"Best X" lists and comparison pages: how AI uses them, how to get on them

Why these questions matter so much

Prompts to an AI assistant sort roughly into three kinds: learn something, do something, choose something. The third kind is where buying happens, and it has two shapes. "Best X for Y" asks for a shortlist ("best scheduling tool for a recruiting agency"). "A vs B" asks for a verdict between options the person already knows ("Fernwick vs Slotwise for a team of ten"). Listen to the questions your sales team hears in first calls: most are one of these two shapes with the names filled in.

Both get answered as a list: five to eight names with a line on each, or a few paragraphs of contrast with a recommendation per situation. In both cases the names, and their order, come from pages that already exist.

How the assistants actually answer them

The search-based assistants (ChatGPT with search on, Perplexity, Gemini, Google's AI Overviews) run one or more web searches for the question, fetch a set of results and write an answer grounded in them. For a "best X" question, the results that rank are overwhelmingly lists: "The 10 best scheduling tools in 2026" on a software blog, a review site's category page, a Reddit thread where people name their favourites. For an "A vs B" question, the results are comparison pages: the two companies' own "vs" pages, a review site's side-by-side, a blog post that compared them.

Two things follow. First, the assistants aggregate; they rarely evaluate. An assistant has not signed up for your product. It knows what the pages it read say about you, which is why a wrong fact on a cited page becomes a wrong fact in the answer (see when ChatGPT gets your company wrong). Second, agreement across pages is what gets a name into the answer. A product named on four of the six lists the assistant read is very likely to appear; a product on one of them might not. The mechanics are in how ChatGPT decides which pages to cite. For commercial prompts, the short version is that the lists are the input.

A "best X" answer is a summary of the "best X" pages the assistant found. Change the pages and you change the answer.

Move one: get onto the lists that already get cited

You do not have to guess which lists matter. Run your commercial prompts through the assistants, collect the citations, and group them by page. The pages that come back again and again, across engines and across weeks, are the lists you need to be on. In Echo this is the Citations page filtered to the "best" and "vs" prompts, but any tool that records citations, or a spreadsheet and an afternoon, gives you the same list.

For each cited list, check three things.

1. Are you on it? If not, that is the gap.
2. Are you described correctly? Lists go stale. A line that says "no free plan" or "no API" when you have both is worse than absence, because the assistant will repeat it.
3. Where are you on it? Position ten on a list of ten rarely makes it into an answer that names five.

Then act, per list, in a way that respects how the author works.

• Software blogs and independent reviewers. Write to the author, not a generic address. Say exactly what is wrong or missing, in one paragraph, with a link to the page on your site that proves it. Offer a trial account with real data in it, or a 20-minute walkthrough. Ask for a correction, not a rewrite: authors update lists far more often than they add to them.
• Review sites (G2, Capterra and their peers). Their category pages are cited constantly, and their rankings are built from listing completeness and review volume, not pitches. Claim your listing, fill every field, keep pricing and integrations current, and ask recent customers for reviews in a steady trickle. Bloggers draw from these sites too. Why AI answers lean on Reddit, reviews and forums explains the weight they carry.
• Community threads. A Reddit or forum thread that lists favourite tools is a list too, and often a cited one. Answer as yourself, with your company in your profile, only where a question is genuinely open, and never pose as a customer.

Keep a log per list (page, status, author, date you wrote, what changed) and re-run the prompts a few weeks later to see whether the answer moved.

Move two: publish your own

The lists you do not control will always be partial. The pages you control can be exactly what the assistants are looking for, provided they are honest enough to be trusted.

Your own "best X" list

A "best scheduling tools for agencies" page on your own site that includes your competitors reads oddly to many founders. It should not. The assistant is going to name your competitors anyway; the only question is whose description of them it reads. A fair, current list that is specific about who each option suits gets cited for the category, with your product in it, in your words, in the position you choose. A list that names only you, or dismisses everyone else, is recognised for what it is and ignored.

"Fernwick vs Competitor" pages

For each competitor your sales team hears about, publish a comparison page with a real table: the same rows for both products, filled honestly, with a date on the page. Rows that matter in most categories are pricing per seat, free plan, integrations, API, data residency, support hours, and the two or three capabilities buyers actually compare. Where the competitor is better, say so. A page that concedes two rows and wins five is believable; a page that wins every row is not, and the assistant has the competitor's page open in the same search.

"Alternatives to Competitor" pages

People searching "alternatives to [competitor]" have already decided to leave. An alternatives page lists several options, including yours, and explains the situation each suits. It is a list page and a comparison page at once, and it ranks for a prompt shape that is almost pure buying intent.

How to write a comparison an assistant will trust

The qualities that make a comparison useful to a buyer make it quotable by an assistant. The general rules are in writing content that AI answers quote; these are the ones specific to comparisons.

• Facts with dates. "Fernwick's Team plan costs X per seat per month (checked 22 September 2026)." A date tells the assistant, and the reader, that the fact is maintained. Put it on the page, not only in the sitemap.
• A table, early. Assistants extract tables well and quote them often. Put the comparison table near the top, with plain headers and one fact per cell. "Yes" or "from X per month" is extractable; "robust and flexible" is not.
• Say who each option is best for. One short line per product: "Best for agencies that bill by the hour" or "Best if you already run on HubSpot." These lines are what a "best X for Y" answer is built from; they let the assistant match the prompt's "for Y" to a name.
• No sneering. Adjectives about competitors cost you trust and gain nothing, because the assistant strips tone and keeps facts. State what the competitor does and does not do and let the reader conclude.
• Keep it current. A comparison with a competitor's old pricing is a wrong fact with your name on it. Put the page on a quarterly review and update the date when you check it, even if nothing changed.

A worked example. Fernwick, a scheduling tool, publishes "Fernwick vs Slotwise: which is right for your team". The table has nine rows. Fernwick concedes "mobile app" and "number of integrations", wins "round-robin by skill", "team-wide availability rules" and "pricing for teams over ten", and ties on the rest. Under the table, three lines: best for solo consultants (Slotwise), best for agencies that schedule on behalf of clients (Fernwick), best for teams already on a particular CRM (depends on the CRM). That page gives an assistant everything it needs to answer "Fernwick vs Slotwise for an agency" with Fernwick near the top, without a single claim Slotwise could dispute.

Position matters, not just presence

Watch the answers over a few weeks and a pattern shows up: the names near the top of the cited lists tend to be the names in the answer, often in the same order. An assistant asked for the best five, reading lists of ten, leans on the first few entries of each. A list where you appear at number eight is a list you are, in practice, not on.

Position on a list is partly the author's judgement and partly their data (review counts, ratings, listing completeness), and you can influence both. On your own pages you choose the order, one more reason to publish them. Track your position in AI answers separately from whether you appear at all; the two move for different reasons.

The ethics line

Everything above works because the pages are honest.

• No paid placements disguised as editorial. Paying for a listing is fine where the site discloses it, as review sites and directories do. Paying a blogger to put you at number one on a page that claims to be independent is deception, and when it comes out, the page and your name on it lose their value.
• No fake lists. Publishing "the 10 best X" on a site you set up to look independent, or seeding threads with sock-puppet recommendations, is the same thing one step removed. The assistants read what buyers read; the buyers find out.

The honest version is also the one that keeps working. A fair comparison page from three years ago, kept up to date, is still cited. A rigged one gets found, corrected in public and dropped.