---
title: "AI page variants for trial signup"
url: "https://trialvalid.com/product/ai-optimization.md"
canonical_url: "https://trialvalid.com/product/ai-optimization"
last_updated: "2026-10-21"
type: "guide"
summary: "A model writes new versions of your trial page, traffic splits evenly between the ones you switch on, and you promote the winner."
related: ["https://trialvalid.com/product.md", "https://trialvalid.com/product/trial-holds.md", "https://trialvalid.com/product/trial-pages.md", "https://trialvalid.com/product/attribution.md", "https://trialvalid.com/product/alerts.md", "https://trialvalid.com/product/trial-reminders.md", "https://trialvalid.com/docs/trial-lifecycle.md", "https://trialvalid.com/docs/webhooks.md", "https://trialvalid.com/product/high-value-alerts.md", "https://trialvalid.com/product/custom-statement-descriptors.md", "https://trialvalid.com/product/dispute-readiness.md"]
---

# AI page variants for trial signup

> **For AI agents:** the complete index of TrialValid pages is [llms.txt](https://trialvalid.com/llms.txt). Every page is also served as markdown at its URL plus .md.

A model writes new versions of your trial page, traffic splits evenly between the ones you switch on, and you promote the winner.

## How it works

1. You describe your product in a few sentences, with an optional angle such as "speed" or "price".
2. A model writes a new variant for the trial page: headline, sub-headline, up to 6 bullet points, a button label and an accent color. It is also shown your current page, so the new version reads as an alternative.
3. The answer is checked against the page schema and its length limits. An answer that does not fit is refused, not repaired, and nothing is saved.
4. The variant is saved inactive. You read it, turn it on or edit it, and from then on traffic splits evenly between your active variants. When one is ahead on your own numbers, you promote it.

## Why it works

Writing a second and third version of a trial page is slow, so most pages never get tested. A model turns that into a minute of reading. You stay the editor: nothing the model writes reaches a visitor until you switch it on.

The split is fair. A visitor is assigned to a variant by a hash of their visitor id, so they see the same version on every visit and the active variants get an even share. Results per variant (visits, signups, conversions, revenue) come from [attribution](https://trialvalid.com/product/attribution.md).

## Traditional approach versus this

|  | Writing and testing by hand | With AI page variants |
| --- | --- | --- |
| Writing a new version | You write it, or brief a copywriter | The model writes it from your product description |
| Going live | You publish it yourself | You switch it on after reading it |
| Splitting traffic | Set up in a separate testing tool | Even split between active variants, built in |
| Picking the winner | Read results in another tool | Promote a variant from your own page results |
| What the copy may claim | Whatever the writer chooses | The model is told not to invent statistics, testimonials or ratings |

## Controls

- Nothing is published automatically. A new variant starts inactive.
- You choose which variants are active, and a retired variant can be switched back on.
- Promoting a variant sends all traffic to it and retires the others.
- Your product description is limited to 10 to 2,000 characters and the optional angle to 300.
- A page with an archived status cannot get new variants.

### Does AI publish changes to my trial page?

No. Every variant the model writes is saved inactive. You read it, edit it if you want, and turn it on yourself.

### How is traffic split?

Evenly between the active variants. A visitor keeps the same variant on every visit, because the assignment comes from a hash of their visitor id.

### What happens when I promote a winner?

All traffic goes to the promoted variant and the other variants are retired. You can switch a retired variant back on later.

### Do you promise more signups?

No. TrialValid makes no promise about how much a variant will change your signups or conversions. You judge each variant on your own visits, signups and revenue.

### What does the model write?

A headline, a sub-headline, up to 6 bullet points, a button label and an accent color, within fixed length limits. It is told to state your trial length and price exactly and to leave out invented statistics, customer counts and ratings.

### Can I still write variants myself?

Yes. Variants you write by hand sit beside the model-written ones and are split the same way.

## Related

- [How card-backed free trials work](https://trialvalid.com/product.md)
- [Card holds for free trials](https://trialvalid.com/product/trial-holds.md)
- [Trial pages and variants](https://trialvalid.com/product/trial-pages.md)
- [Attribution and analytics](https://trialvalid.com/product/attribution.md)
- [Alerts and statement descriptors](https://trialvalid.com/product/alerts.md)
- [Trial reminders and records](https://trialvalid.com/product/trial-reminders.md)
- [Trial lifecycle](https://trialvalid.com/docs/trial-lifecycle.md)
- [Webhooks](https://trialvalid.com/docs/webhooks.md)
- [High-value trial alerts in Slack](https://trialvalid.com/product/high-value-alerts.md)
- [Custom card statement descriptors](https://trialvalid.com/product/custom-statement-descriptors.md)
- [Dispute readiness for free trials](https://trialvalid.com/product/dispute-readiness.md)