Every time Helpie’s assistant can’t answer a question, two things happen. The visitor gets routed to a ticket — that’s the part they see. And the question itself gets remembered — that’s the part you see, here.
The Content Prospector is a running report of the questions your visitors asked that your site couldn’t answer, grouped into themes. It’s your content to-do list, written by your actual customers. Find it at Helpie > Knowledge Base > Content Prospector.
—
What You’re Looking At
The report shows unanswered questions from the last 30 days, clustered into themes. Each cluster shows:
- A short label — the theme, like “International shipping” or “Bulk pricing”
- A count — how many times questions in this theme came up
- Sample questions — up to three verbatim examples, in your visitors’ own words
- First and last seen — when the theme started appearing and how recently
“Couldn’t answer” means exactly what the assistant means by it: the question came in, your indexed content didn’t ground an answer, and the visitor got a clarifying question or a ticket offer instead. The detection uses the same signal as the chat itself, so this panel and the assistant can never disagree about what counts as a miss.
Themes that only came up once are kept but hidden — a single odd question is noise; the same question twice is a pattern.
The sample questions are the gold here. Visitors phrase things in ways you wouldn’t — and when you write the missing content, using their words in your headings means the next visitor asking the same thing gets matched straight to it.
—
Refreshing
The report refreshes itself in two ways:
- Automatically after the nightly indexing run — so it’s never more than a day stale, and it always reflects what the night’s pass indexed
- On demand — the Refresh now button, for when you’ve just made changes and want to see where things stand
The timestamp above the table shows when it last refreshed.
One honest note on cost: grouping the questions into themes uses a single small AI call on your own Anthropic key per refresh. It’s one call for the whole report, not one per question — a tiny line item, but we’d rather you know about every place your key gets used. (The collection of unanswered questions itself is free; only the grouping asks Claude for help.)
—
Acting On It: Fix the Cause, Not the List
Here’s the design decision we’re fond of: there’s no “mark as done” button, and you’ll never need one.
Each cluster stays on the list while the gap exists. Fix the underlying cause, and the cluster clears itself — new questions on that theme stop missing, the old ones age out of the 30-day window, and the cluster shrinks below the surface threshold and disappears. The list heals the same way the gap does: for real.
What “fixing the cause” usually looks like, in rough order of likelihood:
- Write the missing content. The most common case — visitors are asking about something your site simply doesn’t cover. Publish the page or FAQ, reindex, done.
- Make existing content findable. Sometimes the answer exists but hides under vocabulary visitors don’t use. Add their phrasing to the page, or add a route rule so questions on that theme always surface the right page.
- Adjust the persona. If a cluster is really “things I don’t want the assistant discussing at all”, a line in Behaviour & Boundaries handles it more gracefully than a refusal.
Then let the next refresh do the bookkeeping.
—
Common Questions
The panel says “Nothing here yet” — is something broken?
No — it just hasn’t had its first refresh, or the chat hasn’t had any unanswered questions to report. Click Refresh now, or wait for the nightly indexing run to refresh it automatically.
It says my knowledge base is keeping up. Can I trust that?
Yes — it means no theme of unanswered questions crossed the surface threshold in the last 30 days. Enjoy it. (If a refresh ever fails partway — say, an API hiccup during grouping — the panel shows a warning rather than a falsely rosy empty state.)
Do resolved clusters leave any history?
The panel shows the current 30-day picture only — it’s a to-do list, not an archive. The underlying conversations remain in your chat logs for as long as your retention settings keep them.
Will one confused visitor spam the list?
Not meaningfully. Single-question themes stay hidden, and the clustering groups by meaning rather than exact wording — it takes a repeated, genuine gap to earn a row.