ThinkAloud

What do people actually see on your page?

ThinkAloud sends a synthetic audience through your page in a real browser.
An attention model trained on eye-tracking data predicts where the eye lands first, and the personas react to that before they read on.
You get one report: what they see, what they understand, and where they drop off — predicted, before your real visitors arrive.

Try it on your own page See a real report →

Heat map over a product page: the colour sits on the headline and on the phone in the photograph.

The order is the whole difference

Most synthetic users ask a language model to read the page. But a page read in source order corresponds to no human experience at all.

A huge headline and a small footnote get the same attention from a language model. A human is moved by what the eye sees.

The eye lands somewhere before a word is read — and that decides what gets read at all. ThinkAloud runs in that order.

FirstThe eye lands An attention model trained on real eye-tracking data predicts where the eye goes in the first few seconds — before anything is understood.

→

ThenThe persona reads The persona reacts to exactly that: what do I notice, what do I think the page is about. Only then does it read on and navigate.

We do not recreate what happens inside a human head. We tell the persona where the eye landed — a human is never told.

The mechanism is not the same. The order is.

How a run works

You point it at a pageA public address, an HTML prototype or an image. You choose the purpose — conversion, communication or reference — and the test judges accordingly. A missing call to action is not friction on a page meant to inform.
The audience goes through itThe attention model predicts the first glance. The personas then read, navigate and think aloud in a real browser. If you have search queries from Search Console, the personas are built from those rather than from guesswork.
You read the reportHeatmap on top, then coverage, what works, and the problems in priority order — each with quotes, severity and the page it happened on. A PDF is included.

The four fields the report covers

They are the same four the rest of the method rests on. The report shows which fields work, which have problems, and what none of the personas mentioned at all — that last one is often the most expensive.

SafetyDare I do this?

MeaningCan I see why I would?

AbilityDo I know how it is done?

EffortHow much is in my way?

The framework is published research, not a house model — it is open at frictiontheory.org, and Research explains what it rests on.

The personas were satisfied with the page.Four per cent of the first glance landed on something you can click.

See a report before you decide. The one below is not a mockup. It is output from a real run on a public page — our own LMS product page — with personas built from actual search queries.

The think-aloud half found no convergent problems: both personas praised the pricing for being clear and specific.

The first glance said something else. And effort — one of the four fields — neither of them mentioned at all.

That is what the two instruments are for together. A language model that merely reads the page would have signed off on it.

Open the report (conversion test) →

What it costs

From €15 / mo.

Starter €15 a month with 600 credits. Pro €35 a month with 2,000 credits, and unused credits roll over one month. Bureau by agreement, for those who run a lot and want API access. No subscription: 400 credits for €15.

The plans contain the same thing: every analysis, every model, the report as a PDF. Billed in euro, prices exclude VAT.

A run draws on a balance, so you pay for how much you actually test, not for how many people have access.

You sign up yourself at thinkaloud.activero.net. New accounts start with 50 free credits, enough for two focus groups on your own page before you decide.

What do people actually see on your page?

Which page should be tested first?