Tips and Tricks to Reduce Hallucination

👤 Authored by Lashmi Bai Ravindrapandian
📅 Published on Mar 14, 2026

Wrote a post recently on what AI hallucination actually is, and how it forms. Once I understood that, next question was simple…… okay, so what do I actually do about it.

Made myself a rough checklist, honestly, after going through all this, collated from various knowledge sources I read and picked up along the way. Nothing fancy.

Breaking it into smaller pieces here, in plain words, my future bookmark.

1. Be clear about what you want

• Say clearly what you want, don’t leave any room for it guessing. Tighten prompts and system instructions.
• Ask one thing at a time, don’t stack too many asks in one prompt or use numbered list if breaking it out.
• If it’s a fact-based question, tell it “say you’re not sure, don’t guess”.
• Think yourself like a celebrity and always run the “Ask Me Anything” session with AI so there is no room for confusion.

2. Watch how you phrase the question

• Don’t ask leading questions, if your question hints at the answer you want, AI often just agrees with you
• Before you question AI, ensure Role, Context, Assumptions, Format, Tone, Do and Don’ts are given clearly – All these help to reduce AI Hallucination.
• Show it one good example first. It will help to set the context visually.

3. Use the right settings

• If a setting called temperature is there, keep it low for factual stuff, higher only for creative stuff.
• Mention if you want the output to be deterministic or can involve some level of creativity upfront.
• Ask it to show its steps, not just the final answer, easier to spot where it went wrong

4. Give it real information to work with

• Give it the real document, don’t expect it to “just know” your stuff
• If it’s pulling from stored info, keep that info updated, old info gives wrong answers
• Ask it to show where in the source it got the answer from, keeps it honest.

5. In long conversations

Here’s the time to use your project management skills.

• If the conversation’s gone on a while, ask AI to summarize where things stand before continuing, catches anything that quietly got lost
• Break a big task into shorter separate conversations instead of one giant one, easier to keep everything accurate throughout.

6. Verify before you trust the answer

• Don’t trust numbers, dates, or quotes until you’ve checked them yourself
• Be extra careful with anything recent, that’s a weak spot
• Ask the same question again, in different words, see if the answer stays the same

7. Watch for these red flags

• If the answer sounds too smooth and too perfect, that’s your cue to check again
• Don’t trust it more just because the answer is long or detailed, length isn’t proof

8. Common mistake spots

• Double check any maths, even simple ones, this is a common mistake spot
• Watch for mixed-up names, two similar things can get blended into one wrong answer

9. While building AI agent

• Add a person to check before anything big happens, approvals, payments, that kind of thing
• Keep a record of what it did and why, helps a lot when something goes wrong later

Why this need to be a habit, not extra work

None of this is hard, honestly. Just small checks, done regularly.

Felt like extra work at first, not going to lie. After getting it wrong a few times – trusting something that sounded right but wasn’t – I stopped treating these checks as optional.

Now this checklist is just part of how I use AI, not something extra on top.