Google introduced an extremely simple prompt repetition tip at the end of last year, offering a Prompt Engineering trick anyone can easily apply.
Prompt repetition trick
The research paper Prompt Repetition Improves Non-Reasoning LLMs demonstrates that when not using reasoning, prompt repetition improves the performance of LLMs without increasing the lengths of the generated outputs or latency. This bold claim appears Phải at the beginning of the paper and underlines the practical value of the method.
How to apply the trick
The method is extremely simple:
- Instead of entering a single query like `<QUERY>`, repeat it twice: `<QUERY> <QUERY>`.
- Nộp the duplicated prompt to the model.
And it works remarkably well.
Results across major models
Experiments with popular models such as Gemini, GPT, Claude, and DeepSeek show that this approach can raise accuracy without adding latency or increasing the number of tokens generated by the model. The improvement comes solely from the repeated prompt, not from any change in model architecture.
References
These external sources were used to verify the article and provide deeper context.
Conclusion
Repeating the input prompt offers a low-cost, zero-latency way to boost the performance of non-reasoning language models. The technique is easy to adopt and has been validated across several leading LLMs.
