Why do buyers' questions so often land on “X vs Y” pages?
Many of the questions buyers ask AI are comparisons already: “A or B?”, “give me a few X tools”. To answer them, an AI needs a page that puts several products in one place, measured by one yardstick. A comparison page has exactly that shape: two products and the same few measures at one address. This is an inference about the mechanism, not a statistical finding; we found no public study showing comparison pages are cited more often than other pages.
What we can see is our own evaluations. On 2026-10-26 we evaluated GammaSlide: across 21 AI answers, one of the cited pages was a list article, “15 Best AI Presentation Makers Tested & Compared in 2026”, listing 15 products, cited 3x in those 21 answers, and GammaSlide was not in it. On 2026-10-22 we evaluated MochiStamp: across 3 AI answers, the cited pages included “LINEスタンプ作成AIツール比較｜人気ランキング.com”, listing 8 tools, cited 1x, and again MochiStamp was not in it.
These two examples show that in AI answers, comparison pages written by others hold the space and you are not in them. Writing your own comparison page at least gives the AI one source that includes you. Also, Ahrefs analyzed 15,000 prompts in 2025-08 and found only 12% of URLs cited by AI ranked in Google's top 10, which suggests citation depends less on ranking than on whether the page answers the question directly.
What parts does a citable comparison page need?
Every row below matches something that really exists on Pagent's comparison pages; the right column says how it is actually done:
| Part | Why it is there | What Pagent's comparison pages do |
|---|---|---|
| A one-sentence conclusion up front | The first paragraph is what AI most easily lifts when citing | The description says outright: “Muse scores higher on 3 of the 5 tests both took”, with the test date |
| A side-by-side table | One yardstick, so both AI and people can compare item by item | Lists both scores for each test and marks the higher one; also lists overall score, share done first try, reply time and price |
| Test date and number of runs | Readers need to know how fresh the data is and how many times it was tested | States “tested 2026-10-27, Dots 28 runs, Muse 21 runs” |
| Items only one side was tested on, listed separately | No forced winner on things that were not compared | “Only Dots was tested” and “Only Muse was tested” form their own groups and do not count toward the win tally |
| Each side's strengths | A comparison that is not one-sided is more credible | Split into two groups: “Higher score for Dots” and “Higher score for Muse” |
| Unscored facts with their source | Price, channels and privacy are not the same thing as a score | The page says these come from each product's own site and policy and are not scored |
| FAQ that matches the page | Turn the questions buyers ask into Q&A | Three questions (which is better, which replies faster, when were they tested), with answers matching the numbers in the table, also written as FAQPage markup |
What does a real comparison page look like?
Dots vs Muse (Pagent, tested 2026-10-27): 5 tests both took, Muse higher on 3, Dots higher on 1, 1 tied; overall Dots 7.1, Muse 7.8; done first try 66% against 79%; reply time 9 seconds against 6; price per month $100 against $20. Two tests only Dots took (proactive help, memory) and two only Muse took (routines, connected apps) are listed separately and do not count toward the tally.
Poke vs Tomo (tested the same day): only 4 tests both took, Poke wins 3 and Tomo wins 1, yet the overall score is Tomo 6.8 and Poke 6.7. The page does not hide this contradiction: it shows both numbers and lists the tests only one side took separately. Readers, and AI, can judge for themselves how much “3 to 1” weighs.
The numbers on these two pages are measured, not pre-filled. Some products in Pagent have pre-filled scores, and comparison pages for those cannot be cited as measured results, so we did not pick Manus vs OpenClaw as an example.
How do you write your own comparison page?
1. Pick the opponent and the question
Pick opponents buyers really compare you with, and title the page the way buyers say it: “A vs B”, “A alternatives”. See in the evaluation who AI answers recommend and choose from those.
2. Fix a few items to compare, with one yardstick for both sides
Compare only facts both sides can be checked on: price, supported formats, channels, limits. Test what you can test yourself and state the method. Do not compare only the items where you are ahead.
3. Give the conclusion in the first two or three sentences
“A suits X, B suits Y”, with one specific number or difference. Do not write “each has its merits”.
4. Put it side by side in a table and state the test date
Put the table after the conclusion. Put each item's source and date under the table, for example “official pricing page, checked 2026-11”.
5. Write each side's strengths, and call no winner on what was not compared
Write honestly where the competitor is better. List items only one side has data for separately.
6. Add an FAQ whose answers match the table
Three to five questions are enough: which is cheaper, which is faster, who it suits, when it was checked. Take the numbers in the answers straight from the table. If you add FAQPage markup, see Is FAQ structured data still useful?; the page text and the markup must match.
7. Update whenever a price or feature changes
Comparison pages go stale fastest. When you change a number, change the date. Does structured data help AI search? covers keeping markup in sync with what the page shows.
A comparison page only gives AI something to cite. Whether you are in the third-party list articles that AI cites is a separate effort; see how to get into “best tools” articles.
FAQ
Are comparison pages really cited more often than other pages?
We found no public study showing this. What we can confirm: in two kagentx evaluations, the pages AI cited included list articles that put several products together, cited 1–3 times, and the evaluated product was not in them.
Should a comparison page write itself as the winner?
No. Write honestly where the competitor is stronger, and list items only one side has data for separately, outside the tally. Pagent's Poke vs Tomo page shows both “Poke wins 3 tests” and “Tomo has the higher overall score”.
Should a comparison page show test dates?
Yes. Prices and features change, and the date tells readers and AI how fresh the data is. Pagent's comparison pages all show the test date and the number of runs.
Should a comparison page have an FAQ?
Yes: write the questions buyers ask as Q&A, with numbers in the answers taken from the table. The FAQ text must match the page text.
Can I write a comparison page from pre-filled data?
No. Do not present unmeasured data as a win or loss. When data is thin, write only facts that have a source, and state the source and date.
Sources
- Pagent · Dots vs Muse: tested side by side （Checked 2026-11-05, tested 2026-10-27）
- Pagent · Poke vs Tomo: tested side by side （Checked 2026-11-05, tested 2026-10-27）
- Ahrefs · Only 12% of AI Cited URLs Rank in Google’s Top 10 for the Original Prompt （published 2025-08-11, 15,000 prompts）
- kagentx evaluations: the data on the case study pages; each example gives its date.