← The journalField note·September 15, 2026·9 min read

What can and cannot be measured about AI citations in 2026

Google reports impressions, Bing reports citations, Cloudflare reports crawler hits. Position within an answer and agent-completed purchases are not reported.

By Rank Sniper

Start with the boundary, not the dashboard

Three platforms publish some form of AI visibility data to site owners. Three major AI vendors publish none. Knowing which is which saves you from buying a number that does not exist.

This is an inventory of what is documented as of 15 September 2026, with the limits stated as the publishers state them.

Google: impressions, and a rollout

Google Search Console has a generative AI performance report. Its help documentation defines the core metric plainly: impressions measure "how many times links to your site were shown to a user in a generative AI feature."

Four constraints come with it, all from the same help page.

Rollout is complete, but the caveat is still printed. The help page carries "Not all properties have access to the report, as we're rolling out over time", while Google states it finished rolling these insights out to all websites worldwide as of 31 August 2026. Read an empty report as too few generative AI impressions to show, not as a missing entitlement or a queue to join.

No documented click metric. The help page documents impressions. It does not document a click metric for the generative AI report. That absence is the important part. You can see that links to your store were shown. You cannot, from this report, see what happened next.

Labs data is excluded. "Search Console doesn't include data from experiments in Search Labs." Experimental surfaces do not appear.

The 1,000-row cap applies. Search Console's standard export limit is 1,000 rows. For a Shopify catalogue with thousands of product URLs, that cap is binding. You are sampling the top of a distribution, not auditing the catalogue.

What a Shopify merchant should do with it

Treat generative AI impressions as a presence signal at the property level, not as a per-product ledger. With a 1,000-row cap and no documented click metric, the honest questions it answers are narrow:

  • Is this property appearing in generative AI features at all?
  • Which of the queries and pages that clear the cap are involved?
  • Is that set changing month over month?

The question it does not answer is whether any of it produced a session or an order. Do not build a revenue attribution model on an impressions-only report.

Bing: citations, grounding queries, and an explicit disclaimer

Bing Webmaster Tools added AI Performance in public preview in February 2026. It is currently the most specific publisher-side reporting available, and it is also the most careful about what it means.

The February announcement defines citations as "the total number of citations that are displayed as sources in AI-generated answers."

It defines grounding queries as "key phrases the AI used when retrieving content that was referenced." That is a different object from a search query typed by a person. It is the retrieval phrase, which makes it more useful for understanding why you were pulled in and less useful as a demand-volume proxy.

Then the disclaimer, which Bing put in its own announcement rather than burying: the metric "does not indicate ranking, authority, or the role of any page within an individual answer."

Read that sentence as a fence around the entire category. Bing is telling you that a citation count is not a rank, and that it says nothing about whether your page was the answer's backbone or a passing footnote.

In June 2026 Bing added Citation Share, defined as "the percentage of citations attributed to your site out of all citations shown across all sites for that same grounding query." The same release covers intents, topics and a compare view.

Citation Share is a share-of-voice figure scoped to a grounding query. It is the closest thing on the market to a competitive AI visibility metric that comes from the platform itself rather than from a third-party scraper. It still carries the February disclaimer.

The Shopify reading

Bing's AI Performance is per-property. A Shopify storefront on a verified custom domain can use it. Two specifically useful moves:

  • Look at grounding queries against your collection and product pages. These are the retrieval phrases, so they show the vocabulary the system associated with your catalogue. If the phrases describe a different product category than the one you sell, that is a content and taxonomy finding.
  • Track Citation Share on a small, stable set of grounding queries rather than across everything. Share on a churning query set is noise.

Do not convert citations into projected traffic. Bing does not publish a click metric for citations, and the disclaimer forbids the ranking interpretation that such a conversion would assume.

Cloudflare: crawler hits and referral domains

Cloudflare AI Crawl Control reports on the traffic side rather than the answer side. Its documentation for analysing AI traffic lists "Total requests, allowed requests, unsuccessful requests, and total referrals" and "Top domains sending AI-driven referral traffic", with access to the underlying data through the GraphQL API.

Referral analytics are gated to paid plans.

This is a genuinely different measurement. Search Console and Bing tell you about appearances in answers. Cloudflare tells you which AI crawlers hit your origin, whether they were allowed, and which AI domains sent visitors back.

A separate Cloudflare change matters as of today. In its July 2026 post, Cloudflare described three crawler categories, Search, Agent and Training, and announced that from 15 September 2026, on ad-monetized pages, it blocks "Training and Agent" by default while Search stays allowed. The stated rationale: "An ad is a signal that a website owner meant for a person to land there and see it."

The Shopify caveat, labelled as an inference

Cloudflare's data requires traffic to pass through your own Cloudflare zone. A standard Shopify storefront is served by Shopify's infrastructure.

The reasonable reading is that this measurement route is available to you only if you operate your own Cloudflare zone in front of a domain you control, and that a default Shopify storefront does not give you these logs. If you also run a content site, a headless front end, or a documentation domain behind Cloudflare, that property can be instrumented even when the storefront cannot.

Check your own DNS and proxy configuration before assuming either way. This is an infrastructure question about your specific setup, not a general rule.

The vendors that report nothing

This is the shortest section and the most consequential.

OpenAI. The crawler documentation describes OAI-SearchBot as "used to surface websites in search results in ChatGPT's search features", and states that "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." GPTBot crawls content "that may be used in training our generative AI foundation models". ChatGPT-User is "not used for crawling the web in an automatic fashion."

That documentation covers access control. There is no publisher-side citation reporting in it.

Anthropic. The crawler support article describes ClaudeBot collecting "web content that could potentially contribute to their training", states that "Claude-User supports Claude AI users", and that Claude-SearchBot "navigates the web to improve search result quality for users."

Again, access control. No publisher citation reporting.

Perplexity. The bots documentation states that PerplexityBot "is not used to crawl content for AI foundation models", and that Perplexity-User "generally ignores robots.txt rules" because it is user-initiated.

No publisher citation reporting.

So for three of the most visible AI answer surfaces, there is no first-party report telling a merchant that their store was cited. Any number claiming to measure it is inferred from outside, by sampling prompts or by parsing referrer strings, and it should be labelled as such by whoever sells it to you.

What is genuinely unmeasurable today

Two things specifically, and it is worth being blunt because both are routinely promised.

Position within an answer. Bing states its citation metric "does not indicate ranking, authority, or the role of any page within an individual answer." Google's generative AI report documents impressions, not placement. No publisher-side source documents where inside a generated answer your citation sat, how prominent it was, or whether it was the sentence the reader acted on.

There is no "AI rank". The platforms that report citations have said in writing that their citation counts are not ranks.

Attribution of agent-completed purchases. When a shopping agent completes a transaction, no publisher-side report from any of the sources above connects that order back to a specific AI answer, citation or prompt. Google's generative AI report has no documented click metric. Bing's citation metrics carry an explicit non-ranking disclaimer and no revenue linkage. Cloudflare reports referral domains, which is traffic, not orders.

If a tool tells you how much revenue AI search produced for your store, ask which documented platform metric that figure derives from. As of today, none of the sources in this article publish one.

A measurement plan that only uses what exists

Keep it small and keep it honest.

  • Search Console. Check whether the generative AI performance report is available on your property. If it is, record impressions monthly at the property level and note the 1,000-row cap on any breakdown. Do not model clicks.
  • Bing Webmaster Tools. Verify the storefront domain, open AI Performance, and export citations and grounding queries. Pick a fixed set of grounding queries relevant to your categories and track Citation Share on that set only.
  • Cloudflare. Applicable only if a domain you control sits behind your own Cloudflare zone, and referral analytics require a paid plan. If so, record crawler request volume by category and referring AI domains.
  • Your own analytics. Referrer-based sessions from AI domains are your own data and you can segment them. Treat them as one signal, not as citation measurement, and be aware that an agent acting on a user's behalf may not send a referrer at all.
  • Everything else. Label it as inference. If the number did not come from one of the platform reports above, write down how it was produced.

A measurement programme built on these four items will be thinner than the ones being marketed. It has the advantage of being defensible line by line.

Related reading

Sources

Run a free scan →Get the next dispatch →
Keep reading
Field note·September 15, 2026

What Google actually says about being surfaced in AI Overviews and AI Mode

Google documents one entry condition for generative AI features, no special markup, and three separate control planes. Here is what that means for a Shopify store.

Read →Field note·September 15, 2026

How shopping agents actually get your Shopify product data

Agents read Shopify Catalog, Merchant Center attributes and vendor feeds, not your theme HTML. Your structured fields decide what they can say about you.

Read →
What can and cannot be measured about AI citations in 2026 — Rank Sniper