Comparison pages come in four shapes, and they answer four different questions. Lumping them together is why most advice on them is vague. This post separates them, shows what one slice of our data says about each, and gives a procedure for writing one that a skeptical buyer, a competitor's lawyer and an AI system could all read without finding a problem.
What kinds of comparison pages are there?
| Page | The question behind it | What our data says | Main risk |
|---|---|---|---|
| Your brand vs a named competitor | "Should I buy A or B?" | Nothing: the index has no "X vs Y" prompt | Unfair framing, stale facts |
| Alternatives to a competitor | "What else is like B?" | The alternatives question: 5.0 brands per answer, inflated because the prompt names a brand | Disparagement, unsupported claims |
| Alternatives to your own brand | "What if A is not right for me?" | The prompt named brands the index chose, not yours; those brands were rarely recommended in the answers | Little reader value unless honest |
| Best product for a use case | "What fits my situation?" | The use-case and best-overall questions: 2.2 and 2.4 brands per answer | Reads as an ad if it ranks you first with no reasoning |
The only prompt type in our index that resembles any of these is the alternatives question. In the October 2026 edition, every category was asked for alternatives to a named brand, for example "alternatives to Ninja air fryers" on the air fryers page and "alternatives to Merrell Moab" on the hiking boots page. The alternatives research post has the full breakdown; the short version is below.
Two numbers matter for a page strategy. First, an answer to the alternatives question listed 5.0 recommended brands on average, against 2.3 for the questions that name no brand (1,271 answers). Second, the brand named in the question was among the recommended brands in only 21 of the 296 answers where we could match it. Neither number says a comparison page caused anything. They describe the shape of the answers a page would be competing to appear in.
What follows from the shape is modest. The alternatives prompt names a brand in its question, which inflates its brand counts, and the index replaced it in its monthly run for that reason. So 5.0 is a property of that prompt, not a sign of extra slots in the market that a page could compete for. The named brands were ones the index chose, not yours, so the data does not show what a question about your own brand would return. Whether either kind of page changes who gets named is unmeasured. Samples this small are noisy: a brand named in 1 of 8 answers is at 12.5%, but the 95% Wilson interval for that runs from 2.2% to 47.1%.
What do the platforms say about comparisons?
As of October 2026, OpenAI's help page on ChatGPT shopping says you can ask it to compare products, explain trade-offs or suggest alternatives. So comparison is a supported use, which is a statement about the interface, not about which sources feed the answer. Google says there are no additional requirements, and no special optimizations needed, to appear in AI Overviews or AI Mode. Neither says anything about comparison pages in particular.
What stays open: whether a first-party comparison is treated differently from a publisher's, how an assistant weighs a competitor's own page against yours, and whether a table is quoted as a table or rewritten. In our October data, by a hand sort, 19 of Perplexity's 30 most frequent retrieved domains (counted across all three markets) are publishers or review sites, and the other 11 are retailers; shopping aggregators can be classed either way, so read the split as approximate (see the reviews and ratings post). We did not record whether any of those pages was a comparison, so it says nothing about how a first-party page fares beside them.
How do you write a comparison that holds up?
- Choose the criteria before the verdict. Take them from real buyer questions, the way the buying guides post does, and publish them. Criteria chosen after you know the answer produce a rigged table.
- Compare like with like. Same configuration, same price tier, same date. State the date, the model names and any option you excluded.
- Source every competitor fact. Link each competitor row to that company's own product page or specification sheet and record the day you read it. Facts you cannot source stay out.
- Show where the competitor is better. One honest row where you lose makes the rest believable and is more useful to a reader choosing between two products.
- Use plain-text names. Write the competitor's name as text. Do not use its logo, its design cues or anything that suggests a link between the companies.
- Say who wrote it. A comparison on your site is yours, and the page should read that way. Do not build a separate "independent review" site you control; the FTC rule on consumer reviews prohibits a business from misrepresenting that a site or organization it controls provides independent reviews or opinions, including of its own products.
- Keep it as text. Put the table in the page as a real table, not an image, so that anything that reads text can read it.
- Date it, check it, correct it. Prices and specifications move. Add a "last checked" date, set a monthly reminder, and give competitors and readers an address for corrections.
What are the legal limits?
This is information, not legal advice, and it covers the US and Canada only. Check with a lawyer before publishing anything that attacks a competitor's product.
- United States, FTC. The FTC's policy statement encourages naming competitors but requires clarity and, where needed, disclosure to avoid deceiving the consumer. It says it has supported brand comparisons where the bases of comparison are clearly identified, and that it judges comparative advertising the same way as all other advertising.
- United States, Lanham Act. Section 43(a) makes a person liable to anyone likely to be damaged if, in commercial advertising or promotion, they misrepresent the nature, characteristics or qualities of their own or another person's goods. It also reaches representations likely to cause confusion about affiliation or sponsorship, which is why plain-text names and no implied link matter.
- Canada, Competition Act. Section 74.01 covers representations to the public that are false or misleading in a material respect, and performance claims that are not based on an adequate and proper test.
In practice the test is the same everywhere: every claim you make about a competitor should be true, current, supported by evidence you kept, and not misleading in context. Keep the evidence file with the page.
How do you test whether a page helped?
- Write the prompts a buyer would use, such as "alternatives to [competitor]" and "[your brand] vs [competitor]", plus one use-case question for the category. Keep them fixed.
- Ask each prompt in each engine several times, in clean sessions, before you publish. Record which brands are named and which source URLs appear.
- Publish, wait, and ask again with the same prompts and the same number of answers.
- Compare your share of answers with an interval, as in the share of answer post, and track one similar product you did not change as a control.
Expect the first result to be inconclusive at small samples. If the intervals overlap, the honest conclusion is that you cannot tell yet.
Frequently asked questions
Can I name a competitor in the title of a page?
Under the FTC's policy statement, naming competitors in truthful comparative advertising is encouraged. Trademark questions around titles, logos and search ads are separate and depend on the facts, so ask counsel.
Should I publish an "alternatives to us" page?
Only if it would help a reader who is deciding that you are not the right fit, and only if it is honest. In our October data, answers to the alternatives question rarely recommended the brand they named. That prompt named brands the index chose, not yours, so the data does not show whether such a page would be found.
How we write: drafted with AI tools from Dopbot Index data and public sources; every number and outside fact is traced to its source in a claims record and checked before publishing. More about how we write; corrections to hello@dopbot.com.