---
title: "One word made a harmless 256 KB file tool disappear for one assistant only"
canonical: https://dxdev.com/everyday-ai/the-same-honest-answer-meant-something-different/
datePublished: 2026-06-11
---
A small tool of ours [vanished from ChatGPT](/blog/same-tool-different-lie/) the moment we described it as able to write.

Nothing was broken. The same tool kept working exactly the same way for claude.ai and for Manus, the other two assistants using it. Its only job was to create a brand-new file in one safe, boxed-in folder. It couldn't overwrite anything. It couldn't touch a file that already existed. It couldn't reach outside that one folder. It was capped at a small size. About as constrained as a tool can be while still doing something useful.

We had labeled it with the plain, technically accurate description for that kind of capability: it writes. That's true. It is also the exact same generic label you'd put on a tool that could delete a customer's entire account, because the label names the kind of action and says nothing about how narrow or safe this particular version is.

claude.ai and Manus both read that label as a hint and kept right on offering the tool. ChatGPT treated the identical, accurate label as an automatic reason to make the tool unavailable entirely, in chat and in voice. There was no in-between option and no "ask first" step. It was just gone. Same tool, same real behavior, same honest description, and three different outcomes depending on which system was doing the reading.

That's an easy trap to fall into. The instinct toward honesty says to use the same accurate label everywhere, make no exceptions, and play no word games depending on who's listening. But a label is a signal, and different systems weigh the same signal very differently. claude.ai and Manus treated "this can write" as useful context to keep in mind. ChatGPT treated it as a disqualifier, with no way to say "yes, but only this much."

We did not make the tool less safe, and we did not build a second, separate version of it. Every real constraint stayed exactly the same, enforced the same way, for all three: the boxed-in folder, no overwriting, the size cap. What changed was one word, aimed at one listener. The tool started describing itself as read-only to ChatGPT, while claude.ai and Manus kept seeing the plain, accurate "write" label. The tool's actual behavior never moved an inch for any of them.

## If you use the same label for more than one audience

Find the listener that reads your label most harshly. It might be an approvals system that hard-blocks on a single word, a cautious client, or a risk-averse reviewer. Then check what happens to your capability the moment that listener sees your most technically accurate wording. If the answer is "it disappears entirely, with no partial mode," you don't need to loosen the real limits to fix it. You need a second, still-honest word aimed at that one listener.

The tool's real limits never changed for claude.ai, Manus, or ChatGPT: one new file, one folder, no overwriting, a 256 KB cap. The only thing that changed was the single word ChatGPT was shown next to them.