Garbled output — an encoding problem, not an AI problem
If the AI's response shows genuinely broken character sequences rather than a clean substitution, the accented text was almost certainly already corrupted before you pasted it in — commonly from the exact Word-to-web encoding collision described in our copy-paste encoding guide. The model processes whatever text it received; broken input in, broken-looking output out. Check your source text for correct rendering before pasting, not just after.
Missing accents — a tokenization and generation effect
Some AI systems occasionally normalize or simplify accented characters during generation — more so in longer generated passages than short direct quotes, and more often for less common accented forms than high-frequency ones. This isn't a copy-paste artifact — it happens inside generation itself, and is a known, ongoing limitation across language models rather than a bug specific to one tool. If your workflow depends on precise accented output, review generated text character-by-character.
The fix that works regardless of cause
Don't try to fix mangled or dropped accents by retyping them from memory inside the same broken context. Instead, retype the specific characters cleanly using a reliable input method — our symbol picker, accent typer, or text cleaner, or explore our letters with accents hub and keyboard shortcuts (Windows Alt codes, Mac special characters) — and paste that clean text back in, overwriting the corrupted or simplified version.
When it's worth flagging vs. just fixing
A one-off dropped accent in a casual chat is easiest to just retype. If you're running accented text through an AI tool as part of a repeatable content pipeline — translation, transliteration, batch processing — verify the output against the source on every run rather than spot-checking occasionally, since the failure is inconsistent by nature.