Fourteen Vendors Publish Comparison Pages. We Checked Whether They Rank.

Almost every vendor in AI visibility publishes “us versus them” pages. They are cheap to generate, they name a competitor with real search demand, and they look like free traffic. We swept seventeen vendors’ sitemaps and then asked the rank tracker about seven of those pages one at a time. The four that name their own publisher rank for nothing at all.

This is not an argument against comparison pages. Two of the three we tested that compare two other companies do rank, one of them in eighth place for a term with 3,600 monthly searches. The finding is about which half of the pattern earns anything, and the honest answer includes a confound we cannot remove with this data.

Disclosure: EchoWi sells AI visibility measurement, and several vendors named below are competitors, so a study finding that a popular tactic mostly fails is a study with an interest. Everything needed to repeat it is here: the seventeen domains, the seven page URLs, the market is the United States in English, the date is 11 August 2026, and every figure has a row in our measurement register. Run it against us and publish a different answer.

The short version

1. Fourteen of seventeen vendors we swept publish product comparison pages. The largest section is 29 pages. Most sit between two and fifteen.
2. Zero of the four self comparisons we tested rank for a single keyword. Not one, across four different publishers.
3. Two of the three third-party comparisons rank, the best at position 8 for 3,600 monthly searches, on a page whose publisher is neither of the two products being compared.
4. The two explanations cannot be separated here. A page naming a small vendor is both self-serving and aimed at a phrase nobody searches. This sample confounds them perfectly, and we say so rather than pick the flattering one.
5. One section ranks with no discoverable path at all. It is absent from its own sitemap, and headless Chrome finds no link to it from anywhere on the site, including its own hub.

How many of these pages exist

We read the sitemap of every vendor in our catalogue that publishes one, followed one level of sitemap index, and classified the URLs. The classification matters more than the count, because our first pass got it wrong.

A URL like /blog/geo-vs-seo matches any pattern written for comparison pages, and it is not one. It is an ordinary post about two ideas, which every vendor in this category writes. So is chatgpt-vs-gemini. Counting those together with siftly-vs-semrush produced a number that measured nothing: it said AirOps had 93 comparison pages when it has 29 and 64 posts about concepts and engines.

Fourteen of the seventeen swept publish at least one. This is not a niche tactic in this category, it is the default.

Whether they rank, one page at a time

Asking a rank tracker for a domain does not answer this. AirOps ranks for 877 keywords, and the hundred largest by volume stop at 320 searches a month, which is far above anything a page called airops-vs-adobe-llm-optimizer could target. A domain query would report zero comparison pages ranking and the report would be an artefact of the sort order.

So each page was queried on its own.

Before believing any of those zeros, we ran a control, because a query that fails and a page that ranks for nothing produce the same empty result. The same page-level query was pointed at a page we already knew ranked, and it returned that page’s keywords. The instrument works, so the zeros are zeros.

The Scrunch row is stronger than a page-level query, and worth separating out. That domain ranks for 156 keywords in total, we read all 156 rather than a sample, and not one of them sits on a page naming Scrunch on either side of a “vs”.

The part we cannot resolve, and will not pretend to

The obvious reading is that Google discounts a comparison written by one of the participants. It is a tidy story, it fits all seven rows, and we are not going to publish it, because a second explanation fits equally well.

Nobody searches for these phrases. “Semrush vs Ahrefs” has 260 monthly searches. “AirOps vs Adobe LLM Optimizer” has effectively none, because AirOps is not a brand people type into comparison queries yet. Every self comparison in this sample names a small vendor on one side, and every third-party comparison names at least one large one. Self-serving and low-demand are perfectly confounded here, and no amount of staring at these seven rows separates them.

The version that survives is weaker to write and more useful to act on. A comparison page can only capture demand that already exists for the pair it names. Generating a page for every competitor produces a page for every pair, and most pairs have no demand at all. The page is not failing to rank; there is nothing there to rank for.

That is testable, and the test is not ours to run cheaply: it needs a vendor with real comparison demand for its own name publishing a self comparison. If one exists and it ranks, the self-serving explanation dies and only the demand one is left.

The section that ranks with no way in

One case in this sweep breaks the rules badly enough to be worth its own section.

Scrunch’s comparison section lives at /aeo-tools/. Three things are true about it at once:

• It is absent from the sitemap. That sitemap has 179 URLs and not one of them is under /aeo-tools/.
• Nothing links to it. Loaded in headless Chrome with JavaScript running, the hub page exposes exactly one /aeo-tools/ link, which is the hub itself. A comparison page exposes exactly one, which is itself. It does not even link back to the hub. Every page is an island in both directions.
• Four of its pages rank anyway, and one of them holds position 8 for semrush ai visibility toolkit, 3,600 monthly searches, against a keyword difficulty of 52.

We checked the links with a real browser rather than with curl on purpose. Fetching the served HTML shows no links either, and stopping there would have proved only that the markup is client-rendered. It is not: the links are absent after rendering too.

We do not know how Google found these pages, and we are not going to guess in print. What the case does establish is narrower and still useful: the two discovery mechanisms every technical audit checks first are not necessary conditions. A section can be missing from the sitemap, orphaned from the internal link graph in both directions, and still hold a first-page position on a competitor’s product term.

And on the surface these vendors actually sell

Every company in this sweep sells measurement of AI answers, so the second test writes itself. We asked the question their category hub targets, on Google’s AI Overview and AI Mode, three runs each, cache bypassed.

“What is the best AEO tool?” Six answered runs. The vendor whose hub targets that phrase, and ranks 17th for it organically, was named zero times and cited zero times.

The domains cited in every single run were G2, YouTube, HubSpot, higoodie.com and geoptie.com. Two of those five are vendors in this category with ordinary content, not comparison pages. This is the same shape we found measuring who holds the durable citation slots across eleven categories: the slot goes to an aggregator or an intermediary far more often than to a seller.

So the tactic that partly works in one channel does nothing in the other, for the same page, on the same day.

What to do with this if you were about to build them

Check the demand for the pair, not for the tactic. Before generating a page for every competitor, look up the search volume for “you vs them”. If it is zero, the page cannot earn traffic no matter how good it is, and you are about to generate a hundred of those.

Point them at the competitor’s product names, not at yours. Every ranking page in this sample earns its position from a term belonging to a company with demand. semrush ai visibility toolkit is a product name specific enough to be reachable and popular enough to be worth reaching. Your own name is neither, yet.

Do not assume the pages are discoverable. Scrunch’s section ranks despite having no sitemap entry and no inbound internal link. That is a finding, not a recommendation: it worked there, and the base rate for orphaned sections is not something one case establishes. Put them in the sitemap and link them.

And do not expect them to do anything for AI citations. They did nothing here. If citation is what you are buying, the evidence points at being the subject of a comparison somebody else publishes rather than at publishing your own.

Limits

Seven pages. That is enough to notice that four of four self comparisons rank for nothing and not enough to estimate a rate. Treat the direction as the finding and the proportion as an illustration.

One market, one language. United States, English. Comparison demand differs by market and this says nothing about Spain, France or Germany.

One day. 11 August 2026. Rankings move and sitemaps get regenerated.

Sitemap counts are floors. One vendor’s entire section is missing from its own sitemap, so a count of zero there means “not declared”, not “does not exist”. Any sweep like this undercounts by an unknown amount.

Nothing here is causal. No page was changed and nothing was measured before and after. This is what the pages rank for, not evidence that a page’s shape caused it.

Common Questions About Comparison Pages

Do “X vs Y” pages still work for SEO?

They work when the pair already has search demand and fail quietly when it does not. In this sample, every page that ranked named at least one product people search for by name, and every page that named a small vendor on one side ranked for nothing. The tactic is not the variable, the demand is.

Should the comparison include my own product?

The four self comparisons tested here rank for nothing, but so does one of the three third-party pages, and the sample cannot separate self-serving from low-demand. What is safe to say is that naming yourself adds no search demand, so the page still has to earn its position from the other name on it.

Why would a page rank if it is not in the sitemap?

We do not know, and anyone who tells you confidently is guessing. What the case shows is that a sitemap entry and an internal link are not necessary conditions, which is different from saying they do not help. Both are cheap, so declare and link the pages anyway.

Do comparison pages get cited by AI assistants?

Not in this measurement. Asking the category’s own buying question across two Google surfaces, six answered runs, the vendor running the largest comparison hub in that category was neither named nor cited, while G2, YouTube and HubSpot were cited in every run.

How do I check my own comparison pages?

Query them one URL at a time rather than by domain, because a domain report sorted by volume cannot see a page targeting twenty searches a month, and will show you a zero that means nothing. Then check that the zero is real by running the same query against a page you know ranks.

Is this an argument for writing fewer pages?

It is an argument for checking demand before generating a set. A hundred pages for a hundred pairs with no demand is a hundred pages, some crawl budget, and no traffic. The same effort spent on the handful of pairs people actually search would produce the two results we did find.

Where this leaves you

The tactic is close to universal in this category and its results are not. Fourteen vendors publish these pages, the biggest set runs to 29, and among the seven we could test individually the only ones earning anything were the ones that named two other companies.

Both halves of that are worth holding. Something here does work, at a position and a volume worth having. It is just not the half being generated at scale, and one page-level query per URL is the difference between knowing that and assuming it.