ChatGPT’s edits going too far? This prompt keeps them surgical | usagoldmines.com

With the right prompt, you can get ChatGPT to act as a smart “search and replace” tool, snipping out unwanted words or making other document-wide edits.

The problem, of course, is making sure ChatGPT (or Claude or Gemini) doesn’t make other, unwanted edits, subtly tweaking your words or otherwise altering your voice. Indeed, if it does too much rewriting, ChatGPT might end up making you sound like AI.

Those using ChatGPT and other AI chatbots have long grappled with how to keep the models from altering passages of their original code or text when requesting edits, including giving the AI rules like “don’t change anything else.”

I decided to take up the challenge, working with Claude Opus 5.5 and OpenAI’s GPT-6.1-Sol to devise a “surgical edit” prompt that directs an AI to make the precise edits you choose while leaving the rest of your prose untouched.

Now, is it worth using this prompt for a single edit in a document? In that case, no. Just make the edit yourself. But if you want to make a series of targeted edits within a lengthy passage of text, the “surgical edit” prompt could come in handy.

There are two versions of this prompt: a short-and-sweet version and a longer version that makes the AI detail all the edits it made.

First, the short version:

[Your edit]. Keep everything else exactly the same. Return the full text.

Now the longer version:

Edit only what I name below. Preserve every other character, including punctuation, spaces, and line breaks, except punctuation or spacing that belonged only to a deleted word. No smoothing, no tone fixes, no shortening.

Change: [what to change, and how]

Return the full text, then list every change you made. If the same change repeats, list it once with a count. If I specify all occurrences, apply the edit to every match. If the target or scope is ambiguous, ask before editing.

[PASTE YOUR TEXT]

I tested these prompts with a range of more powerful “thinking” models, including ChatGPT (standard chat mode) on “high” reasoning, Claude Opus 5.5, and Gemini Argon. 

For my trials with the prompt, I used a synthetic 1,500-word voice-memo transcript filled with “um’s,” “ah’s,” and “you know’s” (“Okay, um, voice memo for the week, it’s Sunday, uh, November 8th“), and asked the models to “remove the filler words from this transcript” and to “return the full text.” When the models asked what counted as “filler words,” I said to “use your own judgement.” 

In my initial control runs without the “surgical edit” prompt, the models were far more likely to make “collateral” edits throughout the document, formalizing some language (for instance, turning “gonna” into “going to”), erasing hedges (like “kind of agree” and “honestly”), and rewording casual phrases (turning “frame it like” into “frame it as”).

The “high” version of ChatGPT made anywhere from 25 to 35 of these collateral edits in my test transcript when I hadn’t first used the “surgical” prompt, while Gemini Argon went from a low of seven unwanted edits to a high of 51. Claude Opus 5.5 was more careful without the “surgical prompt” guidance, only committing between six and a dozen unauthorized edits.

Once I started the chat with the surgical edit prompt, those unwanted edits dropped to almost nil. Both Claude Opus 5.5 and Gemini Argon stopped with the unsanctioned edits altogether, while ChatGPT still did a handful of extra cuts (chopping a “let me think” from one sentence, for example) that could’ve gone either way.

Interestingly, both the longer and shorter versions of the “surgical” prompt performed equally well in terms of keeping extraneous edits in check. But the expanded prompt does offer a summary of what the AI changed, making it easier to review the edits. 

The full prompt is also more likely to make the AI ask you what counts as filler (five times out of 6 for the full prompt, versus never after 9 runs of the shorter prompt).