GEO & AI Search

One Page, 73K Bing AI Citations and 1 Click: Read the Numbers Before You Trust Them

4 min read Daniel Shashko
One Page, 73K Bing AI Citations and 1 Click: Read the Numbers Before You Trust Them
TL;DR
  • One page: 73K Bing AI citations, 183 search impressions, 1 click.
  • Its queries follow a template with one changing criterion.
  • Weekend citations run at 93% of weekday citations.
  • 4.5% of citations sit on queries unrelated to the product.
  • Citation counts show how often a page is cited, not demand.

One review page of ours has 73K Bing AI citations in three months, 183 ordinary Bing search impressions and 1 click. The queries behind those citations follow a template, run as hard on weekends as on weekdays, and include terms that have nothing to do with the product. The pattern looks automated, and we cannot tell who runs it.

The page is our AthenaHQ review. Bing’s AI Performance report counts how often a page is shown as a source in AI answers, and Microsoft says that reflects how often pages are cited, not page importance, ranking or placement. This post reads one page’s numbers against each other.

The same template runs in two reports

Bing’s Search Performance report lists the queries that showed the page in search results. Its top rows read “evaluate the ai visibility products company athena on generative engine optimization”, “…on best for marketers”, “…on rank higher in”, “…on ai seo”, “…on data accuracy” and “…on llm optimization”. AI Performance lists the grounding queries behind the citations, and they carry the same words: “Athena AI Visibility Products company data accuracy evaluation”, “Athena AI Visibility Products Generative Engine Optimization evaluation”.

80 of the page’s 101 grounding queries start with “athena ai visibility products” and carry 97.7% of its listed citations. 73 of the 101 contain the word “evaluation”. A person rarely writes a search like that. A tool, an agent or a script filling a slot with a new criterion does.

Weekends look like weekdays

The page’s daily citations run from 25 August to 2 November. Weekdays average 1,062 and weekends 989, so an average weekend day carries 93% of an average weekday. After the first day the line stays between 650 and 1,700 for ten weeks with no weekly dip.

The criteria rotate

The slot changes over time. “Data accuracy” is on 11.1% of the last 14 days’ citations against 6.9% across three months. “Model coverage” is on 0.7% against 4.3%, and “prompt volume” on 0.3% against 2.2%. A template that cycles through criteria produces exactly this.

Some queries have nothing to do with the product

22 of the 101 grounding queries, with 3,029 citations (4.5%), carry terms that do not describe AI visibility software. “Athena AI visibility products company evaluation rapid technician onboarding” has 835 citations, “Athena OGM AI Visibility Software company evaluation” has 237, and “Athena AI Visibility Products citation analysis TikTok food & beverage social media” has 131. Bing counted citations against all of them.

Automated tools also search for the brand

Search Performance lists several variants of one long query that joins seven phrases with OR and adds site:reddit.com and date operators, for example “athenahq” or “how do i rank on chatgpt” or “anyone tracking perplexity mentions” and so on. Each variant has 3 impressions and 0 clicks. That reads like a brand-monitoring tool, not a person.

What we can and cannot say

  • We cannot say who runs the queries. The candidates are an AI-visibility tracking tool, an evaluation agent, or someone trying to move the numbers. The reports do not name a source.
  • One click does not prove the citations are fake. Citations and clicks measure different things, and Microsoft does not claim they should match. A page can be cited often and clicked rarely.
  • We do not know whether Bing filters automated prompts. The report does not say. The numbers above are what the report shows.
  • It is one page. The listed queries add up to about 92% of its citations, and the dashboard rounds counts of 1,000 and above.

How to read your own AI citation numbers

  • Put citations next to the page’s impressions and clicks. A ratio of 73K to 183 is a reason to look closer.
  • Read the grounding queries for templates. A fixed stem with one changing slot, quoted phrases, OR chains and operators come from software.
  • Check the weekly rhythm. Open the page chart and compare weekdays with weekends.
  • Look for off-topic terms. Queries about technician onboarding on a software review are noise, and they still count.
  • Report citations as a ceiling. Show them beside real referral sessions, and be wary of any vendor that reports raw citation growth as demand.

For the measurement side, see the GEO KPI measurement framework, best AI visibility tools and best GEO tools. To track Google’s side, see the AI Mode tracking tools list.

Related reading: how Google AI Mode works, robots.txt for AI crawlers, BLUF writing for AI search and best AEO tools.

To see which bots actually visit a site, read AI crawler log file analysis.