What we claim, and what we don’t.
Every figure on this site is either read from the engine that produced it or marked as a forecast. This page says which is which, and what is never claimed.
Two kinds of figure, never mixed.
A modelled figure is what the agents said. A measured one was read or counted.
A forecast
- What the 150 agents say: who would get in touch, where each stopped, why it left.
- Each agent’s reaction to each section of your page, and its notes.
- Which company each agent would contact when competitors are read beside you.
- The attention map on Three readers: where a first look is predicted to land.
Read or counted
- What is on your page: its sections, headings, links, images, forms and structured data.
- Your robots rules, and which crawlers they let in.
- Companies House filings and the facts read from them, each with its source.
- Each run’s own cost and time, stated when the run ends.
- How far answers move between two reads of the same page: printed under the results.
- Real respondents’ answers, when a study is run through Prolific.
Each agent is a large language model briefed as one of your potential buyers, built from published buyer research: a forecast, not a survey of real people. How it works →
What this site will not say.
The rules every line on this site is written against.
No customer counts, logos or testimonials
None appear on this site.
No lift a run did not produce
No percentage, uplift or conversion figure is printed unless a run on your page produced it.
No time a run did not take
No duration is promised. Where a time is quoted, it is one we measured, with when and how.
No claim about AI assistants
Nothing on this site says how buyers use AI assistants or what an assistant will recommend.
Agents are not people
The agents are called AI agents, and every result says what one is: a forecast, not a survey of real people.
No checkout that is not real
Checkout for paid plans is not open yet. Start free; paid plans follow.
Your page, from address to result.
Read from the engine’s code, not from a policy template.
- 01
Fetched once per run
Our server reads your public page once, and the 150 agents read that copy. robots.txt is read first and obeyed on every redirect; a robots refusal is final.
- 02
Public addresses only
An address that resolves to a private network is refused. If the page cannot be read directly for another reason, a page-fetching service (Firecrawl) is tried.
- 03
Zero-retention model calls
Every call to the language model is sent with zero data retention required and data collection refused. A provider that would keep or train on it is not used.
- 04
A run is kept a day
A run’s results are held for one day so you can reload them. After that the run has expired and must be read again.
- 05
The token stays in memory
A run is opened by a 192-bit random token. This site keeps it in the page’s memory and never puts it in a link, so a shared address cannot open your run.
- 06
Competitors are read, never contacted
A competitor’s public page is read the same way as yours. Nobody emails, messages or calls them.
What we hold, and how you pay.
The privacy page lists what is held about you. The terms say what you agree to.
Privacy
The public sources a report is built from, what we hold about you, the cookies, Stripe for payment and Resend for email.
Read the privacy pagePayment
Payment is taken on Stripe’s own hosted checkout. Card details are entered there and never reach our servers.
Checkout for paid plans is not open yet. Start free; paid plans follow.
We hold no security certification today. To see or delete what we hold about you, use the contact form and choose “Your data”.
Where we were wrong, in public.
The research library keeps a section of sources we have retired, and a sentence we published and then retracted, with why.
Send 150 agents to your page.
Type your web address. Free account, no card.