The Stale Answer Audit
Check whether your support bot's answers still match your sources.
Your help centre changes. Prices move, plans are renamed, pages are retired. An answer that was right last quarter can be wrong today, and the bot keeps giving it. We check a sample of your bot's real answers against the live pages they rely on, and show you the evidence for each finding.
Free, from your public help centre only. A person checks every finding before you see it.
- Customer asked
- How much is the Pro plan?
- Bot answered
- Pro is $49 per user per month.
- Live source says
- “Pro — $59 per seat / month, billed annually”
- Source
- help.example.com/pricing
- Read at
- 2026-09-28 09:14 UTC
- Checked by
- A person, against the page
Why answers go stale
The bot didn't change. The pages under it did.
Support bots answer from help-centre articles, docs and policy pages. Nobody tells the bot when those pages change, and nobody tells you which answers the change broke.
Prices and limits move
A plan goes from $49 to $59, a free tier drops from 5 seats to 3. The old number lives on in answers.
Plans and versions are renamed
“Pro” becomes “Growth”, v2 is retired. The bot still recommends what no longer exists.
Pages retire and links die
Answers point customers at pages that now 404, or quietly redirect to the home page.
Three steps, each one optional
Start free. Go deeper only if it's worth it.
The Snapshot costs you nothing and shows you what kind of problems your pages have. The audit checks what your bot actually said.
-
Step 1 · Free
Help-centre Snapshot
We read your public help centre and send up to 10 issues a person has checked: broken links, links that now land on the home page, retired pages still linked, and plain contradictions between pages.
Request a Snapshot -
Step 2 · From US$4,500 (US$2,500 founding)
Stale Answer Audit
Up to 200 (or 500) of your bot's real answers, each checked against the live sources it relies on. You get your stale answer rate, the evidence for every finding, and a source register with re-check dates.
See scope and prices -
Step 3 · After an audit
Monitoring
Each month we re-check the same sources and send a short report of what changed, so you fix answers before customers find them.
How monitoring works
Rules we don't bend
An answer we can't settle is never called fine.
An audit is only useful if you can trust its “all clear”. So the method is built to say “we couldn't tell” rather than guess.
Unclear stays unclear
If the source can't settle a claim, the report says so. It is never counted as current.
A person checks every finding
Nothing reaches you that we haven't opened and checked against the page ourselves.
Every read is logged
Each finding carries the page, the passage and the time we read it, so you can check it yourself.
No AI judge unless you allow it
Checks are rule-based. An optional Anthropic-only AI step for unclear answers stays off unless your signed Order allows it.
Public pages, your permission
The Snapshot reads only public pages and respects robots.txt. We never query your bot without permission.
Independent of your vendor
We don't sell the bot, so we have no reason to grade it kindly.
Why not your help desk's own tool?
They check the articles. We check the answers.
Intercom's Fin and Zendesk's knowledge tools now flag outdated or conflicting articles inside their own knowledge base. That is useful, and we recommend it.
We check something different: what your bot actually told customers, against the live sources it relied on, including pricing pages, docs and policies outside the help centre. And we are independent of the vendor whose bot is being checked.
See what your help centre is hiding.
Send us one public URL. If we find issues worth your time, you get up to 10, each checked by a person, with the link to see it yourself.