Does it match real people?
The claim and the pass mark were written down on 20 October 2026, before anyone took part. The result goes on this page, pass or fail.
The claim
DesignIQ names the same easier version as real people on at least 7 of 8 screen pairs, and does so more often than a check that only compares contrast.
How the study works
Eight pairs of screens, built as plain HTML pages. In each pair, version B differs from A by one deliberate flaw. Real people do the task in the browser, each getting A or B for each pair at random, and the page logs success, time and wrong clicks on its own. Nobody is told which version they are on, and the study pages carry no DesignIQ branding.
Because the pages are ours, we know exactly where every element is. That lets us score how well the model reads screens separately from how well it predicts people.
What it has to beat
Three simpler checks: a coin flip, “pick the version with higher contrast on the target”, and “pick the version with the larger target”. If DesignIQ can’t beat the contrast-only check, the simulation adds nothing, and this page will say so.
The eight pairs
The predictions are in. Real people: pending.
| Pair and flaw in B | Task | DesignIQ predicts | Real people |
|---|---|---|---|
| 1. Tiny button | Book a table for two | A is easier, 100 vs 96 | Pending |
| 2. Faint link | Reset your password | A is easier, 96 vs 77 | Pending |
| 3. Competing buttons | Continue to payment | Too close, 100 vs 100 | Pending |
| 4. Placeholder-only labels | Sign in | A is easier, 100 vs 92 | Pending |
| 5. Buried call to action | Start a free trial | Too close, 100 vs 100 | Pending |
| 6. Icon-only action | Download the October invoice | A is easier, 92 vs 60 | Pending |
| 7. Low-contrast button | Subscribe to the newsletter | A is easier, 100 vs 94 | Pending |
| 8. Icon-only search | Search for running shoes | A is easier, 98 vs 84 | Pending |
Limits, stated up front
The first round of participants are aged roughly 19 to 23, so the older-adult persona is untested by this round. We chose the pairs ourselves. About 15 people and 8 pairs is enough to catch a model that is plainly wrong, not to prove one right.
When the study is done, the raw session log is published here as a CSV.