Skip to main content
  1. Home
  2. Computing
  3. News

Try this two-step trick for better chatbot prompts, repeat and empathize

One paper says repeating a prompt helps for non-reasoning tasks, another shows you get better responses when you share constraints like your skill level and what kind of help you need.

Add as a preferred source on Google
Chatgpt
Unsplash

If your AI assistant keeps dropping specifics or talking past the question, you don’t need a new model to get better results. Small changes to better chatbot prompts can clean up a lot of the mess.

One fix is mechanical. A Google Research paper highlighted by VentureBeat points to a dead simple move, paste your exact request twice in the same message. It’s meant to raise accuracy on straightforward work like extraction, short answers, and basic rewriting, not long, multi-step reasoning.

Recommended Videos

The other fix is human. Separate research from Northeastern University suggests people get stronger outputs when they prompt with perspective taking, sharing what they know, what they don’t, and what kind of help would actually be useful.

Duplicate the request, see gains

The duplication trick is exactly what it sounds like. Paste the same prompt back to back, then send.

In the reported testing, doubling up the input helped on non-reasoning benchmarks across major models. The real-world take is simple, repeating the instruction can make it harder for the model to miss, especially when you want it to stick to specifics instead of sliding into a generic response.

A little perspective changes everything

This isn’t about being nicer, it’s about being clearer. Give the bot the calibration knobs it can’t reliably infer, your level, your time, and the format you want.

The study frames this as theory of mind style prompting, asking in a way that anticipates how another agent will interpret your request. A short line like “I’m new to this, skip jargon” or “Give me three options, then pick one” can reshape the reply fast. Additionally, check the best ways you can use one of the most popular AI now.

Put both moves into practice

Use the double-paste move when the job is straightforward, like “summarize this paragraph,” “pull the key dates,” or “rewrite this email shorter.” Copy it twice, then stop.

After that, add one or two lines that define success. Include your constraints, your background, and where you’re stuck. If the assistant still glosses over key points, keep the duplicate prompt in place. If it keeps giving the wrong kind of output, adjust the perspective-taking line until it matches your goal. Or if you’re experiencing prompt fatigue, we already have experts sharing their prompts.

Paulo Vargas
Paulo Vargas is an English major turned reporter turned technical writer, with a career that has always circled back to…
I spent a fortune on a Copilot+ PC, and I’ve barely ever touched Microsoft’s AI
Microsoft needs to give Copilot+ PC owners a reason to use Copilot
Copilot

There is a dedicated Copilot key on my ASUS Zenbook 14 OLED. Months after buying the laptop, it may be one of the least important keys on the entire keyboard. My Zenbook UM3406 runs on AMD’s Ryzen AI 300 series processor, complete with a dedicated NPU offering up to 50 TOPS of AI performance. That qualifies it as a Copilot+ PC, which makes it a part of what Microsoft once described as the new era for Windows.

AI is already a regular part of my workday. I use it for research, brainstorming, and working through ideas. But rather than relying on something built into the Windows OS, I've relied on the likes of ChatGPT, Claude, and Gemini.

Read more
I bought into a brighter future, but my digital life bills monthly subscriptions
My day runs on streaming, cloud storage, AI, transport, and smart devices. None of it feels expensive until I realize almost nothing is ever fully paid for
Light, Adult, Female

Before I’ve properly started the day, several companies have already clocked in. Spotify or YouTube Premium handles the music, Google One keeps my files available, and an AI subscription is waiting in another tab to help me work faster.

Most of these charges earn their place. They save time, remove friction, and keep the day moving. I barely notice them until I think about what would stop working if one payment didn’t go through.

Read more
What happens when AI detectors fail? Researchers say we must be trained to spot fake AI faces
Researchers say spotting AI faces may soon depend more on people than software
Zuckerberg Deepfake

Artificial intelligence has become remarkably good at creating fake human faces. So good, in fact, that the old tricks people relied on - counting fingers, spotting warped earrings, or looking for distorted backgrounds - are quickly becoming obsolete. According to a new study highlighted by the BBC, the next line of defence may not be a better AI detector at all. It might simply be a better-trained human.

Researchers from the University of Aberdeen, working alongside Australia's National University, found that people can dramatically improve their ability to distinguish AI-generated faces from real ones after a relatively short period of structured training. Instead of hunting for obvious visual glitches, participants were taught to recognise subtle patterns that modern image generators still struggle to replicate consistently.

Read more