From reciting “Star Trek” catchphrases to offering imaginary tips, the strategies users employ to coax better results from Artificial Intelligence (AI) are becoming increasingly bizarre.
However, as technology evolves, experts suggest that the secret to a better response may have less to do with being polite and more to do with treating the machine like a tool rather than a person, reports BBC.
‘Star Trek’ factor
Research into “positive thinking” and AI has yielded unexpected results. While calling a chatbot “smart” or telling it “this will be fun” failed to produce consistent improvements, one unusual technique stood out: when a Large Language Model (LLM) was asked to pretend it was on Star Trek, its performance in basic maths improved.
Rick Battle, an applied machine learning engineer at Broadcom who co-authored the study, noted that in the early days of these models, results were often a “crapshoot”.
However, modern iterations like OpenAI’s ChatGPT, Google’s Gemini, and Anthropic’s Claude have become significantly better at identifying the core intent of a prompt, making such linguistic “hacks” largely obsolete.
Cost of courtesy
Despite the shift in technology, human habits persist. A 2025 survey by the publisher Future found that 70 per cent of users are polite to AI, with 12 per cent admitting they do so to protect themselves against potential “robot uprisings”.
This social nicety comes with a price tag. When a user on X (formerly Twitter) questioned the electricity costs of people saying “please” and “thank you”, OpenAI chief executive Sam Altman replied that it was “tens of millions of dollars well spent”.
Conflicting evidence
The scientific community remains divided on whether politeness works. A 2024 study suggested that LLMs provided more accurate answers when asked politely, though cultural differences appeared; Japanese chatbots performed worse when addressed with excessive courtesy. Conversely, another test found an earlier version of ChatGPT was more accurate when it was insulted.
“It’s not about word choice,” says Jules White, a computer science professor at Vanderbilt University. “It’s about how you fundamentally express what you’re trying to do”.
How to talk to a robot
For those looking to maximise efficiency, experts recommend several practical techniques:
- Ask for multiple options: Instead of one answer, request three or five to force human re-engagement with the content.
- Provide examples: Rather than giving a list of “dos and don’ts” for writing, provide previous emails or samples to establish a style.
- The interview method: Tell the AI to ask you questions one at a time until it has enough information to complete a task, such as writing a job description.
- Avoid factual role-play: While role-playing is effective for brainstorming or creative tasks, Sander Schulhoff, an entrepreneur and researcher, warns it can be “dangerous” for factual queries. It can encourage “hallucination” by making the AI act too confident in its internal knowledge.
Human element
While being nice may not make a model more accurate, it may still be worth doing for the user’s sake. Drawing on the philosophy of Immanuel Kant, some argue that being unkind to anything even a machine can make a person harsher in their daily life.
As BBC senior technology journalist Thomas Germain observes, AI tools are mimics, not living beings. Whether you are polite or blunt, the goal is to get what you want faster and more efficiently, saving both time and energy in the process.
