Target Edits by Highlighting in Canvas

A Canvas technique: select the specific span you want changed and give ChatGPT an instruction on it, so edits land precisely instead of regenerating the whole document.

In plain chat, any tweak re-emits the entire answer and can quietly alter parts you were happy with. Highlighting scopes the edit to a span, so the model changes that and leaves everything else untouched — you keep the good parts and iterate on the weak ones without collateral damage.

Revising a specific paragraph, tightening one sentence, or fixing a single section of a long draft or code block that's mostly right.

Wholesale rewrites or structural overhauls where you actually want the whole thing reconsidered — then a full regeneration is the point.

[In Canvas, with the target text highlighted] Rewrite just this selection to <specific change — shorter / more formal / fix the logic>. Leave the rest of the document exactly as is.

Highlighting one clunky opening line and asking for three punchier alternatives changes only that line, leaving the rest of a polished essay intact.

• Asking for a global change when you only meant one paragraph, and losing edits you liked.
• Not using highlighting, so every tweak risks the whole document.
• Highlighting too much, which reintroduces the regenerate-everything problem.