Accuracy

Outdated AI information does not look outdated. It looks confident.

A model cannot tell you its knowledge has expired, so it writes about last year's prices and last release's API in the present tense, with the same certainty it uses for anything else.

Run it here

Find the claims with no date attached to them.

Full verification report
Freshness scanner

Words

0

Flags

0

Per 1,000

0.0

Reading

Nothing flagged

No term from this list appears in the text.

Paste a draft. The scanner flags claims anchored to a moment, unversioned product statements and recency adjectives, the four ways a page quietly expires. Counted in your browser. Nothing you paste leaves the page.

Models describe a world that stopped

Training data ends somewhere, and everything after that point is either missing or inferred. The model does not experience this as a gap. It answers in the present tense because the present tense is what the training text used.

So a draft can be entirely true of the world eighteen months ago and wrong about today, without a single internal inconsistency to warn you. There is no tense marker, no footnote, no hesitation.

The practical consequence: recency is not something to check after the fact, it is something to build into the claim. Anything that can change needs the date it was true written beside it.

The eight categories that go stale first

Some information has a short shelf life and always has. These are the categories to treat as volatile by default.

  • Prices, plan tiers and included limits, which change without announcement.
  • Product names, packaging and features that were renamed or retired.
  • Version numbers, API endpoints, function signatures and deprecation dates.
  • Legal thresholds, filing requirements and regulator guidance.
  • Availability by country, integration support and hardware compatibility.
  • Company structure: ownership, acquisitions, leadership, headcount.
  • Benchmarks and comparative performance claims, which competitors erode quickly.
  • Contact details, addresses, opening hours and support channels.

Newer is not automatically right

The opposite error is common enough to name: a page is updated with the newest figure and loses the qualification that made the old one usable. A revenue number without its period, a benchmark without its workload, a legal rule without its jurisdiction.

Freshness is a property of the whole claim, not of the numeral. Replacing 2023 with 2024 while keeping a footnote that described 2023 produces a page that is newer and wrong.

Update the source, the scope and the date together, or leave the claim and mark it as historical.

How staleness shows up in a draft

Unversioned present tense is the tell. Currently, right now, the latest version, modern teams, today's tooling, with no release or date anywhere near them.

Second tell: a product described as a monolith of features when it has since split into modules, or a service described as free when it now has a paid tier. The prose is fluent and the nouns are out of date.

Third: comparisons against competitors that have changed. The page still claims a gap that closed a year ago, which is the kind of staleness that damages a brand when a prospect knows more than the page does.

Date-stamping as a default habit

Two words fix most of this: as of, followed by a date. Prices, limits, version support, legal thresholds. It costs a line and it converts an absolute claim into a dated observation.

Add the source beside the date. The pair is what makes a claim maintainable: the next person can open the source, see whether it still says the same thing, and update the date if it does.

Reserve absolute present-tense claims for things that genuinely do not change: mathematical relationships, physical constants, definitions. Everything else gets a stamp.

Version pinning in technical documentation

Code documentation ages faster than any other format. An endpoint renamed in a minor release turns a helpful page into a source of failed builds.

Pin the version in the heading and the body: this page describes 4.x. Add a supported-versions table, and a short note on what changed in the last major release. Readers arrive with a version in mind, and the page should tell them immediately whether it applies.

Where behaviour differs by version, say so in the same paragraph as the instruction rather than in a changelog nobody reads. Documentation errors are expensive because they get copied into other people's code.

Refresh cadence by page type

Not every page deserves the same schedule. A pricing page needs checking monthly. A legal explainer needs checking when the rule changes, which requires a watch on the source rather than a calendar entry.

A conceptual guide can go years without edits if it avoids specifics. The moment it adds a version number or a price, it inherits the volatility of that fact.

Write the cadence down per template, and put the owner's name beside it. A refresh policy without an owner becomes a wish, and pages rot quietly in the gap between two teams that each assume the other is watching.

Finding staleness quickly

Search the page for the markers: currently, latest, new, now, currently supports, as of. Each hit is a claim that either has a date or needs one.

Then check the volatile specifics: every numeral, version, price and legal reference. One source each, reopened rather than remembered, because remembering is exactly what produced the staleness.

Finally, check the names. Renamed products and acquired companies leave pages describing a market that no longer has the same participants, and that is invisible unless someone looks at the entities deliberately.

Two audits: the page and the source

A page audit asks whether the page describes anything volatile without a date. A source audit asks whether the thing the page points at has changed since the date attached to it. The first is quick and the second is the one that actually keeps a page true.

Run the page audit whenever a template is reused, because a page cloned from last year's version inherits last year's facts along with its structure. Search for the year, the version number and the price, and replace all three or none.

Run the source audit on a schedule tied to the claim's volatility rather than the page's age. Pricing sources weekly, version support monthly, legal thresholds whenever the regulator publishes. Put a link to the source in the page's internal notes so the next person starts where the last one finished.

Two audits, two owners, two dates: the pair is what stops a page quietly becoming historical while still being presented as current. One audit alone leaves exactly the gap where the most expensive staleness lives.

What to do with genuinely old material

Sometimes the right answer is to keep the page and label it. An archived guide with a clear as-of date and a note that a newer version exists is useful and honest.

What damages trust is the silent mixing: a page updated in one paragraph and left alone in another, so the reader cannot tell which part is current. Pick a state for the whole page, current or historical, and make it visible.

If a page describes a product you no longer sell, redirect it or mark it retired. A stale page that keeps ranking is a permanent source of confused support requests.

The nine-second test

Ask what date this page describes, and whether that date is stated. If you cannot answer from the page itself, neither can a reader, and a reader who cannot date a claim will discount all of it.

That is the argument for treating freshness as part of accuracy rather than as housekeeping. A page is not correct in the abstract; it is correct about a world at a time.

What this covers

  • Volatile facts dated

    Prices, limits, versions and legal thresholds carry an as-of date and a source.

  • Versions pinned

    Technical pages name the release they describe and list supported versions.

  • Marker sweep

    Currently, latest, new and now searched for and resolved one by one.

  • Entities re-checked

    Product names, owners and competitor claims confirmed against today's market.

  • Refresh owner named

    Each page template has a cadence and a person responsible for it.

Questions this page usually raises

How old is too old for AI-generated information?
Age matters less than volatility. A definition can survive years; a price is stale the day it changes. Decide by category, not by a single age threshold.
Can I ask a model for the date of its cutoff?
It will give you an answer, and the answer is not reliable. Treat every claim about currency as unverified and check it against a live source.
Does web-connected AI fix outdated information?
It helps with facts available on live pages, and it introduces a new failure: citing a current page while describing an older state of it. The source still has to be read.
What is the fastest way to audit an old page?
Search for currently, latest, version numbers and prices, then reopen one source for each. Most staleness is concentrated in a handful of lines.
Should I always show an as-of date?
On anything volatile, yes. On definitions and concepts it adds noise. The test is whether the claim could be different next quarter.
What about pages describing a discontinued product?
Mark them retired or redirect them. A stale page that still ranks generates support requests and undermines the pages that are current.

Outdated AI information

Run the checks, then decide who has to sign it.

The automated pass is free, needs no account and uploads nothing. Escalate to a named reviewer only where being wrong is expensive.