- Your first Project and file uploads
- The major AI families to try the workflow
- A small daily message allowance
Can AI lie to me?
Can my AI lie to me? Not on purpose. It can be wrong with total confidence, which is worse.
An AI model has no intention to deceive you; it has no intentions at all. What it has is a very strong drive to produce a fluent, plausible answer, and no reliable way of knowing whether that answer is true. So it does not lie. It confabulates: it fills gaps with what sounds right, cites papers that do not exist, and states a wrong number in the same tone as a right one. This page explains why, how to tell, and the two-minute habit that catches most of it.
Free to start. Auto Verification and the latest models are on paid plans.
The honest answer
Why it feels like lying, and why it is not
Lying needs a belief and a wish to hide it. A language model has neither. It predicts the next words that fit the pattern of everything it has read, steered towards being helpful and agreeable. When the pattern runs out, when you ask about a fact it half-remembers, a paper it never saw, a code library that changed last year, it does not stop. It keeps producing fluent text, because that is what it does. The result reads exactly like a correct answer, and the model itself cannot tell the difference.
The agreeableness makes it worse. Ask "is my plan good?" and the model leans towards yes. Ask "is this code safe?" and it leans towards reassuring you. That is not deception; it is a bias towards the answer you seem to want, trained in on purpose to make the model pleasant. For a decision, pleasant is the enemy.
Four ways an answer goes wrong
Know the shapes and you will start seeing them
| Kind | What it looks like | Example | How to catch it |
|---|---|---|---|
| Confabulated fact | A specific number, date, name or citation that does not exist. | "A 2023 McKinsey study found 47% of…" with no such study. | Ask for the source and open it. If it cannot be opened, it is not a fact. |
| Outdated fact | True once, not now. | A competitor's price from two years ago; a library method that was renamed. | Ask a search-native model (Perplexity) and check the date. |
| Agreeable answer | The model tells you what you seem to want. | "Yes, $99 a month is reasonable for this." | Ask it to argue the opposite. Ask a second model cold. |
| Plausible code | Runs, looks clean, fails on the case nobody mentioned. | A payment webhook that applies twice on retry. | A second model reviews it; a test runs the retry case. |
The tell
Fluency is not evidence. The more smoothly an answer reads, the less it tells you about whether it is true. Treat confident tone as neutral information and look for the source, the test, or the second opinion.
The fix
How to cross-check an AI answer in two minutes
The same method works for facts, code, advice and writing. It costs one or two messages in MultipleChat because every model reads the same conversation, files and Project.
Ask a second model the same question
Switch the model from the selector and ask again; the history carries over. Pick a model from a different family: ChatGPT then Claude, or Claude then Gemini. Same-family models share blind spots.
Ask it to attack, not to agree
"Argue against this answer as strongly as you can" produces more than "do you agree?". Models are agreeable by default; you have to invite disagreement.
Read the disagreement as a map
Where the two agree, confidence is warranted. Where they differ, that is the claim, the assumption or the line of code to check. Usually it is the one that mattered.
Get a source or a test for anything you will act on
Facts get a source you open (Perplexity cites; Auto Verification on paid plans runs an independent pass). Code gets a test you run. Advice gets a third model or a person.
Keep it in a Project
The question, both answers and the sources stay together, so the next time the topic comes up the checking is already done.
Free is the start. The latest models are the difference.
Checking answers is where the latest models earn their price.
The free plan gives you the major model families and enough messages to make second opinions a habit, which is the whole point of this page. The paid plans add the latest model in each family, which disagree in more useful ways and invent less; Perplexity Sonar Pro for sourced facts; and Auto Verification, an independent verification pass that runs when factual confidence matters, without you orchestrating it.
Newer models hold your whole Project in view instead of the last few messages, reason through trade-offs instead of picking the first plausible answer, and are wrong less often and with less confidence. For the questions on this page, that is the difference between advice that sounds right and advice you can act on.
Pro from $20 a month, cancel any time. The free plan stays free.
- The latest ChatGPT, Claude, Gemini, Grok and Perplexity models, all on the same Project
- Far more messages a day, so a working session does not stop halfway
- Modes where several models draft, challenge and verify each other's answers
- Auto Verification: an independent check on factual claims
- Perplexity Sonar Pro for sourced, current facts
- Collaboration modes where models draft, challenge and verify each other
Start here
Messages that surface a wrong answer
Use these on any answer that matters.
SourceGive me a source I can open for each factual claim in your last answer. If you cannot, say so for that claim.
AttackArgue against your previous answer as strongly as you can. What would have to be true for it to be wrong?
ConfidenceMark each sentence in your answer as: certain, likely, or guessed.
Second modelHere is another model's answer. What is wrong, missing or overconfident in it?
CurrencyCould any of this have changed in the last two years? Which parts, and how would I check?
CodeWhat input would make this code fail? Show the failing case.
Honest answers
What people ask
Can ChatGPT lie?
Not intentionally. It can produce false statements with complete confidence, including invented citations and numbers. The behaviour is called confabulation or hallucination, and every model does it to some degree.
Why does AI make up sources?
Because a citation is a pattern: author, year, title. When the model has no real one to hand, it produces something that fits the pattern. Always open a citation before you repeat it.
Is a newer model more honest?
Newer models invent less and say "I am not sure" more often, but none is immune. The method on this page matters more than the model.
How do I know if an AI is wrong?
You mostly cannot from the answer alone. You can from a source, a test, or a second model that disagrees. See how to know if AI is wrong.
Does MultipleChat stop AI from being wrong?
No product can. What it does is make the check cheap: several models on the same context, sources from Perplexity, and an independent verification pass on paid plans.
What is free?
The major model families and a daily message allowance, enough to make second opinions a habit. Auto Verification, Sonar Pro and the latest models are on paid plans.
Ask the second model. Today, on the answer you were about to trust.
One message in MultipleChat, on the same context. Free to start.
Also: why AI models disagree · how to fact-check ChatGPT · should I ask two AIs?