> ## Documentation Index
> Fetch the complete documentation index at: https://docs.helpalive.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Topics inbox: triage Agent coverage gaps

> The Topics inbox is your queue of questions the Agent couldn't answer well — grouped, ranked, and ready to act on as actionable coverage gaps.

The **Studio → Topics** inbox is one of the most useful surfaces in HelpAlive. Every question the Agent couldn't answer well is captured, grouped into topics, and ranked — so your team always knows what to fix next.

## How questions get flagged

Whenever the Agent answers, HelpAlive watches for three signals:

| Signal               | Meaning                                                                                                           |
| -------------------- | ----------------------------------------------------------------------------------------------------------------- |
| **Nothing matched**  | We searched your knowledge base and didn't find anything relevant. Usually means your docs are missing something. |
| **Weak match**       | We found something, but it wasn't a strong fit for the question.                                                  |
| **User thumbs-down** | The user rated the answer poorly, even when we were confident.                                                    |

A question can carry more than one signal — for example, a thumbs-down on a weak match counts twice toward the topic's score.

## Topics, not individual questions

Individual questions are noisy ("how do I cancel?", "stop my subscription", "delete account please"). HelpAlive groups similar questions together into **topics** so you see patterns instead of single questions:

* **Cancellation flow** — 47 questions in the last 30 days
* **Pricing for additional seats** — 22 questions
* **How to invite a teammate** — 18 questions

Topics stay stable as new questions come in — adding a few new ones doesn't reshuffle your inbox.

## Ranking

Topics are ranked by **volume × gap score** — how often a topic comes up, weighted by how badly the Agent is failing on it. A topic asked 200 times with mostly confident answers ranks below a topic asked 50 times the Agent can't handle.

## Working the inbox

Open **Studio → Topics** in your dashboard. Each topic shows:

* **Question count** in the selected window.
* **Severity** — `high`, `medium`, or `low`.
* **Sample questions** from the cluster.
* **Sample answers** the Agent gave (so you can spot weak ones).
* **Status** — `identified`, `addressed`, or `ignored`.

Click a topic to see all its questions, all its answers, and the rating distribution.

## Resolving a topic

Three options:

1. **Add or update content** in **Agent → Train**. Re-index, then publish.
2. **Add a custom answer** for things that aren't doc-shaped — internal policies, plan-specific responses, time-sensitive notices. Configure in **Agent → Personalization**.
3. **Ignore it** for off-topic questions ("what's the weather?") or things genuinely out of scope.

After publishing a fix, watch the next 24–48 hours of traffic. The topic's score should drop sharply if the new content is the right fit. If not, the topic stays at a lower rank.

## Filters

| Filter          | Use case                                                                        |
| --------------- | ------------------------------------------------------------------------------- |
| **Time range**  | Compare the last 7 days to the last 30 to see what's getting worse or better.   |
| **Page**        | Narrow to questions from a specific page.                                       |
| **Signal**      | Show only "nothing matched" (clear gaps) or only thumbs-downs (quality issues). |
| **Status**      | Show identified, addressed, or ignored topics.                                  |
| **Plan / role** | Find out if pro users hit different gaps than free users.                       |

## What it isn't

* **Not a CSAT survey.** The thumbs-up / down feeds in but isn't the whole picture.
* **Not Agent-tuning telemetry.** Quality issues surface here, but the goal is product action — write docs, fix flows — not LLM tuning.
* **Not a feature-request log.** Some topics will be feature requests; treat those as input to product, not as docs work.

## Privacy

The inbox shows **the actual text** of user questions because reading them is the whole point. To keep this safe:

* Question text lives in a separate, secure store from your behavioral analytics.
* Personal information (emails, phone numbers, IDs) is automatically redacted before storage.
* You only see questions from your own users — never from other HelpAlive customers.

If a user requests deletion, their question history is removed from the inbox. See [Data deletion](/privacy/dsar).

## Next steps

<CardGroup cols={2}>
  <Card title="Train" icon="book-open" href="/agent/train">
    Add or update content after resolving a topic.
  </Card>

  <Card title="Publishing" icon="git-branch" href="/agent/publishing">
    Stage and verify changes before they reach users.
  </Card>
</CardGroup>
