On-Page.ai Research
Websites Caught Lying About Their Last Updated Date: Does Google Care? A Study of 370 Page-One Results
Eric Lancheres · On-Page.ai Research|July 27, 2026
Download PDFAbstract
Site visitors check a page's “Last updated” date to decide whether it is worth reading, and moving the date without touching the content is common enough to have a name, date bumping. No one had published a number for how often it happens. We took the July 2026 snapshot of Google page one from our AI citation study (370 organic results for 50 keywords across 10 verticals) and reconstructed, from Internet Archive snapshots, what each page's content actually did. Of 89 page-one pages that claimed an update within the last year, 46% changed nothing but the date, 15% made cosmetic touch-ups, and 39% genuinely updated their content; these are floors, since the archived snapshot predates the claim by a median of 41 days. Only 40% of page-one results show a date on the SERP at all, and the median displayed date is 296 days old. 54% of page-one content has not materially changed in a year; 32% has not changed in two years. Health pages fake updates the most: 7 of 9 health pages claiming a recent update changed nothing but the date. One consumer-finance guide declared five different modified dates across 12 archive snapshots, every one stamped at exactly 11:00:50 UTC: the date field being edited without a real update. Google's AI Overview leaned toward fresher pages on every measure we tested, but no measure reached significance, and it cited date-bumped pages at the same rate as genuinely updated ones (59% vs 51%). Pages that genuinely updated scored higher on Information Gain than date-only updaters (median 58 vs 52). Google's own documentation says not to artificially freshen content, and Google engineers have described checksumming main content to detect unchanged pages.
Key findings
- 46% of page-one results that claim an update in the last year changed nothing but the date. 15% made cosmetic touch-ups. 39% genuinely updated their content. These are floors: the archived snapshot predates the claim by a median of 41 days, and any change in that window counts against the date-only class.
- Only 40% of page-one results show a date on the SERP at all, and the median displayed date is 296 days old.
- 54% of page-one content has not materially changed in a year. 32% has not changed in two years.
- Google's AI Overview leans toward fresher pages on every measure we tested, but no measure reaches significance. And it cited date-bumped pages at the same rate as genuinely updated ones (59% vs 51%): the AI cannot currently tell a real update from a moved date.
- Pages that genuinely updated score higher on Information Gain than date-only updaters (median 58 vs 52). Real updates and real originality travel together.
- Health pages fake updates the most: 7 of 9 health pages claiming a recent update changed nothing but the date. Developer docs are the most honest vertical (1 of 7).
- One consumer-finance guide declared five different modified dates across 12 archive snapshots, every one stamped at exactly 11:00:50 UTC: the date field being edited without a real update.
- Google's own docs say “don't artificially freshen” and Google engineers have described checksumming main content to detect unchanged pages. The date bump is a tactic Google both prohibits and can mechanically detect.
1. Why we measured this
SEOs may not take a “Last updated” date at face value, but site visitors do: it is one of the first things a reader checks to decide whether a page is worth reading. Google displays a date next to 40% of page-one results, and SEOs treat updating content as a ranking lever. That combination creates an obvious temptation: move the date, skip the work. The tactic is common enough to have a name, date bumping. We were wondering how often this happened and if it had any merit within the SERPs.
We started from the dataset of our July 2026 AI citation study [1]: a one-day capture of Google page one for 50 keywords across 10 verticals, with every organic top-10 result, the date Google displayed next to it, and the pages its AI Overview cited. Each ranking page also carried an Information Gain Score [2] from the same snapshot. To that we added the Internet Archive [3]: for every page claiming a recent update, an archived copy from before the claim shows whether anything actually changed.
Three questions drive the paper:
- How fresh does page one claim to be, and how fresh is it really?
- When a page claims a recent update, did anything actually change?
- Does Google care? We measure what its AI Overview does with date-bumped pages, and review what Google says about them in ranking.
2. Method
Corpus. The 2026-07-15 SERP snapshot from our AI citation study [1]: 370 page-one organic results for 50 keywords (10 verticals, 5 keywords each), plus 278 off-SERP pages cited by Google AI Overviews or ChatGPT for the same keywords. Platform pages (YouTube, Reddit, social, app stores) are excluded from content analysis: 476 article-class URLs remain, 362 of which we could fetch and extract at audit time. 72% had usable Internet Archive coverage.
Update claims. A page's claimed update date is the later of (a) the date Google displayed on the SERP at the snapshot and (b) the page's own declared modified date (article:modified_time, JSON-LD dateModified, og:updated_time, or a visible “Updated on ...” label). Claims older than 365 days are outside the census.
What actually changed. For every claiming page we pulled the latest Internet Archive snapshot at least two days before the claim, extracted the main content of both versions, and diffed them sentence by sentence. Near-matching sentence pairs whose only differing tokens are dates, years, or bare numbers count as date-token edits, not substantive change (“Best tools for 2025” becoming “Best tools for 2026” is not an update). The substantive change ratio is added plus removed plus revised sentences over the larger version's sentence count.
Classes. Under 2% substantive change: date-only. 2 to 10%: cosmetic. Over 10%: genuine. The result is not an artifact of the cut: at a strict 1%/5% split the date-only rate is 43%, at a loose 5%/15% split it is 54%.
Floors, not estimates. The pre-claim snapshot predates the claim by a median of 41 days. Anything the page changed in that window counts as change, which can only move pages out of the date-only class. The real date-only rate is at least what we report. Wikipedia-style pages whose dateModified moves on any minor edit are not driving the number either: excluding wikipedia.org changes the rate from 46.1% to 45.9%.
Real content age. Separately from claims, we compared the current content against snapshots roughly 1, 3, 6, 12, and 24 months old. The oldest snapshot still at least 85% similar to the current content is a floor on how long the content has been stable.
Citations. Within-keyword permutation tests (10,000 permutations, fixed seed) compare AI-Overview-cited vs uncited page-one results, so keyword mix cannot manufacture a result. ChatGPT cited only 4 page-one results in this corpus (consistent with our citation study [1]), so ChatGPT gets description, not tests.
3. Results
3.1 Page one barely claims to be fresh
Of 370 page-one results, Google displayed a date for 148 (40%). The median displayed date was 296 days old. Only 22% of dated results showed a date from the last 90 days.
How old the dates on page one are(148 dated results of 370, snapshot 2026-07-15)
Position does not order freshness. Granular medians by SERP position: position 1 sits at 219 days, position 2 at 160, position 3 at 337, position 6 at 596; there is no gradient from top to bottom. Formally, the rank correlation between position and displayed age is rho +0.02 with a within-keyword permutation p of 0.77 (120 dated results, 30 keywords). Verified content age agrees: positions 1 to 3 and 4 to 10 share an identical 365-day median stability floor. The fresher page does not sit higher on page one. Two scope notes: every page here already ranks, so this says nothing about what freshness does for pages trying to reach page one, and this keyword set leans evergreen, the query class where Google's own freshness systems say recency should matter least.
Does position order freshness?(median displayed age by SERP position)
By vertical, medians run from 98 days (personal finance) and 113 days (travel) to 422 days (health) and 567 days (home improvement), charted in Figure 8 alongside the date-only rates. The freshest-looking verticals are the ones where freshness is a ranking argument; health pages on page one are typically more than a year old by their own labels.
3.2 The update-claim census
89 page-one pages claimed an update within the last year and had a usable pre-claim snapshot. What the claimed updates actually were:
| Update class | Substantive change | Share | Pages |
|---|---|---|---|
| Date-only | under 2% | 46% | 41 |
| Cosmetic | 2 to 10% | 15% | 13 |
| Genuine | over 10% | 39% | 35 |
What claimed updates actually were(one square per page; 89 update claims)
The date-only share is not an artifact of the cut(date-only rate under each class threshold)
Restricted to claims under 90 days old, the date-only share is 40% (18 of 45).
The receipts
Two side-by-side pairs, live page vs its own Internet Archive snapshot:
The metronome
A major consumer-finance site's savings-account guide, reconstructed from 12 archive snapshots between May and July 2026, declared five different modified dates: May 15, June 15, June 22, July 13, July 20. Every one is stamped at exactly 11:00:50 UTC. A real editorial update re-stamps the time of day; editing only the date field in a typical CMS preserves the original time. A frozen seconds-stamp across five updates is the tell of a date-only edit, whether automated or by hand: the date moved, the content did not. Between the July 16 snapshot and the audit, 134 of its 137 content sentences are identical; the other 3 changed “as of July 10, 2026” to “as of July 17, 2026”.
Five modified dates in three months, one exact timestamp
A major consumer-finance site's savings-account guide, 12 Internet Archive snapshots, May to July 2026. Every declared date is stamped at exactly 11:00:50 UTC.
3.3 Who fakes freshness the most
Date-only rate by vertical, census population (small per-vertical n, ranked descriptively, n shown):
Date-only rate by vertical(census population; small per-vertical n, ranked descriptively)
How fresh each vertical looks by its own labels(median SERP-displayed age in days; freshest on the left)
Health, the vertical where stale advice does the most damage and where page one already looked oldest by its own labels (median displayed age 422 days), fakes its updates the most: 7 of the 9 health pages claiming a recent update changed nothing but the date. The marketing and SEO vertical, the people who invented the tactic, run at 50%. Developer documentation is the most honest page one we measured.
3.4 Page one is older than it looks
Across 209 page-one URLs with usable snapshot ladders: 78% have content that has not materially changed in 3 months, 54% in a year, 32% in two years.
How long page-one content has provably stayed the same(209 page-one URLs with usable snapshot ladders)
Among pages that displayed a date under 90 days old, 35% had content unchanged for at least 6 months and 30% unchanged for at least a year (n=20, small; reported as-is).
3.5 Does freshness earn AI Overview citations?
Direction: yes on every cut. Significance: no on any cut.
- Cited pages show a SERP date more often (45% vs 35%).
- Cited pages' displayed dates are fresher: median 250 vs 346 days. Within-keyword delta is 168 days fresher, p=0.107.
- Declared modified dates: median 146 vs 223 days, p=0.441.
- Provable content age: 66 days fresher for cited pages, p=0.207.
AI-Overview-cited pages show fresher dates(median SERP-displayed age; cited n=80, uncited n=68)
Then the kicker. Among pages with an update claim, the AI Overview cited:
| Update class | AIO citation rate | n |
|---|---|---|
| Date-only | 59% | 24 of 41 |
| Genuine | 51% | 18 of 35 |
| Cosmetic | 46% | 6 of 13 |
The date bump performed as well as the real update. Whatever freshness signal the AI Overview responds to, it is not reading what changed.
One place freshness clearly shows: when the AI Overview cites pages from outsidepage one, those pages are fresher (median declared age 108 days vs 188 for page one). The overview's escape hatch skews recent.
3.6 Real updates and originality travel together
Pages whose claimed update was genuine score a median 58 on Information Gain [2]; date-only updaters score 52, cosmetic 51. Content stability alone shows no IG difference (52 vs 53): being old is not the problem, faking new is.
4. Google already told publishers not to do this. And it can tell.
None of this is a loophole Google is unaware of. Google's own documentation is explicit:
“don't artificially freshen a story without adding significant information or some other compelling reason for the freshening”
The same rule appears in the News Publisher Center best practices [5]. John Mueller put it more bluntly in 2022:
“Changing the date without doing anything else is just noise and useless.”
Detection is not hypothetical either. Everything in this study was measured with public archive snapshots; Google has something better, its own crawl history of every page. Google engineers have described computing a checksum (a fingerprint of the main content, boilerplate removed) for every crawled page to detect duplicate and unchanged content [7]. A page whose checksum has not changed is a page Google can mechanically identify as unchanged, no matter what its dateModified says. Google's crawl-scheduling documentation says recrawl frequency follows how often a page actually changes, and its documented freshness systems boost recency only for queries that deserve it [8].
What this means in practice, separating what is documented, what we measured, and what practitioners observe:
- Documented by Google: fake freshening violates its date guidelines, and unchanged content is mechanically detectable. A page that only moves its date should expect no freshness-driven ranking benefit; substantial changes are what trigger reprocessing and rescoring.
- Measured here: the date bump currently works on two audiences that apparently do not run that check, Google's AI Overview (date-bumped pages cited at 59% vs 51% for genuine updates) and the date Google displays on the SERP.
- Practitioner observation, labeled as such: a fresher displayed date can raise click-through rate, and user behavior feeds ranking. So a bumped date may still buy clicks even if it buys no freshness credit. That benefit rides on the SERP display, not on the update itself, and it is exactly the loop Google's guidelines tell publishers not to run.
This study measured citations and content change, not ranking movement. Nothing here shows date bumps moving Google rankings in either direction, and the documented mechanics above predict they should not.
5. What to do with this
Split the decision by what you are optimizing for.
If you are after visibility in AI answers, the uncomfortable finding is that the date bump appears to work.
In our census, Google's AI Overview favored fresher-looking pages on every measure: the pages it cited showed a median displayed date of 250 days old versus 346 for the pages it passed over, and a median declared update of 146 days versus 223. ChatGPT's citations, which rarely overlapped page one at all, skewed the same way: the pages the assistants cited from beyond page one were the freshest group in the study, at a median declared age of 108 days versus 188 for page one. Even Google's own AI is taking the date at its word. A fresher date on the SERP can also lift click-through rate on its own.
If you are after Google rankings, the calculus flips. Google tells publishers not to artificially freshen, it detects unchanged content mechanically (a checksum comparison against its own crawl history, far better evidence than the public archive this study used), and reprocessing follows real change. A moved date on an unchanged page should expect no freshness credit in ranking. If you want that credit, make a substantial update: a real section, new data, corrected claims. Not a date string.
Whichever audience you optimize for, the genuine updaters in this census carried more original information (median Information Gain 58 vs 52), and unique information is the strongest correlate of originality scores we have measured [2]. The date is not the update. And a real update no longer costs more than a fake one: it takes a few seconds to run something like our light page refresh to perform a real refresh, one that actually helps users and keeps the page updated. Same effort as moving the date, but the page gets new information researched online.
6. Limitations
- Wayback coverage gates the census: 72% of article-class URLs had usable snapshots; coverage skews toward better-known sites.
- Sentence diffs are lexical and operate at sentence granularity. Changes inside short table cells or widgets (prices, rates) below the 30-character sentence floor do not count as substantive; a full paraphrase rewrite would count as genuine. Both push the date-only rate down, not up.
- Current-page fetches ran 11 to 12 days after the SERP snapshot.
- One-day snapshot, 50 US keywords, informational-leaning mix: the same scope caveats as the two studies this corpus inherits from [1, 2].
- Citation analysis is observational; within-keyword tests control for keyword mix, nothing controls for everything.
References
- Lancheres, E. (2026). “Do AI Assistants Cite Original Content? 793 AI Citations Measured Against the Pages That Rank.” On-Page.ai Research. api.on-page.ai/research/ai-citation-study
- Lancheres, E. (2026). “Information Gain in Google's Top-Ranking Results: A Measurement of 150 Pages.” On-Page.ai Research. api.on-page.ai/research/information-gain-study
- Internet Archive Wayback Machine. web.archive.org
- “Help Google Search know the best date for your web page,” Google Search Central Blog (2019): the “don't artificially freshen” rule. developers.google.com/search/blog/2019/03/help-google-search-know-best-date-for
- Google Publisher Center best practices (the same rule for Google News). support.google.com/news/publisher-center/answer/9607104
- John Mueller (Google), 2022 public statement: “Changing the date without doing anything else is just noise and useless.”
- “Google Goes Deep On Dupe Detection & Canonicalization,” Search Engine Roundtable (Search Off the Record recap): main-content checksums. seroundtable.com/google-dupe-detection-canonicalization-30376.html
- “A Guide to Google Search Ranking Systems,” Google Search Central documentation (freshness systems). developers.google.com/search/docs/appearance/ranking-systems-guide
Data and metric availability
Aggregate statistics are reported above; per-URL diffs and scores are not published. SERP payloads and citation labels come from the July 2026 AI citation study capture [1]. Information Gain Scores come from the On-Page.ai scan API; the metric's calibration is documented in the information gain study [2]. Any page can be scored against its live ranking cohort with the free checker or programmatically via the On-Page.ai API. Verified content-change tracking is on our roadmap; the Information Gain checker is live today.
Cite this study: Lancheres, E. (2026). The Freshness Audit: Websites Caught Lying About Their Last Updated Date. A Study of 370 Page-One Results. On-Page.ai Research. https://api.on-page.ai/research/freshness-audit