150 agents a run · free account · no card

150 AI agents read your page the way your buyers would.

Type your web address. Each agent is briefed as one of your potential buyers from published research, reads section by section and stops where it would. Watch them read it. See how many would get in touch.

Six products ↓

Free account: your email is the whole form. Then 150 AI agents read your page.

Sample run · an illustration, not your page

150 seats. This is what a run looks like.

Reading the sample page (now)
Building buyer personas for its market
150 agents reading
Writing up what they found

liked it, strongly
liked it
mixed
put off
put off, strongly
reading
would get in touch

Would get in touch18of 150 agents150 of 150 agents so far

What they’re saying (sample)

Marketing manager· would get in touchThe headline is clever but says nothing about the service.
Practice manager· would get in touchThis is exactly the problem we have. I’d book a call.
Owner-manager· would get in touchStrong headline, but I still can’t tell what it costs.
Operations director· would get in touchBy the second section I know what they do.

Sample run · an illustration, not your page

People who’d get in touch16of 150The agents said 18

Agreed with their agent131of 15019 reactions changed

Calibrating the buyer agents with the human answers
Reading the page again with the sharper agents
Found 21 improvements · estimated +4.6% conversion on this sample page

Illustrative figures for a made-up page. Nothing was measured. Your own run reads your page.

Would get in touch0of 150 agents0 of 150 agents so far

What they’re saying (sample)

Each agent’s note appears here the moment it finishes.

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 →

Every time you run a test

The same agents. The same seats.

The same 150 agents, the same seats. Change one section. A difference is real when it is more than twice its standard error.

AYour page as it is

BOne section changed

B − ASeat by seat

read B better
read B worse
no change

Inside the band: run-to-run noise. Keep testing.

Outside it: a real difference. Keep the change.

Illustration: a fixed pattern, not a run. Your own run shows your own tiles.

Fixed seatsSeat 37 in run A is seat 37 in run B: the same persona, the same mood. All 150 of them.
PairedEach seat is compared with itself, so a change in the mix of agents cannot pass for a change in the page.
Noise, measuredRead the same page twice and the engine says how far the count moved on its own. A re-run is how that noise is measured, not guessed.

Products

Six products.

Read a page, iterate it, set it beside rivals, profile the company behind it, read it three ways, capture the evidence, see the ads.

ILfree account
Insight Loop
Change one section. The same 150 agents read it again. The number moves or it doesn't.
Your web address
Fills in the form at the top of this page. Free account, then the run. More on Insight Loop →
CBfree account
Competitor benchmark
Your page and up to three rivals, read by the same agents. A count and one sentence each.
Your web address
Fills in your address at the top of this page. Rivals go beside it on the benchmark page. More on Competitor benchmark →
CPruns here
Company profile
Every Companies House filing downloaded and read. Facts with a source. A corpus you can take into your own AI.
Company name, number or website
The Quick profile is free. No account. More on Company profile →
3Rruns here
Three readers
What people see, what a screen reader hears, what machines read. One page, three reads.
Your web address
One page, three reads, each with its own fix list. More on Three readers →
EPshown here
Evidence pack
Timestamped captures of what Google, ChatGPT, Claude, Gemini and Grok show for your buyers' questions, plus real respondents.
What a pack holds →
Captured inside Insight Loop, on credits.
TAfree account
Their ads
The ads each rival is running, which have run longest, what they lead with, and where they beat your page.
What you get →
A free account is needed.

Method

How a page is read.

Six steps from a web address to a finding. Each agent is a large language model briefed as one buyer: a forecast, not a survey.

01PageYour address is fetched once, the way a browser loads it. robots.txt is obeyed.
02SectionsThe page is split into the sections a reader meets, top to bottom.
03PersonasBuyer personas for your market, drawn from six buyer types in a cited research library.
04LayersEach persona carries decision factors, a Maslow need level, a Big Five profile and a mood.
05150 agents150 seats. Each agent reads section by section and stops where it would.
06FindingsWho would get in touch, where they stopped, why they left, what they would ask.

How the 150 are built →

Pricing

Free to start. Credits to keep going.

Credits on top for runs, comparisons and real-respondent studies (quoted). Developer access (API, connect from Claude) on every account. Enterprise on application.

Free
Free
Insight Loop. Free account. No card.
- Your web address. 150 AI agents read the page.
- Two questions to your guide.
- Where buyers stop, and what to test first.
Read my page
Your email is the whole sign-up.
Starter
£19a month
An account that keeps your work.
- 10 credits every month.
- Every check, query and test kept in your workspace.
- A/B tests: one change, the same 150 agents, a real difference or noise.
- Competitor pages read beside yours, on credits.
See what £19 buys →
Register free. Paid plans open soon; nothing is charged today.
Team
£195a month
More credits. Real people.
- 100 credits every month.
- Everything in Starter.
- Real-respondent studies through Prolific, quoted and set up by our team.
- 80 questions a day to your guide, against 12 on Starter.
See what £195 buys →
Register free. Paid plans open soon; nothing is charged today.

Plans, credit packs and costs →

Your page, read by 150 AI agents.

One web address. 150 seats. Free account, no card.

Read my page How it works