Do comparison pages and listicles help or hurt AI visibility, measured on 507 of our own

Most of this category answers this question by inference. We answered it by accident, at scale, against ourselves. Between 20 and 22 October 2026 we took our stock of recommendation shaped pages from 134 to 641, which is 507 new pages in two days, and almost all of them are the formats this question asks about: best of lists, head to head comparisons, alternatives pages and segment shortlists.

Then we re-ran the buyer questions. On 20 October our domain was a source in 6 of 1,260 readings and our name appeared in 0. On 22 October it was 0 of 635 and 0. Our domain is in none of the 19,793 citation URLs the second run recorded and none of the 3,307 hosts behind them. The method, the reused questions and the detector tests are in the 507 page null result.

What that does and does not establish

It establishes one thing cleanly. Publishing 507 pages in these formats, on a domain nobody was citing, did not move either number in two days on two engines. The format did not carry the pages in.

It does not establish a difference between a comparison page and a listicle. The specification an engine wrote for this question asked for exactly that: the difference in brand inclusion rate between the two formats, in percentage points, with the prompt count and the repeated runs disclosed. We do not hold the format split, so that figure is not on this page and we are not estimating it. The other specification offered a 69 per cent citation to recommendation leakage rate for self published listicles. We did not measure that, it arrived with no underlying data, and we are not repeating it as a fact.

Where the citations went instead

In the same run, 19,793 citation URLs landed across 3,307 distinct hosts, and half of every citation came from 51 of them. The leading hosts were developers.google.com with 1,360, semrush.com with 744, support.google.com with 695, help.openai.com with 467 and blog.google with 394. Review and comparison directories were a thin slice, with g2.com at 125. The full host table and the source mix are in which websites the two engines cite.

The uncomfortable detail is that vendor domains do get cited. In that same list, ahrefs.com took 354, peec.ai 328, otterly.ai 255 and tryprofound.com 227. So the rule is not that a vendor site cannot be a source. Those are the domains the answers were already naming: Semrush appeared in 259 of 635 readings, Profound in 154, Ahrefs in 145, Otterly in 124 and Peec in 91, out of 1,053 namings across a 24 name roster. Publishing in this format did not get us into that set, and the set and the citation list overlap heavily. The naming table is in which brands the assistants put forward.

The risk a self ranked list carries, stated without a number

A page of ours that ranks vendors contains competitor names, their claims and their limits, written by us. If an engine uses that page as material for a question about the category, the names in it are available to the answer. We cannot tell you how often that happens because we did not measure it, and the 69 per cent figure in circulation is an engine's suggestion rather than anybody's observation.

What we can say is what our own pages carry, because it is a deliberate choice. Every page of ours that names us also names where a rival beats us. On 22 October 2026 we fetched tryprofound.com/pricing ourselves: it publishes SSO and SAML with SOC 2 compliance and up to 9 answer engines tracked on Enterprise, against the five surfaces we read, and it publishes no price figure in any currency anywhere on that page. We also fetched brightlocal.com/pricing the same day, where the plan card reads 31 dollars a month on the annual option with 369 billed annually, while the per location table in the same document reads 41 dollars a month billed monthly for one location. Both of those render, so we publish both rather than the cheaper one.

Writing a comparison that way costs the page its sales punch and is the only version worth publishing, because a list with no losing row in it was either written after the answers were seen or covers too little to mean anything.

What to do with the format, concretely

1. Stop counting pages as progress. Our own count went up 378 per cent in two days and both outcome numbers went to zero. Count readings, not pages.
2. Put one checkable fact in the page that only you hold. Of the 224 page specifications the engines wrote, 212 named a single most citable number they wanted the page to carry, and most wrote it out as a fraction with its denominator. The counting is in what makes an assistant cite a vendor website.
3. Date every vendor fact and attach it to the URL you read it on. Not to the brand and not to the domain. Two URLs on one site routinely disagree, and a claim travelling under a bare brand name cannot be checked by anybody.
4. Quote limits rather than reproducing a rival price table. A limit is stable enough to be worth something and a price table is wrong within a quarter.
5. Publish the losing row. It is the cheapest available evidence that the set was not curated.

What this page cannot tell you

Two engines, ChatGPT and Gemini, over two days in October 2026, on a domain starting from almost no citations. No format split, no control group and no third party comparison. A category with more established citation coverage could behave differently and nothing here says which way. We sell the remedy, which is the reason this page leads with the number that does us the most damage.

Questions about comparison pages and AI answers

Should I stop publishing comparison pages

Not necessarily, but stop buying them as a visibility mechanism. On our evidence they are worth publishing when they hold a fact a reader cannot get elsewhere, and worth nothing when they are the hundredth page of opinion.

Is a third party list better than my own

Our data cannot separate the two. What it shows is that the directories were a thin slice of 19,793 citations, with g2.com at 125, while documentation hosts led the table.

Does the format leak my competitors into the answer

A page of yours that names rivals makes those names available as material. We did not measure how often an answer picks them up and we are not putting a percentage on it.

What moved the needle for the vendors who are cited

We cannot say from a host count. What the count shows is that the cited vendor domains are the same ones the answers were already naming, which is a different problem from a format problem.

If the format is not the lever, the next question is where the citations did land. The host table and the places a page can be put instead are in what generative engine optimisation work actually consists of, from our own run, and the surfaces any of it can be measured on are in which AI surfaces can actually be measured through an API, and which cannot.