How we measure accuracy.
Most vendors publish a number and ask you to trust it. This page shows our number, how it is made, and where we fall short. Every figure here comes from claims our customers and design partners graded against what actually happened.
The run finishes. Every finding is written down as one specific claim, with a date.
The campaign runs. The outcome happens, and we have no say in it.
Your team marks the claim against your own data. We have no vote.
It stays on the ledger. Hit or miss, we do not remove it.
The grade feeds back into your model. Hits confirm a pattern. Misses correct one.
Where we stand today
These are from live accounts. Nothing is a pilot study or lab test.
Across all live accounts - customers & design partners. One blended number, not a best case.
Every call is marked HIT or MISS by the customer’s own team, against their own platform data. We never score ourselves.
Every claim that has been graded is in the number above. None removed.
The numbers are not fixed. Every graded claim feeds back into the model for that account. Hits confirm a pattern, misses correct one, and the committee that reads your page in three months has learned from every call your team graded in between.
Four rules
A number is only as good as the rules behind it. Here are ours. They do not change after the fact.
What each grade means
Four outcomes. Only two of them count toward the hit rate.
What we said would happen, happened. The role we named did block. The objection we predicted did come up.
We were wrong. It stays on the record with the original claim text and date, so you can read what we got wrong and when.
The evidence to settle it never arrived. The campaign was pulled, or the data was never captured. Excluded from the rate because there is no verdict, not because it went badly. 21 of the 473 fall here.
The evidence was too thin, so the system declined to make a claim at all. We log these too. A tool that always has an answer is a tool that is guessing.
Every call, against its bar
One overall number hides more than it shows. We are much better at some calls than others. Here is the split, including the row where we are below our own bar.
| By claim type | Accuracy | Our bar |
|---|---|---|
| Does this role block? Does this objection land? | 96.6% | 80% |
| What is already working on the page | 95%+ | 88% |
| Which fix to apply | 71.4% | 70% |
| Why it broke, the root cause | 67.5% | 75% |
Root cause sits below our bar. We would rather show you that than hide it inside an average.
| By engine | Accuracy | Our bar |
|---|---|---|
| Content pages | 94% | 78% |
| Landing pages | 83% | 78% |
| Email campaigns | 76% | 80% |
| Ad campaigns | 78% | 75% |
Email sits below our bar. We would rather show you than calling both “across accounts.”
What we do not claim
WhyUser predicts direction and rank order. Which role disengages, where on the page, and which of two variants wins. It does not do these things, and any vendor who says otherwise is guessing.
Do not take our word for it
The fastest way to settle accuracy is to grade us yourself.
Request accessPick a campaign whose ending you already know. A page that underperformed. An email that flopped. An ad set you killed.
We run it without seeing the result. The claims seal the moment the run finishes, with a timestamp you can check. Then you grade them against what actually happened.
You will get some hits and some misses. Both go on the ledger. That is the whole point: a track record you can audit beats a number you have to trust. More on how the product works in the FAQ.