Why does freshness change whether AI cites you?

Freshness changes whether AI cites you because some answer engines read your page's dates as real, machine-readable data rather than decoration. Perplexity's own Search API returns a publish date and a last-updated date as first-class fields on every result, and Claude tracks how old a page is. Whether adding those dates lifts your citations, though, nobody has measured on a live engine.

Freshness, in Answer Engine Optimization (AEO), means two things at once: a date a person can read near the top of the page, and the same date inside the page's structured data as datePublished and dateModified. AEO is the practice of engineering a website so that AI assistants — ChatGPT, Perplexity, Gemini, Claude — quote it when someone asks them a question. Dates are one of the few AEO levers you can set in an afternoon and verify yourself.

Watch the walkthrough

Do AI engines really read the dates on your page?

Two of them demonstrably do, and it is worth being precise about what is documented and what is not. Perplexity's published Search API schema returns a date and a last-updated field on every result it hands back, which means those values are carried as data, not read out of your prose. Anthropic's Claude tracks page age. That the engines consume machine-readable dates is official. That adding them raises how often you get cited is a plausible hypothesis nobody has measured on a live engine. Publish the dates. Do not let anyone sell you the lift.

Does a recrawl follow my publishing cadence?

Not as a dial you turn. Perplexity's own engineering write-up describes recrawling as machine-learning prioritised: models predict whether a candidate URL needs indexing at all and when to schedule it, with your update rhythm as one input among several. This is the claim most often mangled in AEO advice, so we state it in their terms rather than ours. Publishing more often does not buy you a guaranteed recrawl; it makes you a more plausible candidate.

Why does a good old page lose to a thinner new one?

Because the engine has to choose, and one of you left the field blank. You wrote a strong page two years ago. It is still true and it still answers the question better than anything else out there. But on paper it says two years old, and nothing on it says otherwise. Against a thinner page that says updated last month, the assistant has a reason to prefer the thin one. You did not lose on quality. You lost on a field nobody ever filled in.

Which two dates matter, and where do they go?

At Webappski this is the first thing we set on a client's site, and it is only two moves. One: a visible date near the title, labelled, so a human can tell whether it means published or last updated. Two: the same date inside the structured data as datePublished and dateModified, so a machine can read it. Both places, and the two must agree. A date that lives only in the schema is invisible to your readers; a date that lives only in the page text may never be picked up reliably by an engine.

What does our own page score on this test?

Two out of four — and we show it on camera rather than grading someone else. The page is our hands-free form filling guide. It carries a visible date near the title, and its structured data says datePublished and dateModified, both April 2026, in agreement with what a reader sees. Those are the two checks it passes. It fails the other two: the visible date carries no label, so a reader cannot tell which date it is, and dateModified is identical to datePublished, which means that months on we have not genuinely improved that page once. Two of four is the score most sites actually have.

What does Google itself ask for?

Four things, in its own words, and none of them cost money. Add a user-visible date to the page and feature it prominently. Label your dates with text like "Publish" or "Last updated". Specify dates with structured data, using a CreativeWork subtype such as Article or BlogPosting and its datePublished and dateModified fields. And make the visible date and the structured date match. Google adds one more rule that catches most cheating: the dates must describe the publication or update of the page, not the events described on it.

What do the free AI-visibility tools report about freshness?

Nothing — and that is not a criticism of the tools, it is a gap in what they are built to answer. We checked three of them on 5 August 2026. Ahrefs' free AI Visibility Checker is genuinely free, needs no signup, and shows real things: total AI mentions, a breakdown by platform, your top topics, your top cited domains and your top cited pages. Semrush's free checker gives three runs a day without registering, with a visibility score, mentions, citations and a competitor comparison; the only daily freshness on that page is theirs, not your page's. HubSpot's AI Search Grader scores five dimensions — sentiment, presence quality, brand recognition, share of voice, market competition — and not one of them is freshness. Three good tools, three different reports, zero dates.

What does our open-source tracker do instead?

It ties the lever to the engine rather than to a score. aeo-platform is our own tracker, free and open source on npm. Per engine, its report states why freshness matters there — for Perplexity, that it runs its own index and treats publish date as a first-class signal. Per answer, it flags heuristically whether the engine appears to have retrieved something live or answered out of stale training memory. That second question is the one a single visibility score cannot ask, and it is the one that tells you whether your freshness work has been seen at all.

How long does a fresh update keep working?

Not long, if the one published measurement is representative. A vendor test — their measurement, not ours, and not one we can independently confirm — found the citation lift from a fresh update faded within about two weeks. Treat that as somebody else's observation rather than a law. The practical conclusion holds either way: freshness is a rhythm, not an event. Two consequences follow. Updating an existing page usually beats publishing a new one, because the old page already carries whatever trust it earned. And the update has to be real — new information, a corrected fact, a better answer — because changing one word and bumping the date is a trick that stops working.

What is the full freshness checklist?

Four checks and a habit, in order. A visible date near the title. Labelled, so a reader knows which date it is. The same date in the structured data, as datePublished and dateModified. The two agree. Then the habit: a cadence you can actually keep, even one quarterly pass over your top pages. Our own post scores two of four, and it is on the list to fix — which is the honest version of every checklist anyone hands you.

Frequently asked questions

Does adding datePublished and dateModified guarantee more AI citations?

No, and anyone promising that is selling you an unmeasured claim. What is documented is that engines consume machine-readable dates: Perplexity's Search API returns date and last_updated as first-class fields, and Claude tracks page age. Whether adding them lifts citations has not been measured on a live engine. Set the dates because they are cheap, correct and readable by both humans and machines — not because someone quoted you a percentage.

Should I update an old page or publish a new one?

Usually update the old one. It already carries whatever trust, links and history it earned, and a genuine improvement resets its freshness signal without splitting your topic across two competing URLs. Publish something new when the question itself is new. Rewrite what exists when the answer has moved on.

Is it enough to put the date only in the structured data?

No. Google asks for a user-visible date featured prominently, a label such as "Published" or "Last updated", the same values in structured data, and agreement between the two. A schema-only date is invisible to readers and gives you nothing to reconcile against; a text-only date is not reliably machine-readable. Both, labelled, matching.

Can I just bump the date without changing anything?

You can, and it is exactly what the guidance rules out: dates must describe the publication or update of the page itself. A date bump with no real change is a claim you cannot back up, and it costs you the one thing this lever is for — being trusted about how current your page is. Change something worth changing, then move the date.

How often should I refresh my key pages?

Often enough to keep the promise your dates make, and no more. For most small sites one quarterly pass over the ten pages that matter is enough: check what is out of date, fix it properly, then move dateModified. A rhythm you can hold beats an annual rewrite you announce and never repeat.

Want your own pages scored against this checklist?

We run this same four-check freshness pass, page by page, on real client sites — and the first AI-visibility audit is free. If you want to know how ChatGPT, Perplexity, Gemini and Claude currently see your pages, request a free audit at webappski.com.