Run a search for law firms in a mid-sized city and you might get an AI Overview naming four or five of them. Useful, on the face of it. Now look at where the summary sourced those names. On August 12, 2026, local search consultant Joy Hawkins documented a case where the answer cited exactly one source for every firm listed: a listicle written and published by one of the lawyers appearing in it. None of the named firms were linked. The listicle was.
She reproduced it across browsers, which rules out a personalization artifact. Barry Schwartz, covering it at Search Engine Roundtable, was blunt about the result: this is not quality, this is not making Google's AI results look helpful, it is just a shame.
He is right, and the more useful question is why it happened, because the answer is not that Google's systems malfunctioned. They worked exactly as designed against the content that was available to them.
What Joy Hawkins found on August 12
The mechanics are worth walking through slowly, because the shape of this failure repeats across every category with fragmented local supply.
A user searches for law firms in an area. The query has clear comparative intent: the person wants options, not one destination. An AI Overview appears and produces a shortlist. To produce a shortlist, the system needs a document that discusses several of these firms together, in one place, with something resembling evaluative language. Individual firm websites cannot serve that need. Each one discusses exactly one firm and says it is excellent.
So the retrieval step goes looking for a document that compares them. In this market, one existed. It was written by a practising lawyer, it featured that lawyer's own firm, and it listed competitors around it. That document was structurally perfect for the job and substantively worthless, and the answer could not tell the difference.
| WHAT THE ANSWER NEEDED | WHAT BUSINESS SITES PROVIDE | WHAT THE LISTICLE PROVIDED |
|---|---|---|
| Several named businesses in one document | One business per site | All of them, in one page |
| Comparative or evaluative framing | Self-description only | Rankings, however invented |
| Extractable structure, list form | Prose and service pages | Numbered headings, one entity per section |
| An apparently neutral voice | Explicit first-party marketing | Implied third-party review |
| Local specificity | Present, but single-entity | Present, across the whole set |
| Verifiable independence | Not claimed | Not present, and not checked |
Five of those six rows favour the listicle. The sixth is the only one that should have mattered and is the hardest for an automated system to evaluate. That asymmetry is the whole story, and it is the same dynamic behind the finding we published in the anatomy of an AI citation, where comparison content took 32.5 percent of tracked citations, more than any other format. Comparison content wins because comparison is what the buyer asked for.
Why AI Overview local results reach for a listicle
There is a supply problem underneath this that nobody in local search has solved. In most categories and most cities, genuinely independent comparative content about local businesses does not exist. Directories list without evaluating. Review platforms aggregate scores without narrative. Local press stopped covering business categories years ago in most markets. What remains is content produced by participants, and participants have an interest.
Illustrative view of the comparative-content supply available for a typical mid-market local category, expressed as share of retrievable documents that discuss three or more named local competitors together. Illustrative scenario for framing the supply gap, not measured data.
Where independent comparison is scarce, the participant-authored version is not competing against better content. It is competing against nothing. Any retrieval system optimizing for relevance to a comparative query will find it and use it, and the more structurally clean it is, the more confidently it gets quoted.
This also explains why the problem clusters in professional services, home services and healthcare rather than in retail or software. In categories with real third-party review infrastructure, there is an alternative document to cite. In categories without it, there is one candidate and it belongs to a competitor.
The listicle pattern Google says it protects against
Google is not unaware of this. Search Engine Land reported in April 2026 on the surge of best-of and top-ten pages produced specifically as a generative engine tactic, and the pattern it described is precise: pages that rank competitors the author has never tested, that use subjective or invented scoring systems, that place the publisher's own product first, and that imply firsthand evaluation which did not happen. Search Engine Land's write-up of the trend is the clearest published description of the tactic, and every element of it appears in the page Hawkins found.
A Google spokesperson told The Verge that the company applies protections against manipulation in Search and Gemini, and advised creating content for people while making sure it is understandable to search systems. Both halves of that are reasonable and neither of them describes a system that currently catches the case Hawkins documented.
“A page that ranks eleven competitors the author has never contacted, using a scoring system the author invented, with the author's own firm at number one, is not a comparison. It is an advertisement with a table of contents.”
The gap between stated policy and observed behaviour here is not hypocrisy, it is latency. Manipulation protections operate at scale on signals that generalize. A single self-promotional listicle in a mid-sized legal market generates almost no signal: modest traffic, few links, no spam complaints, no pattern to match against. It is beneath the resolution of the enforcement systems and squarely inside the resolution of the retrieval systems. That asymmetry will persist for as long as retrieval is cheap and verification is expensive.
What this means if you are one of the named businesses
Start with the uncomfortable framing. If a competitor's listicle is the source an engine uses to describe your category, that competitor is writing your positioning, choosing your peer set, and deciding what order you appear in. They are also collecting the click. You are a line item in someone else's marketing collateral, cited by Google, at no cost to them.
The instinct is to file a complaint or ask Google to remove it. That is worth doing when the content is factually wrong about you, and it is not a strategy, because the supply gap that made the listicle valuable will still be there afterward. If that page disappears, retrieval finds the next-best comparative document, which in most markets is a worse one.
| RESPONSE | WHAT IT ACHIEVES | WHAT IT DOES NOT |
|---|---|---|
| Report factual errors about your business | Corrects specific harm, sometimes quickly | Nothing about who owns the comparison |
| Ask to be removed from the listicle | Removes you from a competitor's page | Removes you from the shortlist the answer builds |
| Publish your own competitor listicle | Creates a rival document, fast | Credibility, and it repeats the problem you object to |
| Publish genuine comparative content with a real method | A defensible document with a chance of becoming the cited one | Immediate results, this takes a quarter or more |
| Build out review platform presence | Adds a structured, independent record engines can reach | A narrative document to quote from |
| Do nothing | Nothing | Anything |
The third row deserves a warning. Publishing your own version of the same self-serving page is the most common response and the worst one available, because it works briefly, invites the same criticism you are currently making, and increases the total supply of untrustworthy comparison content in your category. It also ages badly: the enforcement resolution problem described above is a latency issue, not a permanent condition.
The content that displaces a bad listicle
The replacement has to do the same structural job while being defensible. That means a document that names several real competitors, states a method, and reaches conclusions a reader could check. Most companies refuse to write it because it means naming competitors in public, which is a positioning decision rather than a content decision, and it is usually made by someone who has never seen what the current alternative looks like.
The build is not complicated. State what you compared and on what criteria. State how you gathered the information, including where you relied on public sources rather than direct testing. Name the cases where a competitor is the better choice, because a comparison with no such case reads as marketing to a human and correlates with nothing useful to a machine. Keep one entity per section with a consistent structure, so extraction is clean. Date it and update it, because recency is one of the few quality proxies retrieval can actually assess. We set out the section-by-section version of this build in how to construct a comparison page that earns the named mention.
For multi-location brands the same logic applies at every market you operate in, which makes it a content operations problem rather than a one-page project. That is the shape of most of the work we do in local and multi-location engagements, and the pattern holds: the markets where a competitor owns the comparison are the markets where the shortlist forms without you.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.