What does 'sounded like' mean on my result?

It means the specific sounds you produced, evaluated on their own, happen to match an actual different English word — a plain-language shortcut sitting on top of the phoneme comparison, there specifically for the cases where seeing the raw IPA substitution wouldn't land as quickly as just hearing the real word it became.

30-second version: If you're reading "think" and the recording's acoustic evidence lines up with /s ɪ ŋ k/ instead of the target /θ ɪ ŋ k/, the result doesn't just show the phoneme substitution abstractly — it shows "sounded like 'sink'" directly, because that's the concrete, real-word consequence of the substitution. It's not a transcription engine guessing what you probably meant; it's the reverse — the phoneme evidence gets checked against the dictionary after being measured, purely to give you a plain-language handle on the error.

Why this is different from autocorrect

This is worth being precise about, because it sounds superficially similar to what a dictation app does, and it's actually the opposite mechanism. A transcription model looks at ambiguous audio and a language model picks the most likely word, then writes that down as if it's what you said — which means it can silently repair a real mispronunciation into the intended word and never tell you an error happened at all. That failure mode is covered fully in the autocorrect problem.

The "sounded like" hint works the other way around. The phoneme evidence is measured first, completely independent of what word was intended — sayit already knows the target was "think," and it still measures /s ɪ ŋ k/ as what came out. Only after that measurement does it check whether the produced sequence happens to spell a real word, purely to make the feedback easier to grasp quickly. The verdict and the score are based on the phoneme comparison either way, whether or not the substitution happens to land on a real word.

When there's no hint at all

Not every mispronunciation produces a real word — plenty of substitutions land on a sequence that isn't any English word at all, in which case there's simply no "sounded like" hint, and you rely on the raw target-vs-produced phoneme comparison instead. That's not a gap in the feedback; it just means this particular shortcut doesn't apply to every error, only the subset that happens to collide with real vocabulary. Both cases get the same underlying accuracy scoring regardless.

Why the hint is worth paying attention to

Some substitutions matter more than others, and whether the result happens to be a real word is a reasonable, if imperfect, signal for which ones. A "sounded like" hint means a listener in a real conversation could plausibly hear a different, specific word instead of the one you meant — which is a more concrete kind of misunderstanding than a mispronunciation that just sounds slightly off but doesn't collide with anything else in the dictionary. If you're prioritizing which flagged words to fix first, ones with a "sounded like" hint are a reasonable place to start, since they're the errors most likely to cause an actual mix-up rather than just an accent impression.

How to use it in practice

Treat the hint as a comprehension check on the substitution, not a separate piece of feedback. When you see "think — sounded like 'sink,'" say both words back to back, listening for exactly what changed between them, then go drill that specific contrast — think vs sink in this example — rather than re-attempting the original word in isolation, which doesn't isolate the actual contrast the way the minimal pair does.

Try it

Record: "I think the sink needs three thin pipes." Open sayit free — no install, no card — and check whether any word in your result comes back with a "sounded like" hint. If it does, that's the specific contrast worth drilling next, not a general note to "work on pronunciation."