A statistic is the most persuasive sentence in an article and the first one to expire. “68% of marketers say…” lands with authority the day you publish it. Two years later the study has been replaced and the source page has moved. Your article still says 68% in the present tense, quietly turning into the kind of page nobody should trust.
Key Takeaways - Statistics age faster than any other content element, and their sources disappear: 38% of pages from 2013 were gone a decade later
- The cost is trust first, rankings second. Google's raters lower their trust assessment when they notice factual inaccuracies
- AI systems amplify the damage: an outdated number they cite becomes an outdated answer, and pages that produce wrong answers get cited less
- The fix is a routine, not a rewrite: inventory every number, link it to a primary source, and check it on a schedule or with monitoring
Why do statistics expire faster than everything else?
Because a statistic makes three promises at once, and each can break on its own. The number is current. The source still exists. And the source still says that number. Most other content elements make only one.
The evidence on how fast the promises break is uncomfortable. A Pew Research Center analysis found that 38% of web pages that existed in 2013 were no longer accessible in 2023. Even for pages from 2023, 8% were already gone within the year. Every one of those pages was somebody’s source.
A Harvard Law School study of links in New York Times articles, published in the Columbia Journalism Review, put numbers on the decay curve. 25% of all deep links were completely inaccessible, and the rate rose with age: 6% for links from 2018, 43% for 2008, 72% for 1998. More than half of all articles with deep links contained at least one dead one.
Columbia Journalism Review Harvard Law School study of New York Times links, 2021 Linkrot became more common over time: 6 percent of links from 2018 had rotted, as compared to 43 percent of links from 2008 and 72 percent of links from 1998. Fifty-three percent of all articles that contained deep links had at least one rotted link.
The same study names the quieter problem: content drift, “the typically unannounced changes, retractions, additions, replacement, to the content at a particular URL”. The link works, but the page no longer says what you cited. Your reader clicks through and finds a different number, or none.
So a statistic in a three-year-old article carries three failure modes stacked on each other. The research says at least one has probably fired.
What does an outdated statistic actually cost?
Trust first, then citations, then rankings, in that order and with that lag. The ranking loss is the last thing you see, which is why most teams only notice the problem after the damage is done.
Cost 1: trust, with every reader who checks
Google’s rater guidelines are explicit that accuracy feeds directly into trust, the most important E-E-A-T element:
Google Search Quality Rater Guidelines, September 2025, section 3.1 While examining the quality of the MC, you may notice factual inaccuracies that lower your assessment of Trust.
Google’s helpful-content self-assessment asks the same thing in plainer words: does the content have easily verified factual errors? A 2022 statistic presented as current in 2026 is exactly that.
The trust cost is not abstract. A reader who checks one number and finds it stale stops checking and starts doubting. Everything else on the page, including the parts that are still correct, now reads as unreliable. We explained the mechanism in E-E-A-T explained: trust is a multiplier, and a single visible error pulls the whole page down.
Cost 2: citations, human and machine
Journalists, other writers and AI systems all look for numbers to cite, and all of them prefer numbers they can verify. A statistic with a dead source or a superseded study fails the check. The page does not get cited, the fresher competitor does, and the authority that citations build accrues elsewhere.
For AI answer engines the effect compounds. When a system cites your page and repeats the outdated number, its answer is wrong. Systems that are evaluated on answer quality learn, one way or another, to prefer sources that do not make them wrong. We looked at how they select sources in how AI search engines choose which sources to cite; currency and verifiability were two of the five factors.
Cost 3: rankings, slowly
Rankings follow. A page that loses trust and citations loses the signals that kept it on page one. A fresher page takes the query. This is content decay in its most common form. No penalty, no algorithm update. Just a page that stopped being the best answer while nobody was looking.
The order matters for diagnosis. If a page’s clicks are falling, the outdated statistics are usually the cause, not the symptom. Fixing the numbers and their sources often restores the page without touching anything else, because the structure and the intent match were never the problem.
Which statistics go stale first?
The ones tied to a year, a survey or a price. In rough order of decay speed:
- “Latest” survey results. Annual industry reports are replaced every year. The 2024 edition is superseded the day the 2025 edition appears, and your article still quotes 2024.
- Market sizes and forecasts. Analyst forecasts get revised, sometimes drastically. A “will reach $X billion by 2027” claim from 2023 is often wrong by 2025.
- Platform and usage numbers. User counts, market shares, adoption rates. They move every quarter.
- Prices and plan limits. Not statistics in the strict sense, but numbers readers act on. They change without announcement.
- Performance benchmarks. Load times, conversion rates, “average” anything. The underlying reality and the measurement methods both shift.
- Percentages from single studies. Especially small studies that get re-run with different results. The number lives on in articles long after the study was contradicted.
- Demographic and behavioural stats. Slower, but they do move, and they are often quoted from sources that are themselves quoting older sources.
The slowest to decay are physical constants, historical facts and definitions. Almost nothing else in a marketing article is safe for more than a year.
Richard Roth Founder CiteRanger, SEO Strategist The pattern I see most often is second-hand statistics: an article cites a blog that cites a report that cites a survey from 2019. The number has been copied so many times that nobody remembers it is seven years old, and the original source is a 404. When I audit a site, I trace every number back to its origin, and roughly a third of them do not survive the trip.
How do you keep the statistics in your content alive?
By treating every number as a maintained asset with an origin, a date and a review schedule, instead of as a sentence you wrote once. Five steps:
1. Inventory every number
Go through your articles and list every statistic. Into the table go the number, the claim it supports, the page it lives on, the source URL and the source date. For a site with fifty articles this is an afternoon. For five hundred it is a job for a tool. The fact check collects the numbers, the sources and the dates for you.
2. Trace each one to a primary source
Second-hand statistics are the most common failure. Follow the chain until you reach the original study, report or dataset, and link to that. If the chain ends in a 404 or on a page without the number, the statistic is unsupported. Then it gets replaced or cut.
3. Date the number in the text
“According to Pew Research Center (2024), 38% of…” ages honestly. “38% of pages are gone” ages silently. Readers and raters can judge a dated number; they cannot judge an undated one. This is the same principle as honest update dates, which Google’s guidance on dates asks for at page level.
4. Set a review cadence by statistic type
Annual survey numbers: review when the next edition is due. Prices and platform numbers: quarterly at least. Everything else: on the page’s normal cadence, which we laid out in how often to update content.
5. Monitor the sources, not just the pages
A source can disappear or change on any day. Scheduled reviews catch it eventually; monitoring catches it when it happens. For time-sensitive pages that matter, the difference between “eventually” and “now” is the difference between a quiet correction and a visibly wrong page.
When you replace an outdated statistic, leave a trace: a short note like “Updated September 2026: figure revised from the 2023 to the 2025 edition of the report” either inline or in a change log at the end of the article. It shows readers and raters that the page is maintained, and it stops the next editor from wondering why the number changed.
What should you do with a statistic you cannot verify?
Remove it, or rewrite the claim without the number. An unverifiable statistic is a liability, not an asset, and the claim it supported is usually still valid in a weaker form.
“Studies show that 73% of buyers read reviews before purchasing” with a dead source becomes “Most buyers read reviews before purchasing”. Better still would be a fresh number from a current study with a working link. The second version is less punchy and more trustworthy, and trustworthy is what the page is for.
The temptation is to keep the number because it is impressive. Resist it. Anyone who catches one unsupported number assumes the rest are unsupported too. And is right often enough.
Conclusion
Outdated statistics are the most expensive kind of stale content because they are the most trusted sentences on the page. When they expire, they take the page’s credibility with them, then its citations, then its rankings. The research on link rot and content drift suggests as much. In any article a few years old, at least one number has probably broken already.
Inventory your numbers, trace them to primary sources, date them in the text, review them on a cadence and monitor the sources in between. It is unglamorous work, and it is the difference between a site that can be cited and one that cannot.
Finding those statistics by hand is the expensive part; the fact-checking tool extracts every figure from a page and checks it against current sources.