each, and how a manual example is recorded (date, surface, URL, screenshot location) so it is reproducible.
  4. Prohibit unverifiable metrics. Write down that no deliverable may claim access to a search engine’s internal systems, because no third-party tool has it (s2).
  5. Separate permission work from content work. Crawler access is an on/off decision made per user agent; content quality is editorial. Report them as different line items.
  6. Say what will not be claimed. Explicitly state that citations will not be converted into traffic or revenue estimates, and that indexing is never guaranteed (s2).

Questions that expose a blurred proposal

Ask these before signing, and ask for answers in writing:

  • Which of the four questions does each deliverable address, and which system produced the number?
  • If you show a citation, is it a screenshot you captured yourself, and will you record the date, the surface and the exact URL?
  • How will you distinguish “not indexed” from “indexed but not cited”?
  • If we change crawler permissions, what is the expected lag before any effect could plausibly be observed, and how will you avoid attributing unrelated movement to that change?
  • Will any report contain an estimate of traffic or revenue derived from a citation metric? If so, on what measured basis?
  • Which vendor documentation are you relying on, and does it describe the engine you are targeting?

Where your own evidence comes from

You can act on this without buying anything. Server and application logs will show whether specific crawlers requested specific URLs, which is the most direct evidence available for the indexing and permission questions — provided your CDN or bot-protection layer is not blocking them first (s6). Your analytics will show arrivals and conversions, subject to normal attribution limits. What remains genuinely unobservable is the internal reasoning behind any particular generated answer; Google states no third-party tool has access to its internal ranking or AI systems, so no supplier can credibly demonstrate that reasoning.

Next step

Write the four-way distinction into a document your supplier has to agree to. The scope builder produces a private Markdown brief covering your business and public site, audience, a fixed intent choice, the evidence you require, delivery questions, success criteria and constraints. Use it before reviewing proposals: Review an AI-search proposal, Specify evidence before visibility claims, and Define an AI-search scope of work.

Sources and limits

All statements above are attributed to vendor documentation describing the vendor’s own systems, retrieved November 2026; vendor pages carry their own update dates and may change. Google’s pages describe Google Search AI features only; OpenAI’s describe OpenAI products only; Microsoft’s describe Microsoft products, and the AI visibility features cited were in preview. No statement here is an independent test of citation, traffic or ranking behaviour, and this publication publishes no ranking estimates, runs no automated third-party query checks and does not use Search Console. Primary sources: Google, “AI features and your website” (s1); Google, “Optimizing your website for generative AI features on Google Search” (s2); OpenAI, “Overview of OpenAI Crawlers” (s4); OpenAI Help Center, “Advertiser Guidance for Allowing OpenAI Web Crawlers” (s6); Microsoft Bing, “Introducing Copilot Search in Bing” (s8); Microsoft Bing, AI visibility insights in Bing Webmaster Tools (s9); Google, “Introduction to robots.txt” (s11).