01
It starts from the real product
The teardown is built from what a visitor can actually see in the product — screenshots of live pages, or the ones you upload.
Grounds the teardown in what a buyer or evaluator can actually see.
Methodology
What you get from a teardown — and what it can't tell you
Four AI analysts read one product. In a teardown you run, a finding can open the screenshot behind it, and this page lists what we don't measure. Public samples carry the finding text only: see the readout on the Stripe sample.
What a teardown includes
- AI analysts
- PM · Marketing · UX · Design
- Evidence
- In your own run: the screenshot behind a finding, when attached
- Output
- Summary Card, findings, full canvas
What you get
Five things a teardown gives you
Each one comes with the reason it can be trusted.
For reading methods, explore the Guides.
02
Evidence first
In a teardown you run, a finding can open the screenshot it was read from when one is attached, so you can check a claim instead of taking our word for it. Public samples carry the finding text without the screenshots.
A claim you can check is worth more than one you have to trust.
03
Four AI analysts, one product
Four AI analysts read the same captured evidence, one each as PM, Marketing, UX and Design, and flag what that role would worry about. The overall read draws on all four; where they don't agree, you see it.
Where the analysts disagree is usually where the useful finding is.
04
Two numbers, clearly labeled
A UX score over up to six visible dimensions, and an overall read from the four AI analysts. A dimension we couldn't see is marked "Not measured", not averaged in.
A number that admits what it skipped is one you can defend in a meeting.
05
One readout, two views
Summary Card for the 30-second version, the full canvas when you want to argue with it. Both come from the same evidence.
Export the card as a PDF or share the canvas by link.
Output
What's inside the readout
The public Stripe sample shows the Summary Card and the findings as text. A teardown you run shows the findings with their screenshots, when attached, and the full canvas.
Screenshots
Evidence for what was visible in the captured product experience.
What we saw
Page types, components, pricing cues, onboarding signals, and other observable facts.
Four AI analysts
PM, Marketing, UX and Design, each with its own findings and score.
What we concluded
Summary Card, findings, and generated teardown views for the private app experience.
Trust
Trust boundaries
The methodology is useful because it is explicit about what the system can and cannot see.
coExploro analyzes what can be captured from the product experience.
It does not replace customer interviews, analytics, or usability testing.
It can miss gated states, account-specific flows, or content hidden behind access controls.
The UX score does not cover mobile behavior. Captures are taken at desktop width, so nothing in that number reflects how a product reflows on a phone. The personas still rate mobile responsiveness, but that is their read of a desktop capture, not a measurement taken on a phone.
It does not measure load speed. Perceived performance is scored only when a loading state happens to be visible in a capture, which is a design signal rather than a timing measurement.
Scores are decision support, not absolute truth.
Public sample pages are curated and separate from private share links.
Changes to how the UX score is computed
When the formula changes, past teardowns are recomputed with it, so scores stay comparable across time. A teardown you exported as a PDF keeps the number it had when you exported it and will disagree with the live page.
The score averages only what was actually measured
A dimension the model could not judge from the screenshots used to be averaged in at a neutral 50, which dragged every published score toward the middle. Those placeholders are now excluded from the average instead. Across the 50 teardowns in our reference corpus the median published score moved from 79 to 88.5 — most of the change on this page belongs to this step.
Mobile responsiveness retired as a scored dimension
Across the same 50 teardowns, mobile responsiveness took two distinct values and sat at the neutral placeholder on 49 of them. It was never a measurement, so it was removed rather than reported. The score now covers six dimensions, and what the screenshots cannot show is labeled "Not measured" instead of scored.
FAQ
Methodology notes
A few answers that help keep the output useful without overclaiming what the analysis can prove.
Is this raw AI output?
Public samples are not: a person edits each one and removes the recommendations before it is published. A teardown you run yourself is AI output: four AI analysts read what was captured from the product's pages, and nobody edits the result. Where a finding has a screenshot attached, you can open it and check.
Does coExploro replace user research?
No. It helps a team quickly understand a product experience, then decide where deeper research or customer evidence is needed.
What does the UX score measure?
The UX score summarizes six dimensions: visual hierarchy, navigation clarity, content quality, CTA effectiveness, trust signals, and perceived performance. A dimension the model cannot judge from the captured screenshots is shown as "Not measured" and left out of the score rather than counted as an average result.
Why do personas review the same product?
Because a pricing page reads differently to whoever sells the product and whoever has to use it. Each AI analyst reads the same captured evidence from one of those roles, so the four reads can be compared.
Are public samples connected to private share links?
No. Public samples use curated static routes. Private share links stay tokenized and out of the public sitemap.
See the method on a real sample.
Start with Stripe, then run the same format on a product you care about.