# Improve your score (/index-ranking/improve-your-score)

# Check and improve a site's score [#check-and-improve-a-sites-score]

A score tells you where you stand. The sessions behind it tell you why. This
guide goes from a report to a ranked list of fixes, then re-scores the site
once they ship.

You can do all of it at [index.stuntdouble.io](https://index.stuntdouble.io),
or ask your AI assistant to do it through the [MCP server](/mcp), which ships
the `check-agent-readiness` [skill](/mcp/skills).

<Steps>
<Step>

### Read the report [#read-the-report]

Open the site's report, or call `get_index_report` with its domain. If your
Stunt Double project is linked to an Index domain, pass its `project_id`
instead. Start with the lowest categories and providers.

</Step>
<Step>

### Read the sessions behind the weak spots [#read-the-sessions-behind-the-weak-spots]

Filter the sessions by category or provider (`list_index_sessions` with
`category` or `provider`) and read what the agents actually ran into: a cookie
wall before the cart, help content that only renders with JavaScript, a
product page with no structured data. Fix what the sessions show, not what you
guess.

</Step>
<Step>

### Compare with your sector [#compare-with-your-sector]

Browse the leaderboard for your sector (`search_index_domains` with `sector`)
or look up named competitors, and read the reports worth contrasting.

</Step>
<Step>

### Rank the fixes [#rank-the-fixes]

1. Failing probe checks with a high weight.
2. Frictions that recur across providers.
3. Issues that only one provider hits.

Tie every fix to a failing check or a session friction. For legacy pages that
can't be rebuilt soon, [WAO](/stuntkit/wao) repairs the accessibility tree
agents read without changing the page.

</Step>
<Step>

### Re-score once the fixes ship [#re-score-once-the-fixes-ship]

Re-run the domain (`request_index_rerun`) for fresh probes and about 24 agent
sessions. Only the domain's owner can re-run it. Poll
`list_index_sessions` with the returned `run_id` until no session is running,
then compare the new report with the old one, category by category.

</Step>
</Steps>

<Callout variant="tip" title="Keep the fix from regressing">
  Before the re-run, confirm the fix on a preview deployment with a
  [checklist](/concepts#checklists), then run that checklist on every deploy
  with an [automation](/concepts#workflows).
</Callout>

## Next steps [#next-steps]

- **[Connect the MCP server](/mcp/connect)**
- **[Make any site readable by AI agents](/stuntkit/wao)**