A transcript makes repeated words easy to spot. It does not tell you which repetition matters. A speaker may say a word twice while thinking, or repeat it deliberately to make a point. Listen before treating both cases as the same editing task.
Start with the listener's problem
Find a short answer where repeated thinking words make the main point difficult to follow. Mark only the clearest distractions on the first pass. Keep words that connect the answer to the question or show uncertainty that is important to its meaning.
Make one small cut, then play the full sentence. Listen for a clipped beginning, a sudden pitch change or an unnatural jump in room sound. If the cut needs several repairs, leaving a small filler word may be the better choice.
A text-based edit should still be checked against the audio. Transcription errors can make a useful word look like an extra one, and a repeated phrase may not be identical each time.
Keep the speaker recognizable
Compare a longer section before and after editing. The edited version should be easier to understand while still sounding like the same person. Fast, perfectly joined sentences are not automatically better podcast audio.
For a review handoff, mark one example of a substantial wording cut so the host can judge the style. Agree on that level of cleanup before repeating it throughout the episode.
Save a project version before a large transcript-edit pass. That makes it easier to restore the original rhythm if the result becomes too tight. Use the transcript as a map for decisions, with the recording as the final reference.