Club Benefits
If you've spent any time using ChatGPT, Gemini or Claude, you've probably run into one of AI's most frustrating problems: hallucinations.
That's the industry term for when a chatbot confidently presents information that's wrong, fabricated, outdated or simply doesn't make sense. While today's AI models are far better than they were a year ago, hallucinations still happen — and as someone who uses AI every day for work, I've seen plenty of them firsthand.
After one too many confidently incorrect answers, I started experimenting with a simple trick: adding a reusable prompt before asking my question. The results weren't perfect, but they dramatically improved the quality of the responses I received and helped reduce the number of AI mistakes I encountered.
You should always verify important information yourself, especially when it comes to health, finance or legal advice. But these simple prompts have become part of my daily workflow, and they can help you get more accurate, reliable answers from any chatbot.
Here are the prompts I use most often.
Prevent hallucinations before they even happen
Every time I enter a new conversation with AI that requires some form of research and fully informed responses, I make sure to make my request as easy to understand as possible. Alongside that task is one of the following prompts, which are meant to make chatbots produce their most factual answers and ensure that what they’re presenting is backed up by legit sources:
- The all-encompassing prompt for maximum reliability: Answer using only verified information. If any of that information is missing or uncertain, say so clearly. Do not guess or fabricate details. Cite the evidence or reasoning used to find that information. Ask clarifying questions if needed. And after answering, review your response for possible inaccuracies.
- The structured truth prompt: Use this structured format when presenting me with your answers - known facts you’ve found, any assumptions and unverified claims you’ve made, and any missing information you’ve been unable to find.
- The self-check prompt: After generating your answer, be sure to critique it and identify any possible inaccuracies, assumptions, or hallucinations.
- The 'no-fake citations' prompt: When presenting your answers, never invent sources, links, quotes, studies, statistics, or citations. If you cannot verify one, say so explicitly.
- The 'ask questions first' prompt: If the question or request is ambiguous or missing key details, ask follow-up questions before answering.
Since adding these prompts to my workflow, I've encountered far fewer instances of chatbots serving up obviously incorrect, fabricated or misleading information. Having ChatGPT, Claude, Gemini and Perplexity double-check their reasoning, flag potential inconsistencies and present answers in a clear, structured format has made my daily interactions with AI feel much more reliable.
They're not foolproof, and I still verify important information whenever accuracy matters. But these prompts have helped me get better answers more consistently — and that's a win in my book.
Bottom line
AI tools are constantly improving their knowledge and reasoning capabilities, but that doesn't mean users should blindly trust every answer they produce. I've found that using prompts designed to reduce hallucinations can help improve accuracy, but they're only part of the equation.
No matter how advanced ChatGPT, Claude, Gemini or Perplexity become, it's still important to apply basic critical thinking and verify important information through trusted sources. The best results come from treating AI as a helpful assistant, not the final authority.
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. Subscribe to Tom's Guide on YouTube and follow us on TikTok.
More from Tom’s Guide
- Claude Opus 4.8 just launched — and Anthropic says it's far less likely to ‘fake’ answers
- Google’s AI can’t spell its own name — and that’s a terrifying sign for the future of search
- I used the ChatGPT ‘circus’ prompt — and it's a surprisingly effective for multitasking
