Can AI detectors flag dictated text?

Yes, a dictated draft can be flagged, because a detector only sees the words. I found no study that tested dictated text against an AI detector, and no detector vendor has said it treats speech input differently. The rate for dictated text is therefore unmeasured. What the sources do show is how unreliable the scores are for any text, and that matters more than the input method.

Detectors score wording, so dictation adds no separate signal

An AI detector receives a block of text and estimates how much of it resembles model output. It cannot see whether you typed the words, spoke them into a microphone, or pasted them from a notes app. A recognizer's output is ordinary text, so the same detector gives the same kind of score to a dictated draft and a typed one.

That means the question is really about wording. Verbatim output stays closest to what you said, so it reads like your speech. A draft that an AI pass has rephrased may read more like model output, because smooth, even sentences are one of the things detectors look for. That is an inference from how the tools work. I have no measured result that shows it, so treat it as a risk to check, not a fact.

The detectors' own tests show they miss AI text and flag human text

The most direct evidence comes from the companies and researchers who built the tools. The table below lists the results I read.

OpenAI's withdrawal note is the clearest statement. The company did not say its tool was slightly weak. It said the tool was no longer available because its accuracy was too low. Turnitin's detector is a different product, and its later versions may behave differently, but its own warning about low scores applies to any score that sits near the bottom of its range.

Where the risk is higher or lower

Short drafts are the weakest case. OpenAI said its classifier was unreliable on short texts, below 1,000 characters ⁶. A short dictated note or email is therefore a poor test of anything, in either direction.

Writers whose first language is not English may be flagged more often. The Stanford study found that the detectors it tested misclassified non-native English writing as AI-generated ⁵. That is a warning about the detectors rather than about speech. A heavily cleaned draft carries more risk than a verbatim one. If an AI pass has rewritten the sentences, you have added a second source of model-like wording. Keep the verbatim transcript as evidence of what you said. For how to switch off that pass, see how to turn off AI rewriting in a dictation app.

What to do if your dictated draft is checked

1. Keep the raw transcript. Save the verbatim output before any cleanup, so you can show what the recognizer produced. This is useful evidence of your drafting process, though no source says a school will accept it.
2. Ask which detector is used and how the score is read. A score is a prompt for a conversation in some institutions, not a finding. Ask what the institution does with a score below its threshold, and whether it has a process for appeal.
3. Write the way you speak, then edit by hand. Edit the draft yourself, sentence by sentence, instead of accepting a single rewrite. That keeps the wording in your own voice and leaves a record of your choices.
4. Do not rely on a detector to prove authorship. The detectors' own makers have published results that show errors in both directions.

For the wider question of whether cleaned-up dictation still counts as your writing, see whether AI-cleaned dictation is your own writing. For how much a cleanup pass changes your words, see whether AI dictation changes what you said. Our overview of dictation without an AI rewrite is in the guide to speech to text without AI, and the subject is covered in voice and AI. Terms used here are defined in the AI text detector glossary entry and the verbatim transcription entry.