Why Find-and-Replace Fails on Text You Pasted From AI

You can see the word. It is right there in paragraph three. You press Ctrl+F, type it exactly, and Word tells you there are no matches. You retype it slowly. Still nothing. You start wondering whether you have lost the ability to read.

You have not. There is a character inside that word which renders as nothing, and your search — which compares strings exactly — is correctly reporting that the thing you typed does not appear in the document.

Why it happens

Find-and-replace does exactly one thing: it compares the sequence of characters you typed against sequences in the document. It has no concept of what the text looks like. If the document contains analy + U+200B + sis and you search for analysis, those are two different strings, and the search is right to fail.

The invisible character got there because you copied from something that renders HTML — most often a chat interface. The copy path flattens rendered formatting into plain text, and characters that were doing layout work in the browser land in your document as literal content. There is more on the mechanism in ChatGPT hidden characters.

The same failure has a second, sneakier cause: space-like characters. If a phrase contains a non-breaking space (U+00A0) where you expect an ordinary space, then searching for results show fails while the phrase sits there looking completely normal. This one is nastier than the zero-width case because the character does render — as a space, identical to the one on either side of it.

Confirming it in thirty seconds

Pick the word that will not match and try these in order.

Arrow-key through it. Put the caret before the first letter and press the right arrow one keystroke at a time. Count the presses against the letters. If the caret stops somewhere without visibly moving, there is a zero-width character at that position.

Search for half of it. If analysis finds nothing but analy finds it immediately, the break is right after the part that matched.

Search for the character itself. In an editor with regex support — VS Code, Sublime, Notepad++, or Word's own wildcard mode — search for \x{200B} or the range [\x{200B}-\x{200D}\x{FEFF}\x{00A0}\x{202F}]. Every hit is an invisible character. In Word specifically, turning on formatting marks with Ctrl+Shift+8 will reveal non-breaking spaces as a degree-like symbol, though it will not show you zero-width characters at all.

Paste it somewhere that shows bytes. Any hex viewer, or a browser console with [...str].map(c => c.codePointAt(0)), will show you the truth immediately.

The fix that does not work

Retyping the word by hand fixes that one instance. It does not fix the eleven other places in the document with the same problem, and you will not find those until each one wastes its own half-hour. The characters arrived in bulk with a paste, so they are distributed through everything that came from that paste.

Manual find-and-replace on the character itself gets closer, but has two failure modes people hit constantly:

1. You cannot type the character into the search box. You have to paste it from somewhere or use regex mode, and most people give up here.
2. You delete the space-like ones. Removing a non-breaking space rather than replacing it with a normal space welds two words together — resultsshow — and now you have introduced a new error that your spellchecker will catch but your reader might not.

That second point is the reason a naive strip-all-invisible-characters script is worse than doing nothing. The correct operation is different per character: delete the zero-width ones, substitute an ordinary space for the space-like ones.

The fix that works

Applying that table to a whole document in one pass is what the cleaner on the PassMyEssay homepage does before it does anything else. Paste the document, hit Clean, copy the result back. Every character in that table is handled with the correct action for its type, and your text is otherwise untouched.

If you would rather see the problem before fixing it, switch to AI Phrase Detector — it tags every hidden character where it sits, so you get a map of your own document rather than a silent repair.

While you are in there

The invisible characters are the mechanical problem. There is usually a second one riding along in the same paste, and it is the one your reader actually notices: the phrasing. Text that came out of a chat window tends to open with a stock connective, run em dashes at three times the natural rate, and hold every sentence at the same length.

None of that breaks your search function. All of it changes how the document reads. Related: why AI writing sounds robotic, how to edit an AI paragraph.

Try PassMyEssay free → Fix every hidden character in the document at once, not one word at a time.