# Self-improvement

Darwin tests new openings on every page, keeps whichever brings more buyers to booking, and tests again from the winner.

This is the Markdown version of https://www.meetdarwin.ai/features/self-improvement, published by Darwin and generated on 23 October 2026.

## A website chat that gets better on its own.

The first message decides whether a visitor talks at all. Darwin keeps testing it on every page with traffic, and holds back any opening that looks likely to book fewer meetings than the one it replaces.

## Natural selection, on autopilot. Darwin runs the tests. You decide how much it does on its own.

1. **Darwin reads the page.** It drafts a new opening shaped by what a visitor on that page is most likely trying to work out.
2. **It tests the new opening.** Visitors see one opening or the other. Darwin counts who reaches the booking step, and who books.
3. **The winner stays.** When the new opening wins, Darwin applies it or waits for your OK. The next test starts from there.

## Tests you can trust. Run them your way, on every page, without a step backwards.

- **Run it your way.** Let Darwin run tests and apply winners for you, or approve each one yourself.
- **Never a step backwards.** Darwin holds back any new opening that looks likely to book fewer meetings than the one it replaces.
- **A test for every page.** Busy pages get their own test. Quieter pages share one by type, so every page keeps improving.

## Your openings, your way. Add your own, learn from others, send results on.

- **Add your own opening.** Write an opening for any test and Darwin tests it alongside its own.
- **Learn from other Darwin sites.** New tests start from what has worked elsewhere. Only anonymous counts are shared.
- **Send results on.** Webhooks fire when a test starts and when a winner is found.

## Questions

**What exactly improves?**

The opening message that invites a visitor into chat. Busy pages get their own test and quieter pages share one with similar pages; the winner becomes the new starting point. These experiments don't rewrite your website, change your prices or retrain the AI model.

**Does my team have to run the experiments?**

Not if you don't want to. With Automatic testing on, Darwin starts tests, replaces openings that fall behind and applies winners on its own. Want to approve winners first? Turn off Apply winners for me in Experiments settings, and each winner waits for you to apply it.

**What happens when a new opening doesn’t win?**

A challenger doesn’t replace the current opening just because it is new. The current version stays in place unless the experiment meets the criteria for a change. Results and decisions remain available in the experiment history.

**How does Darwin decide a new opening has won?**

By the share of visitors who saw it and reached the booking step: they saw times, asked for a meeting, or booked. Each opening needs at least 200 views, the new one needs at least 10 visitors who reached booking, and Darwin has to be 95% sure it beats your current opening. It never applies an opening that books fewer meetings than the one it would replace.

**Is there a limit on how many pages Darwin can test?**

No plan caps it. Busy pages get a test of their own, quieter pages share one with similar pages, such as your other pricing or comparison pages, and the rest of the site shares a site-wide test. You can also give an important page its own test, whatever its traffic.

**Does Darwin share our test results with other companies?**

Only as counts. With Learn from other Darwin sites on, your new tests start from the kinds of openings that did best on similar pages across Darwin sites, and your results join that pool as counts by kind of page and kind of opening. Your copy, pages and visitors never leave your workspace, and switching the setting off removes your counts.

[Pricing](https://www.meetdarwin.ai/pricing) · [How Darwin works](https://www.meetdarwin.ai/how-it-works)