Just Paste Your Prompt Twice
Pasting your exact prompt twice in the same message sounds dumb and works anyway. Here is the research on why repeating yourself gets better answers out of AI.
2026-05-20
Loading content...
Pasting your exact prompt twice in the same message sounds dumb and works anyway. Here is the research on why repeating yourself gets better answers out of AI.
Large language models pay the most attention to the very start and the very end of your prompt, so anything important in the middle can get lost. If you copy and paste your entire prompt twice in the same message, your real ask lands in both high-attention spots and the model reads your context a second time already knowing what you want.
I know how this title sounds. I mean it literally. One of the most effective things you can do to get a better answer out of an AI is to take your whole prompt, copy it, and paste it again in the same message before you hit send.
I do not mean sending the same prompt as two separate messages. I mean the exact text, twice, in one message.
This sounds dumb. It works anyway, and there is research that explains why.
Large language models have a documented weak spot called the "Lost in the Middle" problem. Researchers at Stanford found that models use information best when it sits at the very beginning or the very end of the prompt, and worst when it sits somewhere in the middle. They called it a U-shaped curve, and they noted it looks a lot like the way human memory favors the first and last items in a list.
So if your most important instruction is buried in the middle of a long message, that is where the model is paying the least attention.
Repeating your prompt fixes the positioning for free. Your real ask now shows up at the top and at the bottom, which are the two places the model is paying the most attention.
There is a second reason, and it is the more interesting one. A paper called "Re-Reading Improves Reasoning in Large Language Models" tested this exact idea: give the model the question twice before letting it answer. Across 14 datasets and 112 experiments, re-reading consistently improved the model's reasoning. The authors explain that the first pass gives the model a sense of the whole question, so the second pass gets read with that context already in mind.
That is the part worth sitting with. A model reads left to right. The first time through, it does not yet know where you are going. The second time through, it does.
Most of us are bad at asking questions. We tell a rambling story, pile on context, and only get to the actual question at the very end.
My 5-year-old son recently wrote a book and wanted me to record him reading it to his class. The story was wonderful and completely ridiculous. He read the whole thing, looked at the camera, and asked, "Now, what was the setting and who were the characters?"
I had no idea. I had been enjoying the story, not studying it. Nobody told me there would be a quiz.
That is exactly what we do to an AI. We narrate the whole situation and only reveal what we actually want at the end, after the model has already decided what to pay attention to. Repeating the prompt is like handing over the questions before the test. The model gets to read your context already knowing what it is looking for. It is the same thing the re-reading research found, explained by a kindergartner.
One of my favorite ways to get a better prompt is to use one AI to write the prompt for another AI. Those generated prompts usually do not repeat themselves. They do something more deliberate: they hint at the goal up front, lay out the details in the middle, and tie it together with a clear ask at the end.
That is still a great approach. If we were doing a consulting engagement to build out a real AI pipeline, that is how I would structure the system prompts.
But you do not always have time to engineer the perfect prompt. Sometimes you just want a faster and better answer right now. Pasting your prompt twice is the lazy version of all that careful structuring, and it gets you most of the way there with almost no effort.
Try this today: Write your next real prompt the way you normally would. Before you send it, select the whole thing, copy it, and paste it again right below. If you want to match the research even more closely, drop a short line between the two copies like "Read that again, then answer." Compare the result to what you usually get.
If you want to talk about structuring your AI workflows so you do not have to repeat yourself, reach out.
Now, what was the setting and who were the characters?