Asking questions of various chatbots, they tried to call them “smart”, encouraged them to think carefully and even ended the questions with comments like: “This is going to be fun!”
None of this had a significant impact, but there was one technique that stood out. When they made artificial intelligence pretend saga Star Trekadvanced in basic mathematics.
People are using all sorts of weird strategies to get better answers from Large Language Models (LLM), the AI technology behind tools like ChatGPT.
Some swear that AI works better when you threaten it, others believe that chatbots work better when you’re polite, and there are those who ask bots to pretend to be experts on the subject they’re discussing.
The list goes on. This is part of the mythology around “signal engineering” or “context engineering,” different ways of constructing instructions to make AI perform better.
Here’s the thing: Experts tell me that many of the conventional wisdom just don’t work. In some cases, they can even be dangerous. But how you talk to the AI matters, and some techniques do matter.
“Many people believe that there is a magic set of words that allows masters to solve a problem,” says Jules White, a computer science professor who studies generative artificial intelligence at Vanderbilt University in the US. “But it’s not about the choice of words, it’s about how you fundamentally express what you’re trying to do.”
Pay attention to manners?
Last year, the social media user
Sam Altman, CEO of OpenAI, the company behind ChatGPT, responded: “Tens of millions of dollars well spent,” he said. – You never know.
Most people interpret the last line as a blatant reference to the idea of a possible AI apocalypse, although it’s hard to see how seriously the “tens of millions of dollars” figure should be taken. But politeness is also a practical matter.
LLM chatbots work by breaking your words into small chunks called “tokens” before analyzing them using statistics to produce a relevant response.
This means that everything you say, from word choice to an extra comma, will affect the AI’s response. The problem is that it is extremely difficult to predict.
Much research has been done looking for patterns in small changes in AI signals, but much of the evidence is contradictory and inconclusive.
For example, a 2024 study found that LLM chatbots provided better and more accurate responses when asked politely rather than simply given orders.
More surprisingly, there were cultural differences. Compared to Chinese and English, Japanese chatbots have fallen a little short in being overly polite.
But overall, there hasn’t been enough research on the topic to draw firm conclusions. Additionally, AI companies are constantly updating their chatbots, which means research can become outdated.
Experts say AI models have improved dramatically in just a few years, making techniques like flattery, politeness, insults or threats a waste of time when the goal is to make AI more accurate.
According to experts, the latest models of artificial intelligence, which can be found in any common product, such as ChatGPT, Gemini or Claude, better capture the most important parts of the instructions.
These small language changes probably won’t affect them, at least not in a consistent, usable way.
This is disturbing in its own way. Companies often design artificial intelligence to behave like humans, so it’s understandable that sometimes they seem to have a mood or personality that can be manipulated.
But don’t be fooled. AI tools are simulators, not living beings. They simply imitate human behavior.
If you want better answers, stop treating AI like a person and start treating it like a tool.
How to talk to your chatbot
AI poses very real problems, from ethical issues to environmental impacts. Some people refuse to interact with him at all.
But if you’re going to use an LLM, learning how to get what you want faster and more efficiently will be better for you and possibly the energy you use up in the process.
These six tips will help you get started.
1. Ask for multiple options
“The first thing I tell people is don’t just ask for one answer, ask for three or five,” White says.
If you need help with text, for example, tell the AI to give you several options that vary widely. “It makes people think about what they like and why,” he explains.
2. Give examples
If possible, provide an AI sample. “For example, I see people who ask a master’s degree to write a letter and then get frustrated because they think, ‘This doesn’t sound like me at all,'” says White.
The natural impulse is to respond with a list of instructions such as “do this” and “don’t do that.” White says it’s much more effective to say, “Here are 10 emails I’ve sent before. Use my writing style.”
3. Request an interview
“Let’s say you want to create a job description. Tell the AI, “I want you to ask me questions one by one until you gather enough information to write a compelling job offer,” White explains.
“By doing it question by question, it can be tailored to your answers.”
4. Be careful with role-playing games
“It used to be that if you told an AI that you were a math teacher, for example, it would be more accurate in answering math questions,” says Sander Schulhoff, an entrepreneur and researcher who helped popularize the idea of rapid engineering.
However, when searching for information or asking questions with only one correct answer, Schulhoff and others say role-playing games can reduce the accuracy of AI models.
“It can really be dangerous,” says Rick Battle, an applied machine learning engineer at Broadcom and co-author of the Star Trek study.
“You’re actually encouraging the hallucination because you’re telling him that he’s the expert and that he should trust his inner parametric knowledge.”
Essentially, this can make the AI act overconfident.
But for very open-ended tasks without a single answer, role-playing is effective. Brainstorm tips, brainstorming, and creative or investigative problem solving.
Or, if you’re nervous about the interview, having a chatbot impersonate the hiring manager can be good practice.
5. Maintain neutrality
“Don’t be a witness,” says Beatle. If you’re trying to decide between two cars, don’t say you’re leaning toward the Toyota. “Otherwise, that’s the answer you’re likely to get.”
6. Please and thank you
According to a 2019 Pew Research Center survey, more than half of Americans say “please” when talking to their smart assistants.
This trend seems to be continuing. A survey conducted in 2025 by the publisher Future found that 70% of people are polite to artificial intelligence when using it. Most said they did it because it was the right thing to do, although 12% said it was to protect themselves in the event of a robot uprising.
Being polite may not protect you from angry robots or make LLMs more accurate, but there are other reasons to keep doing it.
“The most important thing for me is that the words ‘please’ and ‘thank you’ can make you more comfortable interacting with AI,” says Schulhoff. “It doesn’t improve the performance of the model, but if it helps you use it more because you feel more comfortable, then it’s useful.”
You must also consider the sensitivity of your own human nature. Philosopher Immanuel Kant argued that one of the reasons one should not be cruel to animals is that it also harms oneself.
Basically, being hostile towards anything makes you a tougher person. You can’t hurt an AI’s feelings because it doesn’t have any, but maybe you should still be nice. It’s a habit that can benefit other aspects of your life.
*You can read the original English version of this BBC Future article here.
click here to read more stories from BBC News Mundo.
Subscribe here to our new newsletter to receive a roundup of our best content of the week every Friday.
You can also follow us YouTube, Instagram, TikTok, X, Facebook and in ours WhatsApp channel.
And remember you can get notifications in our app. Download the latest version and activate them.
FUENTE: BBC
