Should you write comparison and alternatives pages to get cited by AI?

Comparison and alternatives pages are worth writing, with one correction to the usual advice: they earn you a place in the set an engine considers, and they are weak at making you the name it picks.

Two mechanisms are at work and most guidance on this subject collapses them into one. The first is retrieval and citation, where a page that directly answers "what are the alternatives to X" is exactly the shape of document an engine reaches for. The second is recommendation, where the engine decides which name to put in the sentence, and that decision draws on what the model already knows about the brands in front of it.

Mahe Chen and Nick Koudas measured how far apart those two things sit. Across 158,099 generative engine answers carrying 468,037 citations, they attribute about 32% of the variation in citations to which pages the engine's own search retrieves and 2.8% to page properties, and report that changing only a brand name reordered ChatGPT's product rankings over identical facts (arXiv 2610.15033, submitted 10 October 2026). Their summary is that pages are cited when searched and brands are also named when remembered.

So the honest case for a comparison page is a retrieval case. Published well, it answers your category's buying question on a document that mentions you, and it gets quoted for doing so. Published as a bid to be crowned, it tends to fail in a specific and measurable way, which the next two sections cover.

One disclosure before any of it. We publish two pages of this kind, a comparison of AEO agencies that includes SIGNALS and a comparison of AEO tools that does not, so this article is written by somebody with a stake in the answer. Read our own entries on those pages the way you would read any vendor's account of itself.

Do comparison pages actually get you cited and recommended more often?

Citation of your own comparison page does raise your recommendation rate, by a small and measurable amount, and the best evidence available puts a number and an interval on it.

Trakkr's study 011 tracked 970 first citations of a brand's own page between May and August 2026, across 180 brands drawn from 661 sampled, inside a weekly panel of 1,050,171 AI answers collected from 2 March to 11 October 2026. Using a staggered difference-in-differences design with eight-week windows either side of each first citation, and brands whose pages were never cited as controls, it reports a 3.6 percentage point lift in recommendation rate in the citation week, 95% confidence interval 2.3 to 4.8, and 2.9 points on average across the next eight weeks, interval 1.5 to 4.3, rising to 4.2 points for challenger brands. Its pre-trend test came out at a joint p of 0.58, which is the check that makes the design worth reading.

Three caveats belong with those numbers. Trakkr sells AI visibility software, so the finding that owned pages move recommendation is one that suits its business. The study describes itself as exploratory rather than randomised. And publication-date anchoring could not be completed, so effects are dated to first citation rather than to when a page went live.

Our own category research supports the direction from a different angle. Across 27 buyer queries in three pharmaceutical skid lanes, 70% to 79% of the 4,395 citations behind the AI answers pointed at manufacturers' own capability pages rather than at directories, trade media or third-party rankings. In a category with few independent publishers, vendor pages are most of what is available to cite, and a page that answers the buying question directly is competing against very little.

Read the size honestly. Three percentage points is a real effect and a small one, which makes a comparison page a sensible piece of work rather than a transformation.

Does being cited on your own comparison page get you recommended?

Being cited on your own page frequently does not get you recommended, and the measurement of that gap is the most uncomfortable finding in this subject.

Lily Ray ran 100 B2B "best [category] software" queries through Google AI Overviews on 15 April, 15 May and 8 June 2026, using Ahrefs Brand Radar to capture answer text and cited sources. Where a brand's own self-promotional listicle was cited as a source, that brand was left out of the actual recommendation in 224 of the 323 cited self-promotional listicles, which is 69%, according to her study published on 17 June 2026. The competitors named inside those listicles were recommended instead, and she names specific cases. Her reading is that the recommendation decision is anchored to how much of the web is talking about you.

Trakkr's panel shows the same pattern from the other side: in its first run, a median 28.9% of a brand's cited pages named at least one competitor, and pages a brand owns accounted for a median 4.6% of that brand's AI citations. A comparison page is, by construction, a document that introduces an engine to your competitors in a context where it trusts the document.

Ray also records context worth knowing before you write one: a January 2026 algorithmic demotion of sites using the self-ranking tactic, and new AI Overview disclaimers about self-proclaimed experts. She is explicit that the mechanism is hard to confirm and that some lesser-known brands still get recommended.

What this does not justify is leaving yourself off your own page. Omission buys nothing, because the engine then reads your document as a recommendation of other people with no entry for you at all. The useful response is to be specific about who you are wrong for, which is what readers and engines both repeat.

Why do the two studies disagree, and which one applies to you?

The two studies do not actually disagree, and the reconciliation sits inside Trakkr's own numbers: the lift from your own page appears when the engine searched the web and vanishes when it answered from memory.

Trakkr split its result by how the answer was produced and reports a 6.0 percentage point lift in recommendation rate for answers built on web search and 0.0 points for answers generated from the model's own knowledge. A comparison page is a retrieval asset. Where the engine goes looking, your page can change the outcome; where it answers from what it already knows, your page is not in the room.

Ray's measurement was taken in Google AI Overviews on category head terms, which is the surface where brand familiarity is strongest and where the biggest names in a category are already known to the model. Trakkr's panel spans engines and prompts including many narrower ones. Same mechanism, two different mixes of memory and search, two different answers.

Which applies to you follows from what your buyers ask. A broad category head term, the kind where a dozen established vendors are household names in the sector, is mostly a memory question and your comparison page will be cited and passed over. A specific, qualified question, the kind a buyer types when they already have constraints, is mostly a search question, and that is where a well-built comparison page earns its keep.

Our own demand data shows how specific real queries get. The queries reaching this site from Bing, the index behind ChatGPT's search, run to full sentences and in one case to 38 words, and 55% of the 141 queries Google Search Console recorded for this site in the 90 days to 8 November 2026 were six words or longer. Our guide on what people type into AI search sets out what that looks like in practice.

Separating the two kinds of question is the first thing worth doing before commissioning a set of comparison pages, and a free visibility assessment does it directly: running your buying questions across ChatGPT, Perplexity, Google AI Overviews and Claude shows which answers arrive with cited sources, which arrive without any, and therefore which of your questions a page can still influence.

What has to be on a comparison page for an engine to use it?

A usable comparison page states its criteria before its options, puts them in a table, says who each option suits, carries a date, and discloses who published it. Those five are not stylistic preferences; four of them are the parts engines quote.

Element	Why it earns its place
Criteria stated before the options	Becomes the sentence an engine uses to explain its recommendation
A table of options against criteria	Extracts cleanly; prose comparisons get paraphrased or dropped
A "who this suits" line per option	Buyers self-select on it and engines repeat it almost verbatim
A visible date and review cadence	Buying pages decay fastest; prices and features move in months
A disclosure of who publishes the page	Required of us from 12 November 2026, and it is what readers discount for anyway

One finding should change how you write the thing. Discovered Labs, analysing about 2 million prompt, engine and citation observations across ChatGPT, Claude, Google AI and Gemini, and crawling 10,000 cited URLs with more than 60 features each, reports a coefficient of -0.12 for listicle or review format, against +0.37 for prompt-content alignment with a 95% confidence interval of +0.33 to +0.41. Discovered Labs sells answer engine optimisation services, and a predictor in that study had to survive four of nine robustness checks rather than all of them.

Read those two coefficients together and the instruction is clear. The listicle shape is a mild liability and matching the question is the strongest page-level lever there is, so a comparison page wins on being the direct answer to a specific buying question and loses if it is a generic roundup wearing a number in its title.

Should you name competitors and say when to pick them?

Name them, and say plainly when somebody should pick them, because a page that only describes who you are right for is read as marketing by both the buyer and the engine.

Specificity about fit is the thing that buys credibility, not omission. A line saying "choose this one if you need a large vendor that will pass a formal security review" is useful, repeatable and checkable, and it is the sort of sentence an engine lifts whole. A line saying a competitor is "limited" without evidence is the opposite, and it is also the fastest way to put a factual error about a named company on your own domain.

Three rules keep that safe. Never state a competitor's price, feature or claim without opening their page and citing where you got it. Never invent a weakness: where you cannot substantiate a criticism, describe what they are good at and where the fit is narrow. And do not leave out an obvious market leader to make your own list look better, because an engine cross-checks against other sources and a list missing the obvious names reads as unreliable.

Our own tools comparison carries the disclosure in the form we think works: "This page is published by SIGNALS, which appears in the table above, so read our own row the way you would read any vendor's account of itself", followed by a note that vendor pricing in this category moves within months and that the page is reviewed quarterly. Our agencies comparison carries a section headed "When should you not pick SIGNALS?" for the same reason.

Put the disclosure next to the comparison rather than in a footer, and keep it when the page is refreshed. Google AI Overviews now carry disclaimers about self-proclaimed experts, per Ray's June 2026 study, so a page that volunteers its own authorship is working with that rather than against it.

Is it allowed to publish "independent" comparison sites of your own?

Publishing a comparison site that hides its own ownership is prohibited outright, and the rule became explicit this month rather than being a matter of taste.

Anthropic's usage policy, effective 12 November 2026, prohibits creating or operating "fake personas, accounts, media outlets, or organizations", misrepresenting or concealing "the sponsorship of content or messaging", distributing content through "networks of fake or ostensibly independent outlets, websites, or accounts", and specifically any attempt to "Manipulate the sources from which search engines or AI systems draw answers" by "seeding them with content that misrepresents its origin, authorship, or independence", giving as its example "networks of sites posing as unaffiliated sources corroborating the same claims". Marketing or selling such work as a service is prohibited in the same section.

Google's position covers the ranking side. Its search spam policies state that spam includes "attempting to manipulate generative AI responses in Google Search", in the introduction to the policy page (read 9 November 2026).

Where that leaves ordinary comparison content is unchanged and worth saying plainly, because the policy language is broad enough to worry people who are doing nothing wrong. Publishing a comparison on your own clearly branded domain, with your own name on it, as your own view, is not within any of those bullets. What is prohibited is the pretence of independence: a second site presenting itself as a neutral reviewer, a review network with a common owner, or a persona.

The practical test is a single question. If a buyer discovered who published the page, would anything about how they read it have to change? Where the answer is no, you have a comparison page. Where the answer is yes, you have the thing two policies now name.

What else do people ask about comparison pages and AI search?

Marketers ask these five questions most often when deciding whether to build out comparison and alternatives pages.