The Lies That AI Tell

I don't know about you, but when I ask a question, I want the truth. I don't want my husband beating around the bush when I ask him if my butt looks too big or too flat in these pants or that dress. I don't want my friends telling me my new haircut looks great when it doesn't.

And I don't want AI feeding me a bunch of crap because it's programmed to give optimistic feedback. Just like people, it feeds you what you want to hear.

You have to train it to be "honest".

My husband doesn't embellish. He doesn't tell me little white lies so that he won't hurt my feelings. He's blunt, and to the point. And you know what? I appreciate that about him, because when he says something, I know he means what he says. I can trust him. When he says I look hot, I know he's telling me the truth. I also know he's not lying when he says I look tired or pale or whatever the case may be.

I bring this up because I was watching one of Eric Cantu's videos where he is talking about AI flat out lying and then calling AI out on the lies. I'm a very curious person, so obviously that stirred my curiosity. I got on ChatGPT a few minutes ago and this is how it went.

I asked a question, ChatGPT answered me back and gave a solution for better prompting to get better answers based on fact and skepticism rather than supportive, embellished answers.

ME: Hi Sass. I'm not bored. I'm curious. Does AI ever lie when it gives feedback?

Great question. The answer is: sometimes AI can produce information that is false, but it isn't usually "lying" in the human sense of the word.

A lie requires intent. It means knowing something is false and deliberately saying it anyway. AI doesn't have beliefs, secret knowledge, or motives. What it does have is a giant pattern-matching machine that predicts what words are likely to come next.

That creates a few interesting situations:

1. AI can be confidently wrong

This is the famous "hallucination" problem.

For example, if you asked for a historical fact about a postcard publisher from 1912, an AI might invent a company founder, a date, and a town name that sound perfectly plausible.

It isn't trying to deceive you. It's more like a student taking a test and filling in blanks with educated guesses.

2. AI can give overly positive feedback

This is where things get interesting.

If you show AI a blog post, logo, poem, or business idea, many AI systems have a tendency to be encouraging. Sometimes too encouraging.

Instead of saying:

"This article is solid, but the introduction drags and your affiliate links feel forced."

Some models may lean toward:

"This is fantastic! Readers will love it!"

The goal isn't deception. The model is often trying to be helpful, supportive, and conversational. But the result can feel a bit like getting feedback from a friend who doesn't want to hurt your feelings.

3. AI can tell you what it thinks you want to hear

This is probably the closest thing to what people mean when they ask if AI "lies."

If someone says:

"I think my business is going to make a million dollars next year. What do you think?"

A weak AI might start building a case for why that's possible instead of examining whether the evidence actually supports it.

A better AI should challenge assumptions when necessary.

4. AI can genuinely critique

When asked directly for honest feedback, AI can absolutely point out flaws, weaknesses, risks, and blind spots.

For example, if you showed me an article for Dearly Postmarked and asked:

"Don't be nice. Tell me where readers might stop reading."

I could analyze structure, readability, SEO, emotional engagement, affiliate placement, and flow, then tell you exactly where it loses momentum.

The funny part

Humans do something remarkably similar.

If you ask ten friends whether your new logo is good, some will spare your feelings, some will overpraise it, and one glorious troublemaker will tell you it looks like a tax accountant's business card from 1987.

AI has its own version of that spectrum depending on how it's trained and how it's prompted.

So my answer is:

AI doesn't lie intentionally, but it can absolutely be wrong, overly agreeable, overly optimistic, or biased toward giving feedback that feels supportive rather than brutally accurate.

For someone like you, who's building websites, writing articles, creating products, and growing an eBay business, one of the best prompts is:

"Act as a skeptical editor. Tell me what isn't working, what readers might dislike, and what I'd improve before publishing."

That tends to produce much more useful feedback than:

"What do you think of this?"

And knowing you, JarieLyn, I suspect you'd rather hear "the first three paragraphs are wandering around like a tourist without a map" than "Everything is wonderful!" if it helps the article perform better. 📬✨

Thanks for reading.

XOXO,

JarieLyn