Outdated knowledge base articles detector for help centres
An outdated knowledge base articles detector should find the help articles that are wrong, not just the ones that are old. Veritas reads public help-centre pages and flags claims that outside evidence shows are out of date.
Key takeaways
- Review dates and view counts show which articles are old or quiet. They don't show which articles are wrong.
- Some help-article claims concern your own product; others concern the outside world: third-party versions, laws, standards and dates.
- Veritas checks claims against evidence outside your site and flags the ones that are outdated. It doesn't use your own pages as proof, and can't see pages behind a login.
Help articles go wrong in a particular way. Nobody edits them, and then the product changes: a plan limit moves, a setting is renamed, a policy is updated. The article still reads well and is now incorrect, and customers (and any AI assistant that reads it) act on it.
This guide separates stale from wrong, shows how teams catch stale articles today from the tools' own documentation, explains why wrong articles matter more once AI assistants answer from them, and sets out a routine and a clear statement of what Veritas can and can't reach.
Which "knowledge base" this page is about
The term covers three different things, and search results mix them.
Customer-facing articles on your own domain or a help-desk subdomain. This page is about these.
Team documentation in a tool such as Notion or Confluence, usually behind a login. The tools' own verification features are built for these.
The developer and academic meaning, where "checking a knowledge base" means linting markdown or testing rules. Search for the plain phrase "knowledge base accuracy checker" and most results are of this kind.
If you run a public help centre and want to find articles that are outdated, keep reading.
What goes stale in a help centre
Help articles describe a product, and products change faster than articles do. Some of what goes stale concerns your own product, and some concerns the outside world.
Seats, storage, usage caps and what each plan includes.
A setting moved, a feature renamed, a button relabelled.
Refund windows, support hours, who to email, which form to use.
A checker that looks outside your site can't settle these, because only your own pages know the answer. Reviews triggered by each release are the right tool, and we say more on that below.
Supported browsers, operating systems and versions, and steps that depend on someone else's product.
Statements about regulations, data-protection rules or standards that change over time.
Statistics, deadlines and public announcements your article quotes or links to.
These are claims an outside source can settle, which is what Veritas checks.
Stale is not the same as wrong
One vendor guide on this topic makes a useful distinction. An article is stale when its metadata says it is old. It is wrong when its content no longer matches reality.
The two overlap less than you would hope. A rarely viewed article from three years ago can still be correct. An article updated last month can contain a plan limit that changed last week. Review dates catch articles at a point that has little to do with accuracy.
That is the guide's point, and it comes from a vendor that sells a competing product, so take it as framing rather than data. It matches how the tools' own documentation describes their features, which we look at next.
How teams catch stale articles today
Three tools show the common pattern in their own documentation:
A published article becomes "stale" when its review due date passes. Reminders can be set per article, in bulk or by named schedule, with assigned reviewers. Its examples include pricing, API references, compliance and product features.
Assigned verifiers confirm cards on a schedule. Unverified cards go into the verifier's queue but stay searchable and visible, and overdue cards may be deprioritised in AI answers. Guru recommends reviewing at least yearly.
A page can be marked verified for a set time or indefinitely. When verification expires, the owner is notified, and verified pages are prioritised in search and AI answers.
These are workflow tools. They trigger on a date, an owner or a schedule, and a person decides whether the article is still right. None of the documentation we read describes checking the claims in the text against outside sources.
Why wrong articles matter more with AI assistants
When customers read a wrong article, one person is misled. When an AI assistant answers from it, the mistake can repeat in every conversation.
A widely reported example involved a chatbot rather than a help centre. In 2024 a tribunal in British Columbia, Moffatt v Air Canada (2024 BCCRT 149), found the airline responsible for wrong information its website chatbot gave a customer about bereavement fares, and awarded damages.
Law-firm summaries report that the tribunal rejected the argument that the chatbot was a separate entity, and held the airline responsible for information on its website.
We couldn't open the decision text itself, so we rely on those summaries.
The case concerned a chatbot, not a knowledge base, but the principle applies: what you publish is yours, however it is delivered.
Guru's own documentation reflects the same concern, by deprioritising overdue cards in AI answers.
What a claim-level checker adds
A workflow tool tells you an article is due for review. A claim-level checker tells you which sentence is wrong and why.
The statements in the text that an outside source can confirm or contradict.
Outside evidence, preferring primary sources such as official records, with domains you can list as trusted.
Supported, contradicted, outdated, misleading or unverifiable.
So the owner can fix the article in minutes instead of researching it.
Veritas excludes the publisher's own pages when it looks for evidence, so it can't use your pricing page to settle a plan limit in a help article. That is a deliberate limit, and it is why release-triggered reviews and review reminders still matter for claims about your own product.
Which articles to check first
You can't review everything at once. Start where a wrong answer costs most, and match the tool to the claim:
Statements about laws and standards are checkable against outside sources, and customers quote them in procurement reviews.
Browsers, operating systems and APIs change, and so does what they support.
An outside source can settle these.
These are your own facts. Tie review reminders to releases and price changes; a claim checker won't settle them.
Its answers are only as current as these articles.
See which facts on your site have gone out of date.
Request early accessWhat a finding looks like: an example
This is an invented example to show the format, not a real customer or product.
"Browser X supports feature Y."
Outdated. The browser vendor's current release notes say the feature was removed.
The vendor's release notes, with a link.
"Feature Y is no longer supported in Browser X. Use Browser Z instead."
An editor or support owner checks the evidence, applies the rewrite in one click or edits it, and the issue closes itself once the live sentence changes.
How Veritas checks help-centre articles
Veritas is an autonomous fact-maintenance agent. It runs in the cloud on a schedule and is built for publishers with large archives, which includes large help centres.
Through sitemaps, sitemap indexes and RSS or Atom feeds.
Statistics, dates, who holds a role, quotes, events, comparisons and superlatives. Opinion and prediction are skipped.
Outside the article, preferring primary sources, with domains you can list as trusted.
Supported, contradicted, outdated, misleading or unverifiable.
With evidence, citations and a suggested rewrite.
Fast-moving claims after 7 days, slowly decaying ones after 45, stable ones after 365.
For the same approach applied to any site, see our guide to outdated content detection.
What Veritas needs to see your help centre
Veritas finds pages through sitemaps and feeds and reads the HTML a page serves. That has a practical meaning for help centres.
Zendesk's documentation says a help centre gets an automatic XML sitemap, updated every two hours, unless it requires sign-in. Document360's says it generates a sitemap for public and mixed projects, and private projects have none. Other platforms differ, so check yours.
Pages that need JavaScript to show their text won't be read.
We don't describe support for pages behind a login, so internal wikis in Notion, Confluence or Guru fall outside what we claim.
Veritas excludes the publisher's own pages as evidence, so a claim only your own pages can settle isn't something it will confirm.
Check these points before you decide.
Questions to ask when you choose a tool
Whatever you shortlist, ask the vendor these:
Dates, owners and views are not accuracy.
Check public access, sitemaps and whether text needs JavaScript.
Ask for the evidence behind a flag.
Look for a queue with owners, not a report nobody opens.
A tool that never learns from your decisions creates noise.
A routine for keeping a help centre right
List the articles that mention the changed feature, plan or setting and send them for review the same week. This covers claims about your own product.
And route flagged claims to the owner of the article, not a general inbox.
Clear the queue of flagged claims, starting with the most-viewed articles and the ones about prices, limits and policies.
For structure, screenshots and wording, which a claim checker doesn't judge.
Mark false alarms "not an error" with a reason so the checker stops raising them.
Make sure the articles it draws on are the ones you keep current.
Honest limitsWhen Veritas isn't the right fit
Plan limits, prices, menus and policies need release-triggered reviews; Veritas doesn't use your own pages as evidence.
Use the verification features in your wiki tool.
Support analytics tools do that; Veritas checks existing articles.
Veritas reads text claims.
Veritas reads the HTML a page serves.
Read more about Veritas or other teams it is built for.
How Veritas works
- Finds your pagesReads your sitemaps, Google News sitemaps and RSS or Atom feeds on a schedule.
- Pulls out the claimsStatistics, dates, who holds a role, quotes, events. Opinion is skipped.
- Checks the evidenceSearches outside your site and prefers primary sources.
- Hands editors the fixConfident findings only, with citations and a one-click rewrite.
Frequently asked questions
How do I find outdated articles in my knowledge base?
Start with review dates and view counts to find old or quiet articles. Then check the claims an outside source can settle: third-party requirements, laws and standards, external numbers and dates. Veritas automates that for public help-centre pages and flags claims that are outdated, with evidence. For claims about your own product, tie reviews to releases.
What is the difference between a stale article and a wrong one?
A stale article is old by its metadata. A wrong article's content no longer matches reality. They overlap less than expected: an old article can be correct, and a recent one can contain a plan limit that just changed.
What counts as a claim in a help article?
Any checkable statement: a plan limit, a price, a date, a version requirement, a policy term or a named contact. Veritas pulls out statistics, dates, who holds a role, quotes, events, comparisons and superlatives, and checks them against evidence outside your site. Claims only your own pages can settle are better handled by release-triggered reviews.
How often should help centre articles be reviewed?
Tools set their own cycles; Guru, for example, recommends reviewing at least yearly. Articles about prices, limits and policies deserve review whenever the product changes, so trigger reviews on releases as well as on a calendar.
How is this different from a knowledge base content audit?
An audit usually reviews structure, links, views and ownership. Checking claims looks at what the text says and whether it is still true. The two work together: the audit decides what to review, and the claim check shows what is wrong inside it.
Can Veritas check a knowledge base behind a login?
We don't describe support for pages behind a login. Veritas finds pages through sitemaps and feeds and reads the HTML a page serves, so it is built for public help centres and documentation.
Do AI assistants make outdated articles riskier?
They can repeat a wrong answer in every conversation. Moffatt v Air Canada (2024) is a widely reported case where a company was held responsible for wrong information its website chatbot gave. That case involved a chatbot, not a knowledge base.
Does Veritas work with my help-desk platform?
Veritas finds pages through sitemaps and RSS or Atom feeds, not through platform integrations. If your help centre publishes a public sitemap, Veritas can read it.
Keep every page true
Veritas re-checks the facts across your published pages and hands editors a sourced fix when one goes stale.
Request early accessRelated guides
- An outdated content detector that checks the facts, not the datesMost tools sold as an outdated content detector flag pages by age or by falling traffic. Veritas reads each page, pulls out the checkable claims and tests them against current sources, so you see which sentence went stale and why.
- Outdated statistics checker for affiliate sitesAn outdated statistics checker for affiliate sites looks at the claims in your copy, not just your links. Veritas re-checks the statistics, dates, comparisons and superlatives on your review and buying-guide pages, and flags the ones that are no longer true.
9 sources, checked 31 October 2026
- Document360: Review reminders (updated Sep 2026)
- Document360: Sitemap
- Guru: Verification
- Guru Help: Verifying cards in Slack
- Notion: Page verification
- Zendesk: SEO for your help center (modified 2026-05-15)
- Enterpret: finding help center articles that are out of date or wrong (2026)
- Manatt: Air Canada chatbot ruling summary (28 Feb 2024)
- MLT Aikins: Moffatt v Air Canada summary (25 May 2024)