Article analytics
Views, helpful and unhelpful votes, search queries that surfaced articles. How to read the data and what to do with it.
Article analytics
Every article in Ochre tracks views, votes, and search queries. The data is collected automatically from the moment you publish. The point of looking at it is to find articles that are not pulling their weight, then rewrite them.
Where to find it
Open any article and click the Analytics tab in the sidebar. The full index across all articles lives at Knowledge → Analytics.
The default view is the last 30 days. You can switch to 7, 90, or custom.
Views
A view is one rendered article page in a browser session. Ochre dedupes within a 30-minute window per visitor, so refreshing does not inflate.
Views are interesting in aggregate, less so for individual articles. Use them to:
- Spot the long tail. Articles with zero views in 90 days probably are not doing anything.
- Find traffic spikes. A 10x spike usually correlates with a product change, a new release, or an outage.
- Compare related articles. If one billing article gets 10x the traffic of another, the structure tells you something.
Do not optimize for views alone. A heavily viewed unhelpful article is worse than a sparse helpful one.
Helpful and unhelpful votes
Visitors can mark each article helpful or unhelpful, anonymously, once per article. See Article voting for the mechanics.
Two metrics matter:
- Helpful rate. Helpful votes divided by total votes. Above 80% is healthy. Below 60% means the article is missing something or saying it wrong.
- Vote volume. Total votes regardless of polarity. Articles with high views and zero votes might be answering the question implicitly without prompting a reaction. Or your vote widget might be off-screen.
Sort by helpful rate, ascending. The bottom of that list is your rewrite queue.
Do not overreact to single votes. One unhelpful out of 50 helpful is signal noise. Five unhelpful out of seven votes is a problem.
Search queries
This is the most underused panel. For each article, Ochre shows the queries that surfaced it in help center search.
You will see two patterns:
Queries that match the article
"How to reset password" surfaces the password reset article. Working as intended.
Queries that do not match
"Why is billing wrong" surfaces the password reset article. That is a structural problem. Either the article is mistitled, or the search is matching weakly, or you are missing a billing article.
The fix is usually:
- Rename or rewrite the title to match what people search.
- Add the searched keywords to the article body naturally.
- Write the missing article. Customers are telling you what they want.
Org-wide search queries
The Knowledge → Analytics view also shows the top searches across the whole help center, including searches that returned no results. Treat the no-results list as a TODO list. Each query is an article you should write.
This is the single highest-leverage feedback loop in the help center. Pay attention to it.
AI-driven views
When the AI agent cites an article in a reply, that counts as an "AI surface" rather than a human view. The analytics view splits these out so you can see human reads vs AI uses.
If an article is AI-surfaced often but human-read rarely, it is an internal-grade article (correct, but not searchable by humans). Consider rewriting the title and adding it to a category customers actually browse.
Time-on-page and scroll depth
Ochre does not track these. Most teams do not act on them, and we would rather not load extra trackers on customer browsers. If you want them, you can add Plausible, Fathom, or PostHog through a snippet in the help center theme.
Exporting
Click Export on the Knowledge → Analytics view to download a CSV of all articles with views, votes, and helpful rate over the selected time range. Useful for a monthly review meeting.
What to do every month
A small ritual that pays off:
- Sort articles by helpful rate ascending. Pick the bottom three. Rewrite them.
- Look at no-result searches. Pick the top three. Write articles for them.
- Look at the most-viewed articles. Make sure they are current. Stale top-of-funnel articles damage trust the most.
That is it. Twenty minutes a month, repeated, beats a quarterly content sprint.
Privacy
Analytics are aggregate. Ochre does not store visitor IPs longer than session deduplication needs (a few minutes). Voter identity is never stored. Article view logs are not tied to specific customer accounts even when the help center is gated.
Related
- Article voting for the voting widget.
- Article SEO for SEO knobs that affect indexing.
- The brain: KB graph for how article quality affects AI retrieval.
Was this article helpful?