AI Confessions: Got AI to Admit When It’s Guessing

Ever feel like AI is too smart for its own good, responding with made-up answers so confidently?

We put it to the test working on an article for travel managers. We wanted fresh and compelling ideas, so we turned to AI. But we didn’t want to blindly trust its suggestions.

So, we came up with a fun way to keep AI (and us) in check. See the key parts of the conversation in the carousel below:

• We shared our strategy doc with AI, including target persona, pain points, value props, messaging, positioning, and brand voice.

• We asked AI to suggest article topics.

• Here’s the key: We had AI rate its own confidence in each idea (High, Medium, Low). For anything less than “High”, we asked what info it needed to be more sure.

Check out the screenshots to see how key parts of the conversation played out. It’s like having AI do a self-check!

This simple approach helps us:

• Spot where AI might be guessing

• Know what to double-check

• Give AI better info to work with

We can certainly take it further:

• Add number ratings (1-10) for more detail

• Ask AI to cite sources for high-confidence answers

• Get AI to point out its own potential biases

It’s not about not trusting AI. It’s about working smarter together.

What do you think? Have you tried something similar? Let’s chat in the comments.

#Hallucinations #ResponsibleAI #AIBias #HumanOversight #ContentCreation GrowthPath Partners