Why AI Engines Keep Reaching for Listicles

Ask an assistant for the best waterproof hiking boots under $150 and watch where the answer comes from. Rarely the boot brands. Rarely the retailers. Usually a blog post titled something like "9 Best Waterproof Hiking Boots (2026)" from a two-person outdoor site nobody at your company has heard of.

This is measured now, by several groups with no shared incentive. What they disagree about is how big the effect is, which turns out to be the useful part.

Three studies, one direction, three magnitudes

In December 2025 Ahrefs published research by Glen Allsopp covering 26,283 source URLs that ChatGPT cited across 750 top-of-funnel prompts in software, products and agencies. "Best X" blog lists were 43.8% of all page types cited. Not the largest category by a nose. The largest by a distance.

In May 2026 Evertune ran a wider sweep, covered by Search Engine Land: the 6,000 most-cited URLs per model across ChatGPT, Copilot, Gemini, Google AI Mode, AI Overviews and Perplexity for March and April 2026. About 25,000 unique URLs drawn from close to 400 million citations. Their number was 63%, and of those listicles, 71% to 86% were ranked and numbered rather than unordered.

Then in July 2026 DeltaV Digital published a study of 25,337 citations across five engines and eight brands and put listicles at 19.6%, behind articles at 23.7%.

Twenty, forty-four, sixty-three. Same phenomenon, three answers, nobody lying.

The spread is sample frame and category mix. Ahrefs prompted commercial queries in software, products and agencies. Evertune measured most-cited URLs, which is not the same population as all citations. DeltaV's eight brands include a healthcare nonprofit, a government scholarship program and higher education, categories where nobody publishes "Top 10 Federal Grant Programs" and the engines cite institutional pages instead.

The study that actually resolves it

The smallest one is the most useful. AIVO Research ran a pre-registered test in June 2026: 138 ChatGPT queries built from 46 real conversational prompts, three runs each, split by search intent.

Roundup dominance turned out to be intent-specific, not universal. When someone asks casually for "the best X," roundups appeared 100% of the time. Local queries, 71%. Transactional, 50%. Informational, 13%. Navigational, zero.

So the honest version of the claim is not "AI loves listicles." It is narrower and more actionable: when a shopper asks an assistant to recommend a product, the assistant reaches for a roundup nearly every time. Which happens to be the one query that decides whether anyone buys anything.

The mechanical reason

Think about what the model has to produce. Someone asks for the best insulated bottle under $40 for a bike commute. The answer needs three to six named products, a reason attached to each, and enough attribute detail to justify the order they appear in.

A product page offers one candidate and a sales pitch. A category page offers forty candidates and no opinion. A roundup offers a pre-ranked shortlist with reasons already written, in the exact shape the answer has to take.

The roundup already did the comparison. The model is not synthesizing a recommendation, it is compressing one.

That is why ranked lists beat unranked ones in the Evertune data, and why AIVO found the median cited list ran to ten items with 92% carrying the current year in the title. Ten is enough coverage to look thorough without going mushy. The year is a freshness check that costs nothing and is trivially verifiable.

It also explains the thing merchants find annoying: your product page can be immaculate and still lose, because it is answering a different question than the one being asked.

Publish the roundups for your own commercial clusters

Getting into other people's lists is a public relations program that runs on public relations timelines. Worth doing, slow, largely outside your control.

Publishing your own is neither slow nor outside your control, and most stores skip it because it feels like admitting competitors exist.

If you sell trail running shoes, the buying guides in your category are your job. "Best trail running shoes for wide feet." "Best road-to-trail shoes." "Best trail shoes for winter." The specific sentences your customers say out loud. One per commercial cluster, meaning one per group of products a shopper would actually cross-shop.

What the research says these pages need, drawing on Seer Interactive's February 2026 analysis of over 2 million citations plus the AIVO findings:

• Ten options at minimum. Seer found the surviving listicles ran to ten or twenty rather than five or eight, and AIVO's median cited list was exactly ten. Short lists read as promotional, because usually they are.
• The list itself near the top. Not after 800 words on the history of trail running.
• A dated page and the year in the title, updated when you say it is updated.
• A stated methodology. How you tested, what you compared on, what you did not evaluate.
• Consistent structure per entry. Same fields, same order, every time. This is the part that makes a page trivially extractable and it is the part most brands get bored of by item four.
• External validation where you have it: third-party ratings, review counts, published test results.

And include competitors. Genuinely. A list of ten products where all ten happen to be yours is a category page in a costume.

Ahrefs did find self-promotional lists getting cited, appearing in over a third of software-category responses in its sample, so the tactic has demonstrably worked. Search Engine Land's coverage of the Evertune study flagged the two brakes on it: Google has signaled it intends to penalize spammy self-promotional listicles, and the FTC has rules about misrepresenting independent reviews when the business controls the reviewer. Neither of those is a problem you want to discover retroactively across forty pages.

Rank honestly. Say when a competitor is the better pick for a use case you serve badly. You will lose a handful of sales and gain a page a model is willing to cite, which is the trade you are making.

Structured data helps, because it is what a crawler reads when a shopper arrives from the answer instead of the feed. If that is not set up, adding schema markup to Shopify is a no-code afternoon.

Where this stops working

The share is already moving, and not upward everywhere.

Seer tracked more than 2 million citations on a consistent prompt set from November 2025 through February 2026 and found ChatGPT's citations to "best" and "top" URLs fell roughly 30% between December and January, with listicle share of all ChatGPT citations sliding from 17.2% to 15.5%. Thirteen of sixteen industries declined. On the identical prompts, Google AI Overviews went the other way, flat or rising.

Two engines, same queries, opposite directions, inside eight weeks. That is not a settled system, and it is the whole answer to anyone selling you a listicle strategy as a durable advantage.

The other limit is more basic. A roundup cannot rescue a product that is structurally invisible. If your attributes are incomplete, your category is wrong, or an assistant cannot verify your price and availability, a mention in a list gets you a name-check and nothing after it. Format wins the citation. Data wins the sale. The wider framing on how those two fit together is in what generative engine optimization actually is.

Do it anyway, because the downside is nil. A well-researched buying guide with honest rankings was good content marketing in 2015, ranked in Google in 2020, and happens to be the shape assistants prefer in 2026. Not many tactics survive three regime changes, and the ones that do are usually just competent work with a new label on it.

If you want to know which of your commercial pages an assistant can currently parse and which it skips, Seokai's AI Readiness score will tell you where to start: check your store.