Pages AI search cites for best X carry a date

Most comparison pages that AI search cites show a date, and few name a pick. We read 211 pages our scans cited for "best X", "X alternatives" and "X vs Y" questions on 14 October 2026: about 70% carried a visible date, 55% an FAQ and 12% a methodology section.

Pages fetched 14 October 2026. Published 15 October 2026. Updated 15 October 2026.

How we read the pages

We started from every question in our AI visibility scans that reads as a comparison: it contains best, top, alternative, vs, versus, competitors or comparison. That gave 255 questions, 4,937 citations that resolve to a page, and 3,962 distinct cited addresses once tracking parameters were stripped.

Internal, test and seed accounts, and research runs with no owner, are left out. The pool still holds the ten boards we built as outreach for other people's products, because our seed filter does not catch them.

From those addresses we drew 300 at random with a fixed seed and fetched each once with our own crawler. Of those, 267 came back. The other 33 failed: 21 behind a bot wall (HTTP 403), six returning 404, 410 or 503, one returning 402, three network failures, and two addresses our fetcher refused as unsafe.

We read only the HTML the server sent. No JavaScript ran and no model read a page; the run cost $0. Of the 267 pages, 56 held under 300 words without scripts, mostly pages drawn in the browser, so every share below is over the 211 readable pages, where a missing part means the page really lacks it.

The detection rules, in plain words

• Visible date: a date a reader can see, in a time element, written out in full near the top of the text, or after the words updated, published or reviewed. A date kept only in the page metadata is counted apart.
• Visible byline: an author shown on the page, in an author element or in a short line such as "By" or "Written by" and a name. An author named only in metadata is counted apart.
• Pick in the opening: the first 150 words after the main heading name a winner, with phrases such as "our pick", "best overall" or "we recommend".
• Methodology section: a heading such as "How we tested", "Methodology" or "Why trust us".
• Comparison table: an HTML table of at least 3 rows and 3 columns.
• Use-case table: a table whose header or first column reads "best for", "ideal for", "use case" and the like.
• FAQ: a heading reading FAQ, "frequently asked" or "people also ask", or FAQPage structured data.

The rules were fixed before the numbers were read. Three were tightened after we checked by hand what they had matched, and the date share moved with them. The corrections below say by how much.

What the cited pages carry

A listicle-shaped page is one whose address sits under a blog, guides or reviews path, or carries best, top, alternatives, vs or compar in it. Every part is more common on those 129 pages than across all 211.

A further 19 readable pages, 9%, carry a date in machine-readable form only, where a crawler finds it and a reader does not.

Tap to zoom

What the rules got wrong

A hand check of the evidence behind each match found false matches on five rows. The figure uses the corrected shares; this table shows how each was reached.

How the date share moved, in order. The first run scored 57%. Fixing the date pattern so it caught dates glued to the next word took it to 62%. Spacing out each element's text, so a year and a reading time no longer ran together as "202624 min read", took it to 72%, across the 70% line we had written down before the run.

The year-only clause was in the rule from the start. Without it the share is 67 to 70%, on that line or just under it. Read the date as the most common part, not as a settled 72%.

We checked the pick rule by hand too. Of 12 listicle pages it scored as naming no pick, one arguably does; the rest open on context, a list of names, or the reason people switch. The low share is real, not a measuring fault.

What we expected, and what held

We wrote our guesses down on 14 October, before the run, with a 60% bar for a part to go into our own brief. A visible date on over 70%: held only as first scored, after two rule changes, so call it borderline. A comparison table on over 70%: not held. The pick in the opening on about half: not held.

A methodology section under 30%: held, at 12%. An FAQ on roughly half: held, at 55%. A use-case table under 30%: not held, narrowly, at 34%. A visible byline sits at about 45 to 47% once corrected, under the bar.

What this does not show

Every figure here is a share of pages that were cited. We hold no sample of comparison pages that answered the same questions and went uncited, so nothing here says a date, an FAQ or a table wins a citation. It says what the pages that won one tend to carry.

The rules read text, and text rules slip. An organisation byline such as "Mixpanel Team" counts as a byline. The questions belong to real accounts across devtools, SaaS, real estate, cables and farming, not one niche. It is one sample of 300 from 3,962 addresses, each fetched once, on one evening.

If you repeat the fetch, pin Node 22. Under Node 26 our hardened fetcher rejects every request with an invalid-argument error, so every page here was fetched on the older runtime.

The free trick: count the parts before you write

Before you write a "best X" or alternatives page, look at what is already cited for the question you want to win. It needs no account, only the answer engines you already use.

• Ask your question in ChatGPT with search on, or in Perplexity, and open the first five pages the answer cites.
• On each one, look for three parts: a visible published or updated date, an FAQ section, and a real comparison table.
• A real table sets several products side by side across several columns. A pricing grid for one product is not one, and neither is a list of suppliers.
• Tally the five pages. Build the parts most of them carry first.
• Then write the rest. A pick up top or a methodology section can stay, but few cited pages in our sample carried either.

In our data the date is the part that turns up most. If your page shows none, put a published or updated date near the top where a reader can see it, and keep it true when you edit.

What AfterLaunch does with it

When your Board ranks a comparison or alternatives page as the next move, AfterLaunch briefs it with the evidence behind it and hands it to your AI when you say yes. A brief for a page that names a rival already asks that every claim about the rival trace to a page the writer opened.

That brief does not yet name the structural parts. Adding the one part this data supports, a visible published or updated date, is on the way. The FAQ is under review: it held our guess at 55%, but sits under the bar we set before the run.