Despite mounting pressure from both Google and Anthropic, OpenAI’s GPT-5.1 remains one of the best AI models in the world. And, in a recent episode of the OpenAI podcast, the team behind it shared some tips on how best to use it.
Christina Kim, a research lead in post-training at ChatGPT, and Laurentia Romaniuk, a product manager for model behavior, were guests on the latest episode of the podcast, published on Wednesday.
1. Memory tricks
This is a feature that came to ChatGPT fairly recently. If enabled, ChatGPT will hold memories about you from the conversations you’ve had in the past.
This can be anything from remembering your birthday, a preference for a certain type of genre of film, or that you’re in the market for a new TV. The idea here is that it will allow ChatGPT to know you better, filtering its responses in a way that will appeal to you specifically.
“The model will write down things it knows about you from its conversations with you to refer back to later. This is really nice because you’re not just repeating yourself every time,” said Romaniuk.
“It also helps it to think through its answers for when it responds to you. It has that context and that grounds its answer in being the most useful possible.”
While memory can be turned off, the team highlights just how intuitive ChatGPT can be with it on, personalizing the entire experience to you.
2. Customize the chatbot
In the podcast, Romaniuk shares tips on how to get ChatGPT to perform better for your specific needs.
This is based on a conversation that she had with her brother, who complained that ChatGPT Pro was responding at an undergraduate level. He needed the model to have a better understanding, but hadn’t primed it for that.
Instead, she told him to specify the level of expertise he needed. After telling the model to be a “frontier researcher,” the model produced an insight so advanced that it mirrored a breakthrough his lab had recently had.
This is advice that has been passed around a lot in recent years for using AI chatbots. Tell the model a level of expertise that they should adopt, and it will perform better.
This can be done by starting a ChatGPT prompt with a command. For example, “You’re an expert biologist” or “You’re a sub-editor for a tech magazine”. These starting points help the model understand the level of expertise it should approach the task with.
3. Challenge ChatGPT
AI chatbots are constantly changing and evolving, so it can often be hard to know what they can or can’t do. One piece of advice from the podcast episode is to pressure test the model with really challenging questions.
“Have your super hard questions, things you know really well. I love to pressure test the model to see how it's changing and improving. We’re shipping updates all the time and so it is so easy to say I heard it's great for coding, but it didn’t work. That might be true today, but in a couple of months it could be a completely different story," said Kim.
If there is a part of ChatGPT that you tried before and didn’t get on with, try again in the future and see if it is now something that ChatGPT can handle.
As Kim says, you should also try pushing the model in an area you know well. This can give you a sense of its strengths and weaknesses.
4. Ask for help
A very simple bit of advice they offer, and one that I’ve found success with too, is asking ChatGPT for help with a prompt.
This is something that, in the podcast, they highlighted as a former weakness of ChatGPT. The model would take a guess at how to write a good prompt for itself but it often wasn't sufficient. However, most chatbots now have a good understanding of how to write prompts for themselves.
Try asking ChatGPT, “come up with a prompt to [insert the task you’re trying to achieve]”. This can create something that will push the model more efficiently in the right direction.
5. Try new personalities
One of the big changes that came from GPT-5.1 was the inclusion of new personalities. ChatGPT can be made more fun, friendly, more serious and more.
In the podcast episode, the team talks about the fun they've had experimenting with these different modes.
While they don't neccessarily make the model smarter or more efficient, it can make it more interesting to chat to.
When I've tested ChatGPT's different personalities, I was impressed with just how adaptable it could become to certain situations. Try messing around with these settings yourself.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.
More from Tom's Guide
- I created one Gemini 3 prompt that works in absolutely any situation — here’s how to use it
- Elon Musk’s Grok 4.1 vs Anthropic’s Claude 4.5 Sonnet — here’s the AI model that’s actually smarter
- I tested NotebookLM vs. Perplexity for deep research with 5 difficult prompts — here's the clear winner
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
