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

# Prompt suggestions

> Review the prompt changes Axon proposes from your real mail, edit them, test them, and apply the ones you want.

## Overview

Axon analyses how your requests have been classified and looks for patterns where an intent prompt could be clearer. When it finds one, it proposes a specific change to that prompt — along with the mails it learned from and the accuracy improvement it expects.

Nothing changes on its own. Suggestions are proposals you review, edit, and save yourself.

> [!NOTE]
>
> Prompt suggestions are part of **Axon Premium** and arrive automatically — there's nothing to switch on. See [Auto-tuned intent descriptions](/knowledge-base/axon/auto-tuned-intent-descriptions) for how suggestions build up.

## Suggestion sets

Related prompt changes are grouped into a **suggestion set**. A set can cover one intent or several — when Axon finds that requests are being classified as *Order* that should be *Quote*, fixing that usually means adjusting both prompts, so they travel together.

Each set carries:

| | What it tells you |
| :- | :- |
| **Expected accuracy** | How much classification accuracy should improve if you apply the set, for example **+10.8% expected accuracy**. |
| **Prompt changes** | How many intent prompts the set would change. |
| **Mails per month affected** | How many requests a month this change would apply to. |
| **Mails analysed** | How many requests Axon looked at to reach the suggestion. |

## Find your suggestions

When Axon has suggestions for you, a badge appears on the **Active intent tree** showing how many sets are waiting. Individual intents with a pending change are marked in the tree too, so you can see which prompts are affected before opening anything.

Hover the badge to see the sets, each listed with its prompt count and how many mails were analysed. Select one to open it.

## Review a set

The review dialog shows, for each intent in the set:

* **A summary of the change** — what Axon is proposing, in plain language.
* **The current prompt** — the live prompt as it stands today.
* **The proposed prompt** — Axon's rewrite, which you can edit before accepting.
* **The mails it learned from** — real examples, each showing the subject, sender, date, the intent it was classified as, the intent it should have been, and why.

The mail examples are the evidence behind the suggestion. If the proposal doesn't look right, they usually explain what Axon saw.

> [!TIP]
>
> The proposed prompt is a starting point, not a fixed text. Edit it in the dialog before you accept — your version is what gets loaded, not Axon's.

### When a suggestion no longer fits

Suggestions are generated against your prompts as they stood at the time. Two things can happen in between:

* **The prompt has moved on.** If someone edited that intent's prompt after the suggestion was generated, the proposal is marked as out of date. Axon was rewriting an older version, so read it against your current wording before accepting.
* **The intent is gone.** If the intent has since been deleted or deactivated, the suggestion can't be applied at all, and accepting the set will fail.

In both cases the evidence is still worth reading — the mail examples tell you what Axon spotted, even if the proposed wording needs rethinking.

### Test before you accept

Axon has already tested the set against your recent, verified mail — that's why it was shown to you at all.

If you've edited a proposed prompt, or you want to see the behaviour for yourself, click **Test proposed prompts**. This re-classifies real requests with the proposed wording and shows you exactly which classifications would move. See [Testing prompt changes](/knowledge-base/axon/testing-prompt-changes.md).

## Accept a set

Click **Accept {n} prompt changes & open draft**. This loads the proposed prompts into your intent tree as an unsaved draft.

> [!WARNING]
>
> Accepting does not save anything. The suggestion stays open until you **save the intent tree**, which is what applies the prompts. If you discard the draft, or the save fails, the set is still waiting for you.

That means you can accept a set, keep editing the prompts alongside your other changes, and save everything together. If you accept a set and change your mind, discarding the draft leaves things exactly as they were.

You can't accept a set while another save is still running, and you can't accept one with an empty prompt.

Once saved, the change is recorded in the intent's prompt history with the source **From learnings**, so you can always see which prompts came from a suggestion — and revert if you need to. See [Prompt history](/knowledge-base/axon/auto-tuned-intent-descriptions#prompt-history).

## Dismiss a set

If a suggestion isn't right for your business, click **Dismiss set** and confirm. The set is closed and won't come back.

Dismissing is the right call when Axon has inferred a rule that doesn't match how you actually work. The classification it flagged may be deliberate on your side.

## Who can use it

Reviewing, accepting, and dismissing suggestions requires **Editor** access to Axon. Users with **Viewer** access won't see the accept or dismiss actions.

## What's next

* [Testing prompt changes](/knowledge-base/axon/testing-prompt-changes.md) — check the impact before you apply anything.
* [Navigate and organize the intent tree](/knowledge-base/axon/intent-tree.md)