Start free

Last run 26 Oct 2026 · updated 27 Oct 2026

How accurate is untick?

On our latest run of 100 shops we’d never scanned, 1 critical finding out of 44 checked by hand was wrong, and 2 needed a closer look. Every miss and its fix is listed below.

100shops we had never scanned
44critical findings checked by hand
41were right
1were wrong

Benchmark runs

100 shops never scanned before

26 October 2026

86 of 100 shops scanned; 11 blocked us or didn’t load and failed with no grade. Checked by hand: critical findings on 44 shops.

Right: 41 (93%) Can’t tell: 2 Wrong: 1

What we got wrong, and what changed

A cross-border checkout service (Global-e) was counted as tracking before consent. Now on our list of functional services, never counted as tracking.
A consent tool logging the visitor’s own choice was listed as a tracker. Consent tools’ own logging is now treated as functional.
An analytics SDK on an OpenAI domain was shown as an “unlisted service”. Named as Statsig (product analytics).
Two parked domains on our test list were scanned and given a grade. Parked and “domain for sale” pages now fail the scan with no grade and no charge.

50 shops in 14 EU countries, never scanned before

20 October 2026

41 of 50 shops scanned; 9 blocked us or didn’t load and failed with no grade. Checked by hand: all 40 serious (critical or high) findings.

Right: 39 (98%) Can’t tell: 0 Wrong: 1

What we got wrong, and what changed

“Allow only technical cookies” wasn’t read as a Reject button, so we reported “No Reject on the first layer”. “Only technical / functional” now counts as Reject in every language we cover.

Accessibility on 90 shops (983 serious issues)

20 October 2026

90 of 100 shops scanned; 10 blocked us or didn’t load and failed with no grade. Checked by hand: a sample of issues per check, judged against the screenshot and HTML.

Right: 29 (44%) Can’t tell: 20 Wrong: 17

What we got wrong, and what changed

“Focused item is covered” was wrong in 6 of the 10 we sampled (open drawers, invisible inputs). No longer counted when the item is invisible, inside the covering bar, or under a large drawer or pop-up.
Our keyboard checks (reachable by Tab, text cut off at 200%) were about half right. Look-alike hidden copies, card text and star glyphs are no longer flagged.
Bot-check pages were checked as if they were the shop. Bot checks are detected and the scan says so, with no score.
A score of 100 when no page could be checked. No page checked now means no score and “Accessibility not tested”.

What can’t a scan see?

  • We test like a first-time visitor from the outside. Server-side tracking and what happens after checkout aren’t visible to us.
  • Some shops block automated browsers or show a different page outside their country. Those scans fail and say so, with no grade and no charge.
  • AI-read findings (wording, link purpose, alt text) are marked “likely” or “Check by hand”, and never counted as certain.
  • A report is evidence of what we saw on that date. It isn’t legal advice.

How each check works is on the methodology page. The trackers we recognise by name are on the tracker list.

Common questions

How do you check accuracy?

We scan shops we have never scanned before, with no code changes during the run. Then a person compares each serious finding with its own evidence: the exact request, the cookies, and the screenshot. Each is marked right, wrong, or can’t tell.

What happens when a finding is wrong?

We fix the cause, not the one shop: a rule, a vendor name, or a page we shouldn’t have graded. The fix goes into our end-to-end tests so it can’t come back unnoticed.

Why do some scans fail?

Some shops block automated browsers, are too slow, or refuse the connection. We say so and give no grade, rather than grading a block page. A failed scan doesn’t use up your scans.