Content freshness and AI Overviews: how old are the pages Google cites?

How fresh are the pages Google AI Overviews cite?

Mostly recent, measured by the date the page says it was last updated. Of the 195 cited web pages we could date (and that hadn't changed again since the capture), the median gap between that date and the day the AI Overview was recorded was 88 days. Half were updated within roughly three months. 87.2% were updated within a year.

The tail is short. Just 20 cited pages (10.3%) were last touched one to two years before capture, and only 5 (2.6%) were older than that. For B2B software queries, where products, prices and feature lists change every quarter, that shape makes sense: an AI Overview that names a tool's pricing or integrations has little reason to quote a 2023 page.

Counting each page once instead of once per citation barely moves anything: 189 unique dated web pages, median 89 days, 50.8% within 90 days. A handful of URLs are cited for several queries, but they don't skew the picture.

Are YouTube videos in AI Overviews older than web pages?

Yes, by a wide margin. YouTube supplied 152 of the 742 citations, and the median cited video had been uploaded 314 days before capture. Only 21 of 150 dated videos (14.0%) were under 90 days old, and 20 (13.3%) were more than two years old. The oldest was uploaded 2,343 days before the answer that cited it, a bit over six years.

A video has no "last updated" field you can bump, so this is the purest measure of age in the dataset. It tells you something useful: Google is happy to cite an older explainer or walkthrough when it still covers the question. In our earlier study of 132 B2B software queries, YouTube was the second-largest source of citations at 20.5%, so a fair share of AI Overview real estate goes to content that is far from fresh.

Which kinds of sites get cited with the freshest pages?

Software vendors. Their cited pages had a median of 64.5 days since the last update, and 63.9% (78 of 122) were updated within 90 days. Publisher and media pages came in at 113.5 days. Agency and consultant pages were the oldest web pages in the set at a median of 178 days, with only 6 of 33 (18.2%) updated in the past 90 days.

The vendor number partly reflects habit. SaaS marketing teams refresh "best X software" and "X alternatives" pages constantly, often adding the current year to the title. Agencies cited here tend to be cited for a specific explainer or a client story, which ages more slowly. Review and listing sites (G2, Capterra and similar) refused our fetcher on 33 of 37 citations (32 with a 403), so we can't say much about them; the same goes for Reddit and forums, where only 4 of 71 fetched threads exposed a machine-readable date.

Does the type of query change how fresh the cited pages are?

It does. "Best" and "alternatives" queries pull from fresher pages than head-to-head comparisons. Cited pages for "best" queries had a median of 79 days since the last update, and "alternatives" queries 78 days. For "X vs Y" queries the median was 148 days, and only 34.4% of cited pages were updated within 90 days.

The smaller rows (13, 12 and 19 citations) are directional at best. The gap between list-style queries and "vs" queries is the part we'd trust: a comparison of two mature products can stay accurate for months, while a ranked list goes stale the moment a new tool launches or a price changes. Google's own ranking systems guide describes "query deserves freshness" systems that show fresher content where people expect it, and list queries look like that kind of query.

Are cited pages fresher than the pages Google skipped?

No. This is the finding we'd put at the top of any slide. For 40 queries we compared the organic top 10 results that the AI Overview cited against the top 10 results it didn't. The cited group was, if anything, a little older: 35.0% updated within 90 days against 42.0% for pages left out, and a median of 165 days against 127.

The cited sample is small (20 dated pages), so we wouldn't read the reversed gap as "older pages win." What the data does rule out is the simple story that the freshest page in the top 10 gets the citation. Both groups were overwhelmingly updated within a year. Once a page is in that recent pool, something else decides whether Google quotes it, most likely how directly the page answers the sub-questions behind the query. Our query fan-out guide covers how those sub-questions get built.

Position inside the AI Overview doesn't track freshness either. The first three cited sources had a median of 85 days since update; sources in position 4 or lower had 89.

How do cited pages tell Google their date?

Mostly through markup, not visible text. Of the 265 cited web pages that exposed any date, 245 (92.5%) carried a dateModified value in their structured data or an article:modified_time meta tag. Only 89 of the 474 fetched web pages (18.8%) printed an "Updated" or "Last updated" date in the visible text that our parser could match.

That split matters because Google says it doesn't trust one signal alone. Its guidance on helping Google know the best date for a page asks publishers to show a clear visible date and also to use datePublished and dateModified in schema markup, and notes that Google weighs several factors because any one of them can be wrong. 170 of the 245 pages with a modified date in markup showed no matching label on the page itself.

For the full picture of which schema types cited pages carry, see our crawl of 458 AI-cited pages for schema markup.

Are AI Overviews citing new pages or old pages with new dates?

Mostly old pages with new dates. 168 cited web pages declared both a publish date and a later update date. Their median publish age was 237 days, while their median update age was 83.5 days. 95 of them (56.5%) were first published within a year of capture, but 151 (89.9%) had been updated within a year.

30 of these pages were first published more than two years before the AI Overview that cited them and then updated within the year. 23 were published more than a year earlier and updated within 90 days. That's the refresh playbook at work: an established URL with links and history, kept current.

Seer Interactive's July 2026 study of 7,683 pages cited by ChatGPT, Gemini and Perplexity found the same pattern from the other direction: 75% of cited pages were updated in the last year and 88% in the last two. Our AI Overview numbers for B2B software run even fresher on the update date (87.2% within a year).

There's a catch Google spells out. Its helpful content guidance asks whether you are "changing the date of pages to make them seem fresh" when nothing of substance changed. A date bump is easy to spot when the text is identical to the archived copy.

How quickly do cited pages change after they're cited?

Fast. 70 of the 265 dated cited web pages (26.4%) showed a modified date later than the day their AI Overview was captured, and 33 of those 70 had been modified in the seven days before our fetch. A quarter of the pages Google quoted had already changed again within weeks.

Some of that churn is real editing. Some of it isn't: 7 pages carried the exact date of our fetch as their modified date, which usually means a CMS stamps "today" on every request. We excluded all 70 from the age figures above rather than guess what version Google saw, which pulls our medians slightly older than reality.

What should you do with your own pages?

Use the data as thresholds, not slogans. We'd act on it like this.

1. List pages ("best", "alternatives", "top"): review anything not substantially updated in 90 days. In our data, 56.6% to 66.7% of cited pages for these query types cleared that bar. Add new products, remove dead ones, fix prices.
2. Comparison pages ("X vs Y"): update when either product changes pricing, packaging or a headline feature. The median cited page here was 148 days old, so a calendar-driven refresh isn't the point.
3. Explainers and videos: leave them alone while they're correct. Cited YouTube videos had a median age of 314 days.
4. Dates: ship both a visible "Updated" date and dateModified in markup, and change them only when the content changes. Check that your CMS isn't stamping today's date on every page view.
5. Year in the title: fine when the page really reflects the current year. It didn't separate cited from uncited pages in the top 10 (30.0% versus 39.4%), so don't expect it to do the work.

Then measure. Pick five buyer queries, note which pages the AI Overview cites today and their dates, and check again after your refresh. Our guide to ranking in Google AI Overviews covers the rest of the page-level work, and the citability grader checks whether a page answers clearly enough to quote.

How does this compare with other freshness studies?

It lines up on direction and differs on scale, mostly because of method. Ahrefs analyzed about 17 million citations across ChatGPT, Perplexity, Gemini, Copilot, AI Overviews and organic results in July 2025. It found AI assistants cite content that is 25.7% fresher than organic results on average (1,064 days versus 1,432), and that Google's AI Overviews and organic results lean toward older pages more than ChatGPT does.

Our median is far lower than Ahrefs' average because we measure something narrower: the date the page itself declares, on B2B software queries, where pages turn over quickly. Ahrefs used the date its crawler first saw each URL, across every topic. Averages are also pulled up by a long tail of very old pages that medians ignore. Seer's 75% updated-within-a-year figure for ChatGPT, Gemini and Perplexity sits between the two.

How we collected the data

Citations. We started from the 742 AI Overview citations in our AI Overview citation study: 150 B2B software queries in 10 categories (CRM, project management, marketing automation and others), of which 132 were measured and 122 showed an AI Overview. Each citation has the query, its position in the AI Overview, the cited URL, a source type, and the date the SERP was captured from Ahrefs' US SERP snapshots (28 August to 10 October 2026). For 40 of those queries we also had the organic top 10, which gave us 297 control rows: 84 top-10 results the AI Overview cited and 213 it did not.

Fetching. On 21 October 2026 a Python script requested every unique URL (931 in total) with a desktop Chrome user agent, a 25-second timeout and 24 requests in flight. 789 returned HTTP 200. Most failures were 403 responses from bot protection (123), concentrated on review sites and some publishers.

Dates. For each page we took the publish date from, in order: datePublished or uploadDate in JSON-LD on an article, web page or video object; any other JSON-LD publish date; the article:published_time family of meta tags; YouTube's upload date; the first <time datetime> element. The update date came from dateModified in JSON-LD, then article:modified_time or og:updated_time. Dates before 1995, after the fetch day, or update dates earlier than the publish date were discarded. The last update date is the later of the two.

Age. Age is the number of days between the capture date of the SERP and the page's last update date. If the page now declares a date after the capture, we can't know what version Google saw, so we flagged it as "changed after capture" and left it out of age figures. Unless stated otherwise, figures count citations, so a URL cited for two queries counts twice; the unique-URL figures are almost identical.

Other fields. "Year in title" is any year from 2010 to 2029 in the HTML title tag. "Visible updated date" means the page text matched a pattern such as "Last updated October 3, 2026" or "Updated: 10/03/2026".

Limitations

• We read the date a page declares, not the date it actually changed. Some sites stamp a new modified date without editing anything, and some edit without changing the date.
• The citation snapshots are from late August to early October 2026; the pages were fetched on 21 October. Pages that changed in between are flagged and excluded, which biases ages slightly upward.
• 44.1% of fetched web pages had no machine-readable date at all, Reddit threads above all. Review sites mostly blocked our fetcher. The age figures describe pages that publish dates.
• The sample is B2B software queries in the US. Freshness patterns for health, news, travel or local queries will differ, probably a lot.
• The cited-versus-not-cited comparison rests on 20 dated cited pages and 100 uncited ones from 40 queries. Treat the direction as solid and the exact percentages as rough.

Frequently asked questions

Dataset: content-freshness-ai-overviews-data.csv (1,039 rows: 742 AI Overview citations and 297 organic top 10 control rows). It is published under CC BY 4.0, so you can reuse the numbers as long as you credit AI Ranked and link here.