Is dictated text still your own writing if AI cleans it up?

Usually yes, but two things can change the answer: how much the tool rewrote, and what your rules say. Under US copyright law, using AI as a tool to assist your creativity does not by itself take away protection for the output. Schools and employers set their own tests, and some do not allow AI to rephrase whole sentences or paragraphs. The question is less about dictation as a method and more about how much of the final wording came from a model, and whether you can say so when asked.

Copyright asks whether a person made the creative choices

The US Copyright Office says that using AI tools to assist human creativity, rather than to stand in for it, does not affect whether the output can be protected ¹. Its January 2025 report also says that, on current technology, prompts alone do not give a user enough control over the output to make that user its author ².

Those two findings point the same way for dictation. When you decide what to say, speak it, and approve the result, the choices are yours. A cleanup pass that removes fillers and fixes spelling is closer to an assistant than to a co-author. A draft that a model writes from a short spoken prompt is much further from your authorship.

This is a copyright test, not an academic one. The Office's report is about protectable expression under US law. It does not define plagiarism, and it does not say what your teacher or employer must accept.

Meaning can change even when the words look like yours

A cleanup pass can change more than spelling. A 2026 preprint on large language models found that even when an LLM is asked only for grammar edits, it still changes the text in a way that significantly alters its semantic meaning ³. That study tested language models on written text, not on dictation, so it does not measure a dictation app's output. It is still a reason to read the cleaned version against your own speech.

The check is simple. Compare the cleaned draft with the verbatim transcript, and ask whether each claim, qualifier and example is one you would stand behind. If the tool dropped a "probably" or made a hedged point sound certain, the ideas are no longer exactly yours, even if every word was typed by you.

School and employer rules turn on rephrasing and disclosure

The policies I read turn on two questions: whether a model rephrased your words, and whether you can declare that you wrote the work. The table below lists the two I read.

Neither passage addresses speech recognition. A policy that counts only generative AI may treat a recognizer as outside its scope, but a policy that asks for disclosure of any tool may not. Read the rules where you study or work, and do not assume that dictation falls outside them.

When your authorship is weaker

• The cleanup rewrote whole passages. The more the wording came from a model, the weaker your claim to it, and the more likely a policy about rephrasing applies.
• The cleanup changed the meaning. Then the ideas on the page are not exactly the ones you chose.
• You dictated a prompt and kept the model's draft. The copyright report's finding on prompts applies here: the instruction is yours, but the output may not be.
• A policy asks you to declare tools and you did not. The Burren policy treats the duty to declare authorship as separate from whether a tool counts as AI, so skipping the declaration can be a problem even when the tool itself is allowed.

What to do

1. Decide what the tool did. Note whether it only recognized your speech, added marks, or rewrote sentences. The how to turn off AI rewriting steps show how to limit each change.
2. Keep the verbatim transcript and the draft history. It shows what you said and how the text changed. No source I read says this is required, but it is useful evidence if your authorship is questioned.
3. Read the cleaned text against your speech. Restore any claim or qualifier the tool removed, and delete any sentence you did not mean.
4. Check the policy and disclose if it asks. Name the tool and say what it did, such as "dictated, with punctuation added by the tool".

For the related question of whether an AI detector can flag dictated text, see whether AI detectors can flag dictated text. For what cleanup changes in your words, see whether AI dictation changes what you said. The guide is speech to text without AI, and the subject is covered in voice and AI. For the term, see the generative AI glossary entry.