What one grader rejected, another cited more than 370,000 times.
Google AdSense classified five of our properties as “Low value content”. Over 90 days to 23 August 2026, those same five sites were cited more than 370,000 times by Microsoft Copilot and its partners, across 4,168 distinct grounding queries. Same URLs, same period, opposite verdicts.
We are not appealing and we are not claiming anyone is wrong. Both graders are measuring what they exist to measure. The finding worth publishing is that those two things have come apart far enough to be measured, and we can measure it because we are on both sides of it.
Each carries the AdSense status “Needs attention: Low value content”, read from the account console on 2026-08-26. Citation counts and share are from Bing Webmaster Tools AI Performance on the same day.
strongest measured query: “business grants for starting a business”
strongest measured query: “paid vs free AI tools for studying effectiveness”
strongest measured query: “lowest prices common prescriptions pharmacies”
strongest measured query: “tax software self-employment income”
strongest measured query: “canine genetic testing”
Display advertising has spent twenty years being gamed by thin affiliate pages built to collect impressions, so the review is tuned to recognise that shape: comparison articles, outbound commercial links, templated structure across many pages. Our properties are affiliate-monetised comparison sites. They have the shape the screen is built to catch, and it caught them. That is the screen working, not failing.
A retrieval system is asking something narrower: does this page contain a specific, well-structured answer to the question in front of it, in a form that can be lifted into a response and attributed. Comparison structure, explicit tables, named entities and direct answers all help there. The same properties that read as templated to one grader read as structured to the other.
Ad-network approval has been the default proxy for whether a site is any good, because for most of the web’s commercial history it was the only grader that paid. It is now possible for a site to fail that test and still be one of the most-cited sources in its category, and a publisher optimising only for the first signal cannot see the second. The two are worth measuring separately because they have stopped agreeing.
Nine other properties were not rejected, they were never reviewed. 9 of our 14sites show the status “Getting ready”, which means the review has not run. It would be wrong to describe the network as having been rejected, and we are not describing it that way.
The account was never approved, so nothing was lost. No advertisement has ever served on any of these properties. This is not a publisher who was earning and got cut off, and it would be dishonest to let it read that way.
The citation figures are a sample, not a census. The Bing Webmaster Tools console states that the data it reports “represents a sample of overall activity”. We have not established whether that qualifier applies to the headline totals or only to the query breakdown, so we treat the whole figure as sample-qualified.
The totals are rounded because the inputs are. Bing reports each property’s count to three significant figures, so every number here is known to roughly fifty either way and the total to roughly two hundred and fifty. We round down and say “more than” rather than print a sum that would imply precision the source does not provide.
Citation share is a floor.Bing lists the top 25 grounding queries by citation count, not by share, so the share figures above are each property’s best among its most-cited queries. A higher share on a less-cited query would not appear.
One engine family, one network. These are Microsoft Copilot and partner citations for fourteen sites owned by one operator. It is a worked example, not a study, and it should not be generalised into a claim about publishing at large.
AdSense statuses were read from the signed-in publisher console on 2026-08-26. Citation counts, grounding query counts and citation share were read the same day from Bing Webmaster Tools AI Performance, covering 90 days to 23 August 2026, with Microsoft Copilots and Partners named as the citation source. Nothing here is modelled or estimated.
The per-property citation figures behind this page are the same ones published on the placement inventory, and the measurement methodology this network uses is described in the AI Consensus Index.
Written by Vincent Couey for Deep Synthesis.