Why won't AI just say “I don't know”-

Ask ChatGPT, Claude, or Gemini almost anything and you'll get a confident answer, often opening with "Absolutely!" or "Great question!" What you almost never hear is "I don't know." Think of a waiter who refuses to say "we're out of that" and just brings you something else with a smile.

Four reasons AI guesses instead of admitting uncertainty

1. Next-word prediction. AI writes by picking the likeliest next word, and on the open internet "I don't know" is rarely the likeliest choice.
2. Tests reward guessing. OpenAI research in 2025 argued that many benchmarks score like a quiz where blanks lose and lucky guesses win, so models learn that guessing pays.
3. Human raters prefer confidence. During training, a sure-sounding answer usually beats "I'm not sure."
4. AI doesn't know what it doesn't know. Sounding sure and being right are not the same thing.

When AI can say "I don't know"

It can, when it knows what it knows. Grounding answers in trusted sources lets a system tell when a question falls outside its library. Ask a health companion who the president of Uganda is and a well-built one will stay in its lane.

The trickier case is on-topic but missing pieces. A patient asking an insulin dose should get a handoff to the care team, not a confident guess. A marketer asking a competitor's price should hear "I couldn't find current pricing" instead of invented numbers.

Three questions to ask before you choose an AI product

• What does it do when it doesn't know?
• Does it show its sources?
• Can it hand off to a person?

Bonus: a Stanford-led study in Science in 2026 found that across eleven models, chatbots backed the user's actions 49% more often than people did. When AI takes your side, get a second opinion from a human. For more on grounding and trusted sources, see our Responsible AI-by-Design Framework or Can AI stay grounded in facts?.