Free

Our AI grades stricter than students expect

A common worry about AI scoring is that it flatters you. Our own feedback points the other way. Here is what real test-takers told us.

After a practice test, ParrotTalk asks one quiet question: how did the band you received compare with what you expected? These are the answers from 88 test-takers who left feedback, internal tests excluded, collected through 21 July 2026.

49%
said their band came out lower than they expected
42%
said it was about right
9%
said it was higher than they expected

Nearly half of test-takers found the AI harder on them than they thought it would be, and fewer than one in ten found it generous. For a practice tool, that is the useful direction: a score that leans strict prepares you for an exam that will not flatter you either.

When we asked what would help most, two answers came out on top: more practice tests, and more detailed corrections. That is the work we focus on. Every essay you send to the writing checker comes back with a full correction: what the band is, where it was lost, and the specific correction to make on grammar, vocabulary, task response and coherence.

We do not claim a measured accuracy figure yet, and we will not invent one. Your estimated band can still differ from an official result by roughly half a band to a full band. We are comparing ParrotTalk bands against real, official Test Report Forms to publish that gap openly once we have enough data. Until then, treat every band as a directional practice signal, not an official score.

ParrotTalk is 100% free, with no sign-up. Figures come from in-app feedback and will be updated as more responses come in.