# Track what changed after you improved a topic

Source: https://frosvo.com/docs/track-support-improvements
Updated: 2026-11-06

Record a change you made to cut contacts about a topic, such as a new help article, and see how much of your conversations that topic was before and after, with warnings when there is too little data.

You wrote a guide, fixed a confusing step or changed a message, and you want to know if customers stopped asking. This page is for owners and managers who read the reports. It keeps the change, the topic and the numbers together in one place.

## Before you start

- You need reports, and conversations tagged by topic. Turn on [auto-tagging](/docs/auto-tagging-and-sentiment) for a website so conversations get topic tags such as billing or shipping, or tag them by hand.
- Recording or removing an improvement needs the reports permission. Everyone who can open Reports can read the list.

## How do I record an improvement?

1. Open **Reports**. Under the topics card, find **Improvements and what changed after**.
2. In **Topic**, type the tag the change is about. The topics already in your report are suggested.
3. In **What you changed**, say it in a few words, for example "Wrote the domain transfer guide".
4. Set **Went live on** to the day customers could first see the change.
5. Choose **Add**.

The date cannot be in the future, and up to 50 improvements are kept. Remove old ones with **Remove**.

## How do I read the result?

- For the first seven days it says **Too soon to compare**.
- After that it shows, for example, "billing was 50% of conversations before and 15% after (−35 points)". Below it are the counts it came from, so you can check them against your reports.
- The period before is the 28 days ending the day before the change. The period after starts the day after the change and runs up to 28 days. The day of the change is in neither.
- Spam is not counted.

## When should I be careful?

- **Few conversations.** If either period has fewer than 30 conversations, or the topic has fewer than five in both, the card says to read it as a hint, not a result.
- **Other changes.** If you also ran a campaign, changed pricing or had an outage, the difference may not come from your change. The card always says it does not prove cause.
- **Moves under two points** are small enough that you should treat them as no change.

## Common problems

- **"Choose a topic: a single word…"** Topics are tags: lowercase letters, numbers and dashes, such as domain-transfer.
- **The topic shows 0% before and after.** No conversations carry that tag. Check the spelling against your report's topics.
- **I cannot add one.** You need the reports permission. Ask an owner. See [roles and permissions](/docs/roles-and-permissions).

## Related

See [reports](/docs/reports), [auto-tagging and sentiment](/docs/auto-tagging-and-sentiment), [drafting help articles from chats](/docs/draft-help-articles-from-chats), and [re-testing the assistant after you change your knowledge](/docs/retest-after-knowledge-changes).

## Common questions

### Does this prove my change worked?

No. It compares the share of conversations on the topic in the weeks before and after. Seasons, campaigns and outages also move the numbers, so use it as evidence to look at, not as proof.

### Why does it compare shares and not counts?

A busy month raises every topic's count. Looking at the share of all conversations keeps a busier month from looking like a worse topic.

### When will I see a result?

After seven days from the day the change went live. It compares up to 28 days before with up to 28 days after.
