There are times when we think AI is changing the world… And then we discover that one of the most effective optimizations is simply… repeat the instruction twice.
Yes, exactly like when you're talking to someone who seems to be listening, but you'd rather "play it safe." The idea is simple: instead of giving your instruction just once, you repeat it exactly the same way, right after. And surprisingly, it often works better than you'd think.
Why does repetition boost AI?
Language models (LLMs) do not read a prompt like a human. They process a sequence of "tokens", with priorities that vary according to the structure of the text.
In the reality of work, we don't always produce "clean" prompts. We pile on the constraints: a tone, a format, a target audience, prohibitions… The classic result: the template overlooks a detail along the way. You asked for "no list," it includes one. You wanted "short," it pulls out an encyclopedia.
According to an analysis of Journal of the Net, this repetition technique (often called Repeat Prompting) allows for a drastic reduction in inattentional errors in models, a method validated by Google researchers themselves to improve the accuracy of responses.
In what cases is this technique essential?
This tip is particularly effective for highly structured tasks:
-
Strict formats: JSON, tables, numbered steps.
-
Length constraints: Maximum number of characters or words.
-
Editorial style: Specific tone, absence of emojis, prohibition banners.
-
Ready-to-use outputs: LinkedIn posts, scripts or summaries without introductory "blah blah".
The more constraints there are, the higher the risk of forgetting. Repeating the instruction is like putting up additional safeguards.
How can you use it without adding to your workflow?
The goal isn't to write a novel, but to solidify your framework. The simplest method:
-
Write your prompt as usual.
-
Copy the critical instructions line exactly at the end.
Example:
“Task: Write a LinkedIn post about UX. Tone: Professional. 1200 characters max. No lists. Short conclusion. REMINDER: 1200 characters max, no lists, short conclusion."
What this actually changes
There's a lot of talk about increasingly powerful models or complex reasoning. But in everyday life, the real issue is... regularityWhat wastes time is not the “worthlessness” of the model, but its lack of consistency.
Gaining even one iteration of correction is a huge victory in a workday. Repeating your prompt might seem like a simple technique, but it's incredibly effective. It's not futuristic; it's simply practical.
Philippe Elovenko, Product designer at UX-Republic
