The Make No Mistakes Prompt: The Meme and What Helps

Adding "make no mistakes" to a prompt does not make your AI more careful, because it gives the AI nothing to act on. What helps is saying what done means, what must not change, and that "I do not know" is a fine answer.

Where does the joke come from?

The meme is a one-line request for something enormous with "make no mistakes" tacked on the end. A September 2026 article on ClaudeFolio describes how it became a running joke about vibe coders. We found no solid measurement of whether the phrase changes anything, so what follows is reasoning and advice from the people who make these tools, not a test result.

Why does it not work?

• There is no careful mode to switch on. The AI writes likely text based on what it was given. "No mistakes" does not say which mistakes, or where.
• It does not know what counts as a mistake in your game. Is a slow balloon a bug or a design choice? Only you know that.
• Pressure can backfire. The Claude Code docs say that piling on emphasis "mostly makes Claude over-apply the rule", and that phrases like think hard are ordinary text, not hidden switches.
• It leaves no room for doubt. If only perfection will do, "I am not sure" looks like failure. Anthropic's guide to reducing made-up answers does the opposite.

What helps instead?

That guide lists three basics: give the AI explicit permission to say "I don't know", have it work from exact quotes, and have it find something that backs each claim, taking the claim out if it cannot. It also warns that these reduce made-up answers but "don't eliminate them entirely". So you still play the result.

Here is the difference, using Cloudbank Couriers, an invented balloon delivery game.

The meme version:

Add weather to my game. Make no mistakes.

The version that gives your AI something to work with:

Add storms to my game. Done means: a storm crosses the map every two minutes, and any balloon inside it flies at half speed. Do not change fuel or the map.

The second one has a finish line and a fence, and every word of it can be tried out in the game. For more on backing up claims, read show me where that is written.

Paste this to your AI

Delete "make no mistakes" and use this instead:

Here is the change I want: [describe it].
Done means: [what I should see in the game].
Do not change: [what must stay as it is].

Before you build, draw what is supposed to happen as a small Mermaid diagram. Walk through every path in it, including odd cases, and show me any dead end or clash. Then tell me what you are unsure about. Build from the drawing once I agree.

If something is not in my project or in what I told you, ask me or say "I do not know". Do not fill the gap.

When you finish, list each claim you make about the result, such as "the storm slows the balloon". Next to each, put what it rests on: the place in the game, what I said, or what you tried and what happened. If a claim has nothing behind it, take it out and tell me you took it out.

Or skip this and connect the MCP: your AI then finds all of this by itself.Connect the MCP

How to check it worked

• Ask about something that does not exist. Ask what the weather is like on level 40 when you only have three levels. It should say there is no level 40.
• Spot-check one claim. Pick a line from the list and go to the place it names.
• Ask the same question in two fresh chats. If the answers differ, at least one of them is guessing. Anthropic's guide suggests comparing runs like this.
• Read the drawing for odd cases. What happens if a balloon is in a storm when it runs out of fuel? If the drawing does not say, your game will decide by accident.
• Play it. A list of claims is not a game.