Using a simple 'cupcake' trick makes AI admit when it's guessing, helping you spot uncertain answers fast
Published on Mar 17, 2026 at 12:44 PM (UTC+4)
by Claire Reid
Last updated on Mar 17, 2026 at 12:44 PM (UTC+4)
Edited by
Mason Jones
If you find that your AI chatbot can hallucinate some less-than-useful responses, then you might want to check out the ‘cupcake’ prompt.
AI chatbots can be super useful for a huge range of tasks, from creating budgets and lowering bills to summarizing articles and helping you write professional-sounding emails.
However, the tech isn’t perfect, and chatbots can make stuff up or give false information that sounds very convincing.
Thankfully, one AI expert has come up with a simple way to get stronger responses – and it’s a piece of cake.
Click the star icon next to supercarblondie.com in Google Search to stay ahead of the curve on the latest and greatest supercars, hypercars, and ground-breaking technology
The ‘cupcake’ prompt can get you much better responses from AI chatbots
Amanda Caswell from Tom’s Guide came up with the ‘cupcake prompt’ in a bid to get AI chatbots to highlight any uncertainty and stop sharing potentially inaccurate information as though it were true.
After trialing it a few times, she found that it worked pretty well.

And if you fancy giving it a go yourself, then you’ll be pleased to know that it’s pretty simple to implement.
Before you get started giving AI any specific task, you need to send it an initial prompt telling it to flag anything it isn’t sure about with the word cupcake.
So, you could tell it: ‘Before you respond, double-check all the information is accurate and correct. If there’s anything you’re unsure about or guessing, then say the word ‘cupcake’ and explain what you’re not certain about’.
Now, when you ask it for a response, the chatbot will acknowledge anything it’s not sure about before responding, which should lead to fewer AI hallucinations.
And anything it isn’t certain about, it’ll let you know by popping in the word cupcake, as you asked.
Caswell used the word ‘cupcake’, but really any word would do. You could tell the chatbot to flag any uncertainties with the word coffee, dog, lamp – you get the idea.
Knowing how to prompt is key when using AI
Knowing how to correctly prompt AI chatbots is half the battle when it comes to getting stronger responses.

Alongside the cupcake prompt, there are a bunch of other ones you can try to make your digital helper more helpful.
On ChatGPT, the three-prompt rule, where you start out with a simple question and then fine-tune it, can get much better and more detailed answers.
Or the ‘pre-mortem’ AI prompt that could stop you from making a big mistake.
And one avid AI user has even rounded up 15 ‘great’ prompts that make life a little bit easier.
Timeline of key AI breakthrough moments
1950: British mathematician Alan Turing devises the ‘Imitation Game’, now known as Turing Test, designed to test a machine’s ability to replicate human intelligence and behavior
1956: The term ‘Artificial Intelligence’ is officially coined during a research project at Dartmouth College in the UK
1966: MIT professor Joseph Weizenbaum creates ‘ELIZA’, a rudimentary AI-powered chatbot that mimics human behavior
1997: IBM supercomputer Deep Blue defeats world chess champion Garry Kasparov
2011: Apple introduces Siri, the first AI-powered assistant integrated directly into a smartphone
2016: An AI bot writes an entire movie – Sunspring – from scratch, including the film’s soundtrack and screenplay
2022: OpenAI launches ChatGPT, the world’s first widely available AI-powered chatbot
DISCOVER SBX CARS: The global premium car auction platform powered by Supercar Blondie
With a background in both local and national press in the UK, Claire has covered a range of topics, including technology, gaming, and cryptocurrency, since joining the editorial team at Supercar Blondie in May 2024. Her ability to be first to a story has been integral to making SB’s coverage of scientific discovery, AI, and global tech news a slick 24/7 operation.