# The Prompt Engineering Trick That Could Be Making Your AI Slower

DevFeed: [The Prompt Engineering Trick That Could Be Making Your AI Slower](<https://devfeed.tech/articles/the-prompt-engineering-trick-that-could-be-making-your-ai-slower-116246.md>)

Original publisher: [Read original article](<https://hackernoon.com/the-prompt-engineering-trick-that-could-be-making-your-ai-slower?source=rss>)

Author: Cameron M. Palmer

Published: 2026-11-04T08:37:25Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [Prompt Engineering](<https://devfeed.tech/topics/prompt-engineering.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [prompt](<https://devfeed.tech/topics/prompt.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cost](<https://devfeed.tech/tags/cost.md>), [data-science](<https://devfeed.tech/tags/data-science.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [hackernoon-top-story](<https://devfeed.tech/tags/hackernoon-top-story.md>), [llms](<https://devfeed.tech/tags/llms.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [prompt-engineering](<https://devfeed.tech/tags/prompt-engineering.md>), [prompting](<https://devfeed.tech/tags/prompting.md>), [software-development](<https://devfeed.tech/tags/software-development.md>), [speed](<https://devfeed.tech/tags/speed.md>), [study](<https://devfeed.tech/tags/study.md>)

## AI overview

In a 5,000-question MMLU-Pro experiment across four models and five system-prompt conditions, expert personas did not reliably improve accuracy and often increased reasoning-token use, cost, or latency. Effects varied by model: Deepseek used more reasoning tokens with every expert persona and with no system prompt, while any system prompt improved its accuracy over the blank baseline. Ling used 12%-36% more reasoning tokens across the compared conditions and was about 15%-25% slower with expert prompts. Claude showed a statistically significant 8% token increase only for the longest expert prompt; GLM showed no statistically clear token change. The author favors neutral prompts for straightforward production requests, while cautioning that the results may not generalize to agentic or coding tasks.

## Source excerpt

A study of expert persona prompting on cost and speed.