Knowledge bases do not fail loudly. No page throws an error because it was last updated nineteen months ago, or because it is ninety words that promise more than they deliver, or because it never got a description and so shows up in search results as a blank.
They just quietly stop being useful, and the first sign is usually a support ticket about a feature you documented.
What the scan looks for
FinalDoc’s content health scan reads every published article and flags four specific, checkable problems:
- Thin — under 80 words. Usually a page someone started and never finished.
- No description — nothing to show in search results, previews or link cards.
- Uncategorised — the article exists but sits outside your navigation, so readers reach it only by search or luck.
- Stale — untouched for over 180 days. Not automatically wrong, but worth a look.
These are deliberately concrete. Each flag is something you can check yourself in a few seconds and either fix or dismiss — no judgement about whether your writing is any good, because that is not a thing software should be asserting.
It tells you what changed, not what it already told you
The scan compares each run against the previous one. If the same twelve articles were flagged last week and you have not acted on them yet, it does not stack another identical card on your feed. You hear about what is new, with a count of what you have resolved since last time.
And when nothing is wrong, it says so — content health looks good, nothing flagged — instead of manufacturing findings to look busy. A maintenance tool that always finds something is a tool you stop reading.
Run it on a schedule
Content health is one of the recurring tasks you can give an AI Writer. Hire one, add Run content health scans, and choose weekly or monthly. It runs on its own and posts findings to the writer’s activity feed, where you can jump straight into the first flagged article — or have the writer draft a repair for review.
Weekly suits a knowledge base under active development. Monthly is plenty for a mature one — the flags are about slow decay, and checking for slow decay every day is just noise.
If you would rather have a score
The scan deals in flags. If you want a single number to watch over time, Analytics → Content Health is the other view: a 0–100 score built from five dimensions — freshness, completeness, engagement, SEO and category coverage — with drill-down lists behind each one.
The two are separate tools with separate thresholds, so their counts will not line up. The score is for tracking a trend; the scan is for getting a specific list of things to fix this week.
Freshness is the dimension worth understanding before you read the number. It counts articles you have marked as reviewed, using the Mark Reviewed button on each stale row — a deliberate "I have read this and it is still correct" rather than a side effect of touching the file. That makes it a real record of when a human last checked each page, and working through the list is what moves the score.
What to do with the findings
Not everything flagged needs fixing. A glossary entry is legitimately thin. A policy page that has not changed in a year may be perfectly accurate.
The value is in the ones that surprise you: the setup guide nobody has revisited since the product changed, the article that fell out of its category during a reorganisation, the page with no description that has been quietly under-performing in search for months. Those are the ones costing you tickets.