One Point in Your Audit, and a Cheap Test: Does Your Site Show AI a Date?

The date that says 2019

That's not a publication date. It isn't a "last reviewed" date either. It's a footer somebody's theme generated once and nobody has touched since.

A copyright year doesn't count as a date signal for our audit, and it doesn't count as one in schema either. So it's worth asking whether your pages show a machine any date at all, and whether it matters.

There's some data on the second question. It's smaller than the headlines make it sound.

What Seer found, and how old it is

Seer Interactive published a study on June 25, 2025, written by Sonny Vasquez. It's over a year old, so "the past year" and "2023-2025" in it are both measured from mid-2025, not from today.

They looked at more than 5,000 URLs with extractable publish dates that were cited across ChatGPT, Perplexity, and Google AI Overviews. Citation data came through Peec.ai. Then they split it by engine. Roughly 85% of AI Overviews citations came from content dated 2023 to 2025. For Perplexity it was about 80%. For ChatGPT it was about 71%, and ChatGPT reached back as far as 2004. Seer noted that some of those older sources are Wikipedia articles, which hints that authority can beat freshness when the source is strong enough.

The server log side is narrower than it sounds. It came from three OpenAI bots only, not "AI crawlers" as a category. Of those hits, nearly 65% went to content published within the past year, and 89% went to content updated in the last three years.

Industries differ. Financial services showed extreme recency bias, with nearly all hits landing on 2024 and 2025 content. For how-to content, Seer's example was a decking site whose pages from as far back as 2004 still drew AI crawler activity. So old instructional content can still get crawled. Nobody has to panic about a ten-year-old how-to page.

And this is one agency's own study, using correlational data. It shows that recency travels with crawling and citation. It doesn't show that a date causes either one.

What Otterly tried

Otterly.ai published an experiment on September 24, 2026, and it's being read as "just change the date." That reading is wrong. (Otterly sells AI-visibility tracking, so they have a horse in this race.)

On August 4, 2026, they took 10 of their own older articles and changed only the visible "Last updated" date. Their words: "The underlying content remained unchanged." Two other unchanged articles served as controls. They tracked 320 prompts across seven engines, about 67,200 responses, comparing July 5 to August 3 against August 5 to September 3.

The changed pages went from 4,800 citations to 6,312, up 31.5%. The controls dropped from 358 to 320, down 10.6%. By engine, ChatGPT was up 692%, but that's 37 citations to 293, a tiny base. Claude rose 87%, Perplexity 217%, Google AI Overviews 15%. Google AI Mode fell 20% and Gemini fell 37%. They gave no Copilot figure.

Ten pages. Two controls, with a baseline about one-thirteenth the size of the test group's. One vendor, its own site. And Otterly says so themselves: "Citation trends were already moving before the date change, so we interpret the findings as directional rather than conclusive."

Directional is the right word. It's a reason to check your own site, not a reason to rewrite every date you own. Their post also doesn't say whether the schema dateModified changed along with the visible text.

Don't bump a date you didn't earn

The lazy takeaway is to go change every "last updated" to today. Don't. A date that says you reviewed the page when you didn't is a small lie on your own website, and the first customer who notices will notice it about everything else too.

The honest version is boring. Next time you actually go through your pricing page, your service-area page, or your FAQ and confirm the hours, prices, and services are still right, put the date on it. Those are the pages where "reviewed this month" means something to a person and to a machine. A page about how a water heater works doesn't need one.

The part our audit cares about

Otterly changed a visible text date. A plain-text "Last updated: March 2026" line does not pass our audit's date check.

The audit reads machine-readable dates only. It counts three things: a <time> element with a non-empty datetime attribute, a <meta property="article:published_time"> tag, or JSON-LD with dateModified or datePublished (including inside the @graph that Yoast and RankMath generate). That's the list.

It does not count a <time> tag with no datetime or a footer copyright year.

The fix for a page that has a visible date and fails is small. Wrap it:

<time datetime="2026-09-15">September 15, 2026</time>

Or add dateModified to the page's JSON-LD. Either one does it. Same date on screen, and now a machine can read it.

How much this is worth

One point. Technical Health is 12 points, spread across HTTPS, sitemap, canonical tag, viewport, date signals, landmarks, heading hierarchy, internal linking, and an informational check. Date signals is one of those, pass or fail, no partial credit. Out of a 100-point audit, that's one point.

Two more limits. The check is site-wide, so one crawled page with a date signal passes it for the whole site. A dated blog post passes it even if every service page is bare. (The free Grade Check crawls only your home page, so it sees only what's there.) And the check never reads the date's value. A 2015 date passes the same as one from last week. It doesn't judge age, accuracy, or honesty. It answers one question: is there a machine-readable date on the pages we looked at?

If you want the rest of the Technical Health list, the audit tool checklist covers it.

Check it in five minutes

Open your page, view source, and search for four strings: <time datetime=, dateModified, datePublished, and article:published_time. If you find one in a JSON-LD block or in one of those tags (not an itemprop="dateModified" attribute, which the audit doesn't read), you're covered. Find none and only a visible date in plain text, wrap it or add dateModified. Find nothing at all, decide whether the page is one you actually keep current, and date it when you review it.

If you want to know whether a change like that moves anything, Bing gives you a free way to watch. In February 2026 they launched AI Performance in Bing Webmaster Tools, as a Public Preview. It reports citations across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations, with total citations, average cited pages, grounding queries, page-level activity, and visibility trends. Note your numbers before you change a date, look again a month later. That's your own before and after. It's one site and one engine family, so it won't prove cause, but it's real data about your pages instead of somebody else's ten.

Go look

The free Grade Check on aeocheck.net scans your home page and hands you a score and a letter grade. It doesn't break out individual checks, so use the view-source test above to see the date signal itself. Takes about a minute and costs nothing. Worst case, you learn your site has a "© 2019" and a one-point fix waiting.