AI tools can request your pages and cite your content while your analytics records no AI referral visits. Understanding that apparent gap starts with what each report measures.
Bot requests, citations and referral visits sound like steps in one journey. They are measured by separate reports, and none of them tells you what the others mean.
To show the difference, we used a small, seven-day snapshot from our own Microsoft Clarity dashboard, read on September 27, 2026: 1,348 AI bot requests, 12 AI citations and zero tracked AI referral sessions. The numbers demonstrate how to read a dashboard. This snapshot alone cannot establish whether an AI visibility strategy is working.
Key takeaways
- A bot request means a machine asked your server for a file. It does not show that a page was read, cited or seen by a person.
- A citation means an AI answer referenced one of your pages. It does not show that anyone clicked.
- Zero tracked AI referral sessions is not the same as zero people finding you through AI.
Why these numbers are not a funnel
Each layer of AI visibility answers a different question, and each one comes from a different report.
Access: 1,348 requests. Did AI systems request files from the site?
Source: Clarity Bot Activity, via Cloudflare.
Citations: 12 citations. Did AI answers reference our pages?
Source: Clarity Citations, “Microsoft Copilot and partners”.
Visits: 0 tracked sessions. Did people arrive from an AI tool?
Source: Clarity AI referral metric.
Outcomes: Not verified. Did any of this produce inquiries, calls or sales?
Not verified in this review.
These reports do not track the same journey from request to citation to visit. Microsoft’s Bot Activity documentation says so directly: bot requests do not indicate content was cited or surfaced in AI answers, and may not translate into traffic or outcomes.
For the same reason, don’t treat ratios between these totals, such as 12 ÷ 1,348, as conversion rates. The totals do not describe a linked group of people.
Each report was set to “Last 7 days.” We did not verify the exact start date, end date or reporting timezone, so this is a dated snapshot rather than a defined study period.
What do AI bot requests actually include?
A request is one machine asking the server for one file. It is not a visit, and it is not a reader.
Clarity classified 984 of the 1,348 requests as AI crawlers, 215 as AI search and 149 as AI assistants. Of those requests, 696 were classified as HTML, not necessarily 696 different pages. The rest were style sheets, scripts, images, data files and resources Clarity could not classify.
- /robots.txt: 82 requests
- /sitemap.xml: 43 requests
- /start: 28 requests
- /blog: 27 requests
- /css/service-links.css: 27 requests
- /css/agency.css: 27 requests
- Homepage: 26 requests
- The article that received the citations: 5 requests
These are single path rows, not totals combined across www, non-www, HTTP and HTTPS versions. The request total includes instructions, style sheets and other resources, not just articles.
The requests also got different responses: 850 succeeded, 337 were redirected and 161 were unsuccessful. Redirects are not automatically a problem, and nothing here shows the unsuccessful requests affected citations or visits.
Clarity’s operator breakdown included Google 156, OpenAI 129, Anthropic 120, Perplexity 96 and Microsoft 37. Those counts show who requested files, not who cited the site.
What does a citation count actually show?
A citation means an AI-generated answer referenced one of your pages. In this snapshot, all 12 went to one article: The Em Dash Is Not Proof of AI. Treating It That Way Is the Mistake.
- Citations to our site: 12
- Pages cited: 1
- Grounding queries shown: 2
- Share of authority: 26.09%
- Citations to other domains in the same query set: 34
- Branded citations: 0
Four limits apply to numbers like these. First, the report’s stated source is “Microsoft Copilot and partners,” so it is not a comprehensive measure of citations across all AI platforms.
Second, grounding queries are the searches an AI system runs to find material before it writes an answer. Microsoft’s Citations Dashboard documentation notes they may differ from what a person typed, so they show how the AI went looking, not who was asking.
Third, share of authority means 12 of the 46 citations in this report’s query set. Microsoft counts a query’s citations on a given day only if the site was cited for that query that day, which can make the share higher than a broader calculation would.
Fourth, twelve citations does not mean twelve people. Microsoft counts how often a page was referenced, not its prominence in the answer.
One more thing is worth checking in any citation report: which topics the cited pages cover. Here the cited article was about punctuation and AI detection rather than a service topic. That describes what was retrieved in this window, not the value of the people reading it.
What does zero tracked AI referrals mean?
It means Clarity recorded no sessions it could attribute to an AI assistant in this window. It does not mean AI sent nobody.
The zero could reflect no click-throughs, visits that arrived without referral information, gaps in tracking or consent coverage, or some combination. This snapshot does not tell us which, and we are not claiming to know.
Clarity’s separate three-day overview showed one “Contact us” smart event session. The event alone does not establish a submitted inquiry or an AI-attributed lead. Business outcomes were not verified in this review, which is different from recording zero outcomes.
How to check the same things on your own site
You can run the same checks on your own site without any special tools beyond your analytics.
Look at what your cited pages are about. If your analytics has a citation report, compare the cited pages with the questions your customers ask before they buy. If it does not, ask a few AI tools those questions and note which sites they cite.
Test whether an AI visit would even show up. When an AI answer links to your site, click the link, then check your analytics for that visit and see how it was labeled. If it lands as “direct” or unknown, real AI visitors may already be hiding in your reports.
Record the platform, browser and time, and identify the visit as test traffic. One successful test checks that setup; it does not validate attribution across every AI platform.
Ask every new inquiry how they found you. Add a “How did you hear about us?” field to your contact form, with options for Google, ChatGPT, Perplexity, Copilot and other. It captures an answer analytics may miss, and it is a separate kind of evidence from a tracked referral.
If your bot activity report shows unsuccessful requests, also check which URLs failed. An unsuccessful response is worth investigating, but the status code alone does not show what content the crawler received.
What can we honestly claim?
Supported: AI-classified systems requested files from our site.
Not supported: Those requests came from prospective customers.
Supported: One article received 12 reported citations.
Not supported: Twelve different people saw or clicked those citations.
Supported: Clarity recorded no AI referral sessions in scope.
Not supported: Nobody found us through AI.
Supported: Some requests were unsuccessful.
Not supported: Those responses caused the missing referrals.
The reusable AI visibility reporting worksheet
Copy this for each reporting period and add a column each time. One rule keeps it honest: record 0 only when the source reports 0, and otherwise write “not available” or “not measured.”
Access
AI bot requests: 1,348
HTML requests: 696
Successful requests: 850
Redirected requests: 337
Unsuccessful requests: 161
Source and scope: Clarity Bot Activity, CDN connection, AI bots filter. All figures are from our September 27 snapshot.
Citations
Reported citations: 12
Source: Clarity Citations, Copilot and partners.
Distinct cited pages: 1
Source: Cited pages table.
Grounding queries shown: 2
Source: Grounding queries table.
Share of authority: 26.09%
Scope: This report’s query set only.
Visits
Tracked AI referral sessions: 0
Source: Clarity AI referral metric.
Engaged AI referral sessions: Not available
Use your analytics tool’s definition.
Outcomes
AI-attributed inquiries: Not verified
Source: Form answers or CRM records.
AI-attributed booked calls: Not verified
Source: Booking records.
AI-attributed customers or revenue: Not verified
Source: CRM or payment records.
Record this context for each source separately, because two reports rarely share the same settings.
- Exact start and end date: ________________
- Reporting timezone: ________________
- Date and time captured: ________________
- Domain or URL filters: ________________
- Which AI platforms the report covers: ________________
- Internal or test traffic excluded?: ________________
- Tracking or consent limitations: ________________
- How the tool defines the metric: ________________
- Screenshot or export saved as: ________________
- Site changes made during the period: ________________
For every business outcome, note how it was attributed: a tracked referral, the customer telling you, or another documented method. “The customer said ChatGPT” and “the analytics said ChatGPT” are different kinds of evidence.
What we are checking next
Our next steps are to test AI referral tracking, investigate the 161 unsuccessful requests and review the attribution evidence for any verified inquiries.
FAQ
Q: Why would AI cite my website but send me no visitors?
A: Citations and visits are measured separately. People may not have clicked, visits may have arrived without referral information, or tracking may have missed them, so test your tracking before assuming which.
Q: Does a high number of AI bot requests mean my site is doing well in AI search?
A: No. A bot request only shows that a machine asked your server for a file. Microsoft’s documentation states that requests do not indicate your content was cited or surfaced in AI answers.
Q: What is a grounding query in Microsoft Clarity?
A: It is the search an AI system runs to find content before writing an answer. Microsoft notes it may differ from the exact words a person typed.
A dashboard total is easy to read and easy to misread. Read each layer on its own terms, and you will know which questions your data can actually answer.
AI Visibility Studio helps websites become easier for AI systems to find, read and cite. aivisibilitystudio.com
Originally published on Medium ↗