Transparency report · measured 2026-08-23

768,234 AI citations in ninety days, and 20 Google clicks.

Both numbers are ours, both are measured, and the second one is why you should believe the first. The industry argues about generative-engine optimisation using vendor blog posts and inference, because almost nobody who has the data publishes it. We are a 14-property network with per-URL citation exports, crawler logs and search console access. This is all of it, including the parts that are unflattering.

That is roughly 8,536 citations a day — about 6 times a minute, an answer engine reaches for one of these pages.

768,234
AI citations · trailing 90 days
14
measured properties
1,001
individually cited URLs
20
Google search clicks
the click gap

Those two numbers are the entire report. 1,001 pages cited by AI assistants 768,234 times, against 20 clicks from Google search on 56,965 impressions — a ratio of roughly 38,412 citations per click.

We call that the click gap: the distance between how much the answer layer uses a corpus and how much the click economy rewards it. A network can be heavily used by one and almost invisible to the other at the same time. Every conventional instrument — Domain Authority, organic traffic, authority scores — reads this network as worthless, and each of them is measuring the side of the gap that is no longer where the reading happens.

§1 · the answer layer

What was measured, and with what

  • Instrument: Bing Webmaster Tools, AI Performance, source "Microsoft Copilots and Partners"
  • Level: PROPERTY (Bing per-property totals, summed)
  • Measured: 2026-08-23 · window UI 3M range, 2026-05-23 to 2026-08-22 (data reports through 2026-08-21 for all 14)
  • Daily average: ~8,536 citations/day
  • Prior reading: 683,677 → +12.4%
  • Arithmetic: the 14 property rows sum to exactly 768,234.
propertycitations · 90dshare
omnirx.org174,61022.7%
grantprobe.com126,41316.5%
nesyona.com103,67613.5%
edubracket.com70,7829.2%
rxgrab.com68,1448.9%
ceocult.com49,9066.5%
bagengine.com44,5385.8%
lenspov.com39,4325.1%
healthbritannica.com33,1524.3%
petmaxxing.com20,1592.6%
rinzara.com18,5412.4%
lucreya.com12,5671.6%
deskdeploy.com4,9680.6%
filmfont.com1,3460.2%
§2 · concentration

Citations are not spread evenly. They pile up.

Of 1,001 cited URLs carrying 737,197 page-level citations, the top ten pages alone hold 18.8%. This is the single most practically useful finding in the report: answer-layer value concentrates at the page, not the domain, so a site-level authority metric cannot see it at all.

2.9%
held by the top 1 page
18.8%
held by the top 10
50.1%
held by the top 50
67.6%
held by the top 100

Fastest movers between the two most recent readings

Pages with fewer than 50 citations in the prior reading are excluded. A page going from 1 to 129 is a “+12,800% increase” and is nothing of the kind — a percentage is a claim about its denominator, and a tiny denominator makes any numerator look like a discovery.

pagebeforeafterchange
bagengine.com/articles/best-shopify-review-apps2851,128+296%
filmfont.com/articles/best-ai-typography-tools-2026134451+237%
rinzara.com/articles/ai-indemnification-matrix62204+229%
nesyona.com/articles/best-ai-meeting-assistants4661,208+159%
bagengine.com/seller-tools/pirate-ship91222+144%

Page-level always runs below property level, because pages under Bing reporting thresholds appear in the property total but not as rows. The two are never compared as growth.

§3 · compounding

Two readings, and the gap between them is the finding

683,677
prior reading
768,234
current reading
+12.4%
citations
+5.4%
cited pages

Citations grew 12.4% between readings while the number of individually cited pages grew 5.4% (9501,001). Growth is arriving faster than the corpus is, which means most of it is landing on pages that were already being cited rather than on new ones.

If that holds, a cited page does not plateau — it gets cited harder, and publishing volume is the wrong lever to pull. We are stating the direction, not a decomposition: the two percentages come from different captures (a network-level total and a per-URL export), so read them as two readings that point the same way, not as an exact split of the gain.

fastest movers · individual pages, both readings
bagengine.com/articles/best-shopify-review-apps2851,128+296%
filmfont.com/articles/best-ai-typography-tools-2026134451+237%
rinzara.com/articles/ai-indemnification-matrix62204+229%
nesyona.com/articles/best-ai-meeting-assistants4661,208+159%
bagengine.com/seller-tools/pirate-ship91222+144%

Movers require a prior base of at least 50 citations, because a page going from 2 to 20 is a 900% rise and means nothing.

§4 · humans

The traffic side, including the number that argues against us

16,305
sessions · 28d trailing
2,640
sessions referred by AI
16.2%
of sessions from AI
20
Google clicks · 2026-05-27 to 2026-08-23

Google Search Console across 14 properties: 56,965 impressions, 20 clicks, 9,746 distinct queries.

We publish this because hiding it would make the citation figure less believable, not more. A publisher claiming enormous AI presence and enormous search traffic is describing an ordinary successful site. A publisher with enormous AI presence and no search traffic is describing something new, and the second claim is only credible with the first number attached.

Caveat, verbatim from the snapshot: JS-based, so ad blockers / privacy browsers / no-JS clients are missed. Treat as a floor.

§5 · the edge

Everything that touched the infrastructure

886,065
HTTP requests · 2026-07-26 to 2026-08-25
142,060
unique visitors
7.83 GB
served
10
zones measured

Read this section narrowly. Unlike every other figure on this page, it is account-wide across 10 zones, not restricted to the 14 editorial properties — the account also hosts unrelated projects, and their traffic is inside this total. It is a measure of infrastructure reach, not of the network the citation figures describe, and it should not be compared with them or summed against them. A narrower, network-only version of this figure is the next correction to this report.

Cloudflare GraphQL Analytics, httpRequests1dGroups. Includes bots, crawlers and assets. Not comparable to GA4 sessions and never summed with them.

§6 · the limits

What we cannot measure, stated plainly

A report that lists only what its author can see is marketing. These are the gaps, and the first one is the most important open question in this field.

limit 1

Brand-level extraction. We can measure that an assistant read a page. No instrument exposes whether it then named the vendors on that page in the answer a human saw.

limit 2

Non-Microsoft engines. ChatGPT, Claude, Gemini and Perplexity publish no citation counts. Every citation figure here is a floor across one engine family, not a total.

limit 3

Downstream conversion. With near-zero search clicks and citations that fire no click event, attribution from citation to revenue is inferential.

limit 4

Crawler intent. Cloudflare records the fetch, not which assistant later used the content.

§7 · method

How to check us

Every figure here is generated from dated JSON snapshots by a script, not typed by a person, and the generator refuses to publish if the per-property rows do not sum exactly to the headline. The data was generated on 2026-08-25 from telemetry measured on 2026-08-23, and it is baked into this page at build time — so if those dates are old, the numbers are old, and you should read them as of then rather than as of today.

Citation counts come from Bing Webmaster Tools because Microsoft is the only operator that publishes them. ChatGPT, Claude, Gemini and Perplexity publish nothing comparable, which means every citation number here is a floor across one engine family and not a total. If any of those operators ever publish counts, this report will carry them.

This report is re-generated from a fresh capture roughly every two weeks, and each edition keeps the prior reading beside the current one so the direction is checkable rather than asserted. The section above is the second edition to carry both.

what these numbers are attached to

The pages behind the figures are the inventory.

Every citation counted above landed on a specific page. Those pages are listed individually, with their own citation counts, their movement, and whether a placement slot on them is open. You can search that shelf for your own company and see exactly which pages already name you.

The AI Consensus Index is the other half of this work: what the assistants actually say, rather than how often they say it.