Your own listicle recommends your rivals

A self-promotional listicle is a "best [category] tool" page published by one of the vendors in the category, and in AI Overviews it now tends to get the publisher cited without getting the publisher recommended.

Those are two different outcomes, and the gap between them is the whole story. A citation puts your URL in the source strip. A recommendation puts your product name in the sentence a buyer reads. Teams have been reporting the first and assuming it delivers the second.

#The measurement

Lily Ray ran 100 B2B "best [category] software" queries through Google AI Overviews on three dates in 2026: April 15, May 15, and June 8 (lilyraynyc.substack.com). Of those, 80 prompts returned an AI Overview. The design separates two metrics that most reporting collapses into one: which sources the answer cites, and which brands the answer actually names as picks.

Across that sample, self-promotional listicles were cited 323 times. In 224 of those, Google used the brand's own page as a source and then recommended someone else, a rate of 69% (Search Engine Land). Measured per prompt instead of per citation, 74% of AI Overview responses cited a self-promoter without recommending it (lilyraynyc.substack.com). The dataset covered 184 self-promoting listicle pages across 146 brands.

The concrete shape of the failure is worth stating plainly. Pylon ranked itself first for help desk software; the AI Overview cited Pylon and recommended Zendesk, Freshdesk, and Help Scout (almcorp.com). You paid for the page, and the page argued your competitor's case.

#Why citation and recommendation came apart

The mechanism is not mysterious. A listicle that covers a category is topically excellent retrieval material: it names the vendors, describes the use cases, and answers the query's shape almost perfectly. So it gets retrieved and cited. The ranking inside it, though, is authored by an interested party, and the answer engine has other evidence about who is actually well regarded. The citation survives; the self-ranking does not.

What fills the recommendation slot instead is third-party and user-generated content. Forbes, Reddit, and YouTube were the most-cited domains in AI Overview responses containing "best" (Search Engine Land). That matches what we have seen in the reverse direction: the pages an engine is willing to quote are not automatically the entities it is willing to endorse.

There is a second cost. Ray reports that visibility declines began around January 20 across dozens of sites leaning on self-promotional listicles, and continued through Google's May 2026 core update (Search Engine Land). Whatever the AI-answer effect, the classic organic channel appears to have repriced this content type too.

#The tactic still generates citations, which is why it persists

If you only look at citation counts, the play still looks like it works. Peec AI tracked 232,000 citations from 13,000 listicles over 12 weeks between December 2025 and February 2026 in the software and software-review niche, and found roughly 11% of citations coming from self-promotional listicles (peec.ai).

The per-engine spread is large enough to matter for planning (peec.ai):

Engine	Self-promo share of citations
ChatGPT	3.6%
Google AI Mode	10.3%
Perplexity	10.4%

Peec found no systemic correction inside the observation window: rates held roughly flat across the 12 weeks rather than trending down (peec.ai). Their own caveat is the important one, and it applies to every number on this page. The findings are vertical-specific, and software reviews is a category with unusually heavy self-promotional supply.

A third dataset points the other way on volume. Seer Interactive analyzed over 2 million citations from November 2025 to February 2026 across ChatGPT and AI Overviews, classifying URLs containing "best" or "top" as listicles, and found ChatGPT listicle citations falling about 30% from December to January, from roughly 160,000 to 111,000 (seerinteractive.com). Listicle share of citations moved from 17.2% to 15.5% in the same step.

Read that decline carefully before acting on it. Total citations per response fell 23% over the same period, from 11.96 to 9.18, so part of the listicle drop is the answer format getting shorter rather than listicles getting deselected (seerinteractive.com). The comparison is also a single month-over-month step, and ChatGPT and AI Overviews moved in opposite directions on identical prompts. Thirteen of sixteen industries declined; three rose.

None of these three studies contradict each other once you line up what each one counts. Peec counts citation share and finds the tactic still gets picked up. Seer counts citation volume and finds the format shrinking on one engine. Ray counts recommendations and finds the citation does not convert. Different denominators, different questions.

#What to change in reporting

The practical fix is a reporting change before it is a content change. If your AI-visibility dashboard has one column for "we appeared," split it.

• Cited. Your domain shows up in the source list. Useful as a retrieval signal, weak as a demand signal.
• Recommended. Your product name appears in the answer's actual list of picks. This is the column tied to pipeline.
• Cited but not recommended. The gap. On category and "best" queries, track this as its own number, because it is where a self-promotional page quietly funds a competitor.

We track those as separate fields for exactly this reason: an answer can name your page as a source and name a rival as the pick, and a single appearance counter cannot tell those apart.

#What to publish instead

Comparison content is not the problem. A comparison page that a buyer can trust still earns retrieval, and the query shape is genuinely high-intent. The problem is the self-awarded first place, which is the one element in the page an answer engine has independent evidence to overrule.

Two adjustments follow. First, write the category page so it is accurate about where you lose. A page that says you are the wrong choice for a named use case is more citable and is not making a claim the engine will contradict. Second, put the effort that would have gone into ranking yourself into being ranked by someone else. The recommendation slot is filled by third-party and UGC sources, so presence in review sites, analyst coverage, and the communities in your category is the thing that moves it.

The short version: stop measuring whether the answer mentioned your URL, and start measuring whether it recommended your product. In this sample, those two numbers disagreed 69% of the time (Search Engine Land).