What do I do when the app flags a word I said right?

Check the recording conditions and the specific verdict before assuming either the app or your own ear is right — a handful of known, specific conditions make a false flag meaningfully more likely, and most disagreements trace back to one of them rather than to the scoring being simply broken.

30-second version: No acoustic model has perfect certainty on every recording, and sayit's scoring is built to be cautious about that: a flagged word only reaches the strongest "you mispronounced this" verdict (check) when the evidence is both serious and confident, and it deliberately softens to unclear when the acoustic confidence is low rather than forcing a call it isn't sure about. If you're seeing a firm check on a word you're confident about, the most useful next step is looking at why — the recording conditions, the word length, the specific sound flagged — rather than dismissing the flag or accepting it blindly.

Conditions that make a false flag more likely

Background noise or a weak microphone. The acoustic model reads the actual waveform, so anything that degrades that signal — room echo, background chatter, a laptop's built-in mic picking up more room than voice — genuinely reduces the evidence quality, and a genuinely unclear signal can produce a genuinely uncertain verdict. This is common enough to warrant its own dedicated guide and microphone-specific tips.

Very short words. Words below a small phoneme-count threshold are deliberately excluded from the strongest check verdict — they resolve to good or unclear only — precisely because short function words carry too little acoustic evidence for a confident, specific accusation to be fair. If a short word is flagged, it will never be flagged as harshly as a longer one, by design.

Genuinely ambiguous accent-neutral variation. Some sound differences are legitimate regional variation rather than errors — the classic case is cot and caught, which are distinct words in some English accents and merged in others. If a flagged sound is one with well-documented accent variation, that's worth knowing before treating the flag as a correction rather than a difference.

How to check which one applies

Re-record the same word in a quieter environment, closer to the microphone, and compare the two takes directly. If the flag disappears with better recording conditions, that's your answer — the original flag was a signal-quality issue, not a pronunciation one, and it's worth fixing your setup rather than your speech.

If the flag persists under clean conditions, look at the specific substitution named rather than just the overall verdict. A specific, plausible substitution (target /θ/, produced /s/, for a learner whose background doesn't have /θ/) is much more likely to be a genuine, correctable pattern than a coincidence. If the named substitution doesn't match any pattern you'd expect from your accent or background at all, that's more likely a measurement fluke on that specific take — try the word again in a fresh recording before concluding anything.

What not to do

Don't treat one flagged take as a final verdict, and don't treat one clean take as proof a sound is fixed either — a single recording is one sample, and the useful signal is always a pattern across several takes, not any individual result. This is the same reasoning behind checking a sound across multiple sessions rather than trusting a single score, covered in how to read a per-word IPA verdict.

Drill it

Record: "I think the sink needs three thin pipes." Record it twice — once in your normal environment, once as close to the mic as comfortable in a quiet room — and compare the two results directly rather than trusting either one alone.

Try it

Open sayit free — no install, no card — and re-record a word you disagreed with under cleaner conditions. If the verdict changes, that's useful information about your setup; if it doesn't, that's useful information about the sound.