What Share of AI Overviews Answers Cite a Source Published in the Last 30 Days? (2026 Data)

27.1% of Google AI Overviews answers cite a source published in the last 30 days. That figure rises to 37.1% (36 of 97) if you only count the answers where a reference was actually captured — the other 36 of 133 AIO answers in our sample returned an AI Overview with no attributable source at all. We captured Google AI Overviews for 160 commercial queries across 8 industries, pulled every cited URL, and fetched each one to extract its datePublished and dateModified. The median cited source is 8 months old by original publish date — but just 3.2 months old by last-modified date, because AI Overviews leans heavily on pages that get refreshed often even when their original publish date is years old.

Last updated: July 2026. First-party data — full methodology below.

How fresh are AI Overviews' cited sources, really

We ran the numbers two ways, because "freshness" means something different depending on which date field you measure — and AI Overviews' cited pages tell a genuinely different story by each one.

Under 30 days old

Published

11.5%

Modified

25.4%

Under 90 days old

Published

27.1%

Modified

48.4%

Under 1 year old

Published

59.1%

Modified

85.3%

By original publish date (datePublished) — 384 dateable citationsBy last-modified date (dateModified) — 279 dateable citations

Share of dateable citation instances under each age threshold. Median age: 250.1 days by publish date vs. 96.3 days by modified date — AI Overviews leans on pages that get refreshed often, even when the original publish date is old. LumenGEO first-party study, Google AI Overviews, 160 queries, captured 2026-07-02.

Read the two columns side by side and the pattern is unmistakable: at every threshold we measured, "last modified" freshness runs higher than "originally published" freshness — most sharply at 30 days, where the modified-date rate (25.4%) is more than double the publish-date rate (11.5%). Only 40.9% of dateable citations by publish date are over a year old — but by modified date, that drops to 14.7%. The mean age (585.3 days) is heavily skewed by a long tail of genuinely old evergreen pages still getting cited, some with 2014-2017 publish dates; the median (250.1 days) is the more representative "typical" figure for a page in our sample.

Why published and modified dates tell different stories

The gap between 250.1 days (median publish age) and 96.3 days (median modified age) is not measurement noise — it is the single most useful finding in this dataset. Many of the pages Google AI Overviews cites are not new content. They are old URLs — sometimes years old — that get substantively updated on a recurring cadence, and the update is what keeps them in the cited pool.

This lines up with what we found in our earlier study of why AI citations decay: AI-cited content has a short measured half-life, and the winning defense is a maintenance cadence, not a constant stream of net-new URLs. This freshness dataset is the AI-Overviews-specific evidence for the same mechanic — a page's dateModified field is doing real work in what gets cited, even when datePublished is old.

By industry: who gets the freshest citations

Freshness is not evenly distributed. Some categories reward recency far more than others — and the spread between the leader and the laggard is large enough to change how you'd plan a content-refresh calendar by vertical.

*Home services has only 9 AIO-present answers and 7 total citation instances at 28.6% date coverage — too sparse to treat as more than directional; the 0% figure is likely an artifact of the small sample, not a real freshness gap.

SaaS/software stands well apart: 63.2% of its AIO answers surfaced a source published in the last 30 days, roughly triple Marketing/SEO's 20.0% among the industries with a reasonably-sized sample. That tracks with the categories' behavior — a plausible explanation is that SaaS pricing, feature, and comparison queries change fast enough that Google's AI favors recently-published pages, while marketing/SEO informational content skews toward evergreen guides. Finance/fintech shows the oldest typical source by publish date (510 days, ~17 months) despite decent date coverage (57.6%) — high-stakes categories lean on established, slower-moving sources rather than the newest thing published.

Why 40% of citations don't have a usable date

Of 699 unique URLs cited across our 133 AIO-present answers, 420 (60.1%) yielded an extractable datePublished or dateModified — above the 40% honesty-gate threshold we hold ourselves to before reporting a rate as if it applies broadly. The remaining 39.9% splits into two structurally different buckets:

• 96 URLs (13.7%) failed to fetch outright. 80 of those were HTTP 403 bot-blocks, concentrated on a handful of large, well-known publishers: Forbes, Investopedia, PCMag, Quora, Serious Eats, and Trustpilot. These sites are cited by AI Overviews but actively block the kind of automated fetch our extraction pipeline uses — so their real freshness is unknown to us, not old.
• 183 URLs (26.2%) fetched fine but carry no byline date at all. This bucket is dominated by Reddit threads (85 URLs, 0% dated) and Google SERP/shopping result fragments (35 URLs); the remainder is a long tail of small-count domains, including Facebook (4 URLs) and Amazon (3). This is less an extraction failure than a feature of what AI Overviews cites: forum threads and search fragments simply don't carry a datePublished meta tag the way an article does.

Neither bucket should be read as "old." A 403-blocked Forbes page could be five days old; we cannot tell. Our 27.1% / 37.1% headline and the age-bucket percentages describe the 60.1% of citations we could actually date — the true rate for the rest is unknown, and any tool reporting one clean number without naming this gap is glossing over almost 40% of its own sample.

Method

Built to be reproducible, with every judgment call stated plainly.

• Engine and window: Google AI Overviews, captured for 160 US commercial queries spanning 8 industries (Marketing/SEO, SaaS/software, Finance/fintech, Ecommerce/retail, Travel, Healthcare/wellness, Legal services, Home services), all fetched the same day, 2026-07-02.
• AIO presence: 133 of 160 queries (83.1%) returned a visible AI Overview; the remaining 27 are excluded from the freshness analysis (no AIO, no citations to date).
• What we counted: every reference URL the AI Overview displayed — 701 citation instances across 699 unique URLs (some URLs were cited more than once across different queries).
• Date extraction — a 3-tier fallback: for each unique URL we fetched the live page and tried, in order, (1) JSON-LD structured data (datePublished/dateModified) — the source for 268 of 420 dated URLs, (2) meta tags (Open Graph article:published_time, etc.) — 112 URLs, and (3) a visible-text regex looking for a byline-style date on the rendered HTML — 40 URLs, the least reliable tier. We ran no DOM/JS-rendering library, so a page that only renders a date via client-side JavaScript shows as undateable even if a date exists.
• Two caught errors, disclosed: the visible-text tier — the least reliable of the three — produced two confirmed false positives, both nulled before analysis. The first: a Semrush tool page whose regex match read as a publish date more than three months in the future (2026-10-13, against the 2026-07-02 capture date). The second: royalcaribbean.com's homepage, whose regex match read as the day after the capture date (2026-07-03). Neither was caught by the codified extraction bound, which only rejects a matched year after 2027 (see extract-dates.mjs) — both were caught in a separate manual cleaning pass. Treat any single visible-text-sourced date with more skepticism than a JSON-LD-sourced one.
• Two denominators for the headline number, reported together, not collapsed: 27.1% treats the 36 of 133 AIO answers with zero captured references as "no recent source" (conservative). 37.1% excludes those 36 and asks the question only of the 97 answers where a reference was actually captured. We can't tell whether a zero-reference answer reflects a genuinely sourceless AI Overview or a gap in our own scrape of Google's UI, so we report both instead of picking one.
• Age math: every "age" is computed against the fixed capture date, 2026-07-02, not the date you're reading this — reproducing this analysis later against the same raw data returns identical numbers.

Limitations

• Date coverage is 60.1% (420/699 URLs) — above our 40% honesty-gate threshold, so we're not required to strictly reframe every number as "of dateable sources only." Still, every age-bucket percentage here is implicitly a rate among dateable citations unless stated as an answer-level metric (27.1% and 37.1% are answer-level and already account for undated references).
• The undateable 39.9% is not one thing. 96 URLs (13.7%) failed to fetch, mostly HTTP 403 bot-blocking on major publishers. 183 URLs (26.2%) fetched fine but structurally carry no date — dominated by Reddit and Google SERP fragments. The second bucket is a feature of what AI Overviews cites, not a tooling gap.
• Denominator ambiguity in the headline. 36 of 133 AIO-present answers (27%) had zero references captured by our scraper — we can't distinguish "genuinely no visible sources" from "our scrape missed them," so we report both the 27.1% and 37.1% figures.
• The mean is not the typical value. Mean age by publish date (585.3 days) is pulled upward by a long tail of older evergreen sources, some dated 2014-2017, still being cited today. The median (250.1 days) is the more representative figure.
• Published and modified dates measure different things — any freshness claim needs to specify which one it means, and this piece has tried to be explicit every time.
• Per-industry breakdowns for Home services (9 answers, 7 citations, 28.6% date coverage) and Legal services (20 citations) are small-n — directional only.
• Single-snapshot capture, one engine. All fetches happened the same day (2026-07-02) against 160 US commercial queries — not a statistically representative sample of all Google AI Overviews traffic, and Google AIO only, not ChatGPT, Perplexity, or other AI search engines. Treat the directional finding (modified-date freshness roughly doubles publish-date freshness) as durable and the exact percentages as a July-2026 snapshot. For how citation sets behave over time more broadly, see our ChatGPT domain-authority study and Perplexity's Reddit-citation breakdown — first-party snapshots with the same discipline about what a single measurement can and can't tell you. Our live citation tracker shows how these numbers move as we keep sampling.

What this means for GEO

• Stop optimizing for "newly published." Only 11.5% of dateable AIO citations were published in the last 30 days. Chasing a constant stream of net-new URLs is the wrong lever for most categories.
• Start optimizing for "recently substantively updated." 25.4% of dateable citations were modified in the last 30 days — more than double the publish-date rate. A real refresh (new data, a rewritten section, updated numbers) on a year-old page is a stronger freshness signal than a thin new page you never touch again. Update the visible dateModified timestamp together with a genuine change — see Google AI Overviews optimization for the fuller playbook and why AI citations decay for the maintenance cadence that holds citations once you have them.
• Weight your refresh calendar by category. SaaS/software rewarded recency at 3x the rate of Marketing/SEO — in a fast-moving vertical, a monthly or quarterly refresh cadence is defensible. In slower-moving, authority-driven categories like finance or legal, a well-maintained but less frequently updated page still gets cited.
• Know what you can't measure. Roughly 40% of the pages AI Overviews cites — bot-blocked publishers, Reddit threads, SERP fragments — don't expose a usable date. Don't assume a source is old just because you can't verify its date; plan your own site to be the opposite: dated, structured, and fetchable.

FAQ

What share of Google AI Overviews answers cite a source published in the last 30 days?

In our July 2026 first-party study of 160 commercial queries, 27.1% of AI Overviews answers (36 of 133 AIO-present answers) cited at least one source published within the last 30 days. That rises to 37.1% (36 of 97) if you only count answers where a reference was actually captured by our scrape — the other 36 of 133 answers returned an AI Overview with no attributable source in our capture, and we can't tell whether that reflects a genuinely sourceless answer or a gap in the scrape itself.

How old is a typical source cited by Google AI Overviews?

By original publish date, the median cited source in our sample is 250.1 days old — about 8 months. The mean (585.3 days, ~1.6 years) is much higher because a long tail of older evergreen pages, some dated 2014-2017, are still getting cited; the median is the more representative "typical" figure. Measured by last-modified date instead, the typical cited source looks far fresher: a median of 96.3 days, about 3.2 months.

Does Google AI Overviews prefer newly published content or recently updated content?

Recently updated, based on our data. Only 11.5% of dateable citations were published in the last 30 days, but 25.4% were last modified in the last 30 days — more than double. Modified-date freshness was higher than publish-date freshness at every threshold we measured (30, 90, and 365 days), though the gap narrows at longer horizons: about 1.8x at 90 days and 1.4x at 365 days. This suggests AI Overviews' cited pool leans on pages that get substantively refreshed on a recurring cadence, even when their original publish date is old, rather than requiring a constant stream of brand-new URLs.

Which industries get the freshest AI Overviews citations?

SaaS/software had the highest rate of recent-source answers in our sample at 63.2% — roughly 3x Marketing/SEO's 20.0% among the industries with a reasonably-sized sample (both had a well-sampled 18-20 AIO-present answers). Finance/fintech had the oldest typical cited source by publish date at a 510-day median (~17 months), consistent with high-stakes categories leaning on established, slower-moving sources. Home services showed 0%, but with only 9 answers and 7 total citations at 28.6% date coverage, that figure is too small-n to trust as a real finding.

Why don't all AI Overviews citations have a traceable publish date?

Of 699 unique URLs cited in our sample, 60.1% (420) yielded an extractable date. The remaining 39.9% splits into two different problems: 96 URLs (13.7%) failed to fetch, mostly HTTP 403 bot-blocks on sites like Forbes, Investopedia, PCMag, Quora, Serious Eats, and Trustpilot, so their real freshness is unknown to us rather than old. The other 183 URLs (26.2%) fetched fine but structurally carry no byline date — dominated by Reddit threads and Google SERP/shopping fragments, where a datePublished field simply doesn't exist to extract.