I Don’t Trust AI. Should I?

If someone tells me, “I don’t trust artificial intelligence,” my first reaction isn’t to convince them they’re wrong.

My reaction is:

Good.

You shouldn’t automatically trust AI.

You shouldn’t automatically trust Google.

You shouldn’t automatically trust something you read on Facebook.

You shouldn’t automatically trust a YouTube video.

You shouldn’t automatically trust a salesperson.

And you definitely shouldn’t automatically trust some confidently written paragraph simply because a computer produced it in three seconds.

Skepticism isn’t the enemy of artificial intelligence.

Used properly, skepticism might be one of the most important AI skills you can have.

The goal isn’t to learn how to trust AI.

The goal is to learn how to use AI intelligently.

Those are two very different things.

Why AI Feels Different

We’ve used computers for decades.

Traditionally, though, computers felt like machines.

You clicked buttons.

You typed commands.

You searched databases.

You filled out forms.

Generative AI feels different because you talk to it.

You might write:

“I’m considering changing jobs. Help me think this through.”

And the AI responds in natural language.

It might ask questions.

It might remember context within the conversation.

It might organize your thoughts.

It might even sound sympathetic.

That conversational interface creates an interesting psychological problem.

Something that talks like a person can begin to feel like a person.

It isn’t.

No matter how natural the conversation feels, you are interacting with software.

That distinction matters.

AI Can Be Wrong and Still Sound Excellent

This is probably the first lesson everyone should learn about generative AI.

AI can produce information that is incorrect.

Worse, the incorrect information can sound extremely convincing.

The National Institute of Standards and Technology uses the term confabulation for situations where generative AI confidently produces erroneous or false content.

You’ve probably heard the more common term:

hallucination.

NIST specifically warns that generative AI can produce incorrect statements, inconsistent information, and even false citations or reasoning that appears to support an incorrect answer. (NIST Publications)

That’s an important limitation.

Traditional computer errors often looked like errors.

You’d get an error message.

Something wouldn’t load.

The program crashed.

Generative AI errors can arrive in a beautifully formatted paragraph with bullet points and perfect grammar.

That’s more dangerous because polish can create the illusion of accuracy.

Confidence Isn’t Evidence

Imagine asking an AI:

“Who invented this particular device and in what year?”

It gives you a name.

A date.

Maybe even a quote.

That feels authoritative.

But what happens if it invented the quote?

Or mixed up two people?

Or found outdated information?

The correct reaction isn’t:

“The AI said it, so it must be true.”

The correct reaction is:

“Interesting. Now let’s verify that.”

That’s especially important when a fact actually matters.

The Risk Depends on What You’re Doing

Not every AI mistake has the same consequences.

Suppose you’re using AI to brainstorm names for your fantasy football team.

The AI gets something wrong.

Society continues.

Suppose you’re asking AI about whether two medications interact.

Different situation.

Suppose you’re asking AI to interpret a contract.

Different situation.

Suppose you’re making an investment decision based on an AI-generated claim about a company.

Again, different situation.

The biggest mistake people can make is treating all AI answers as though they carry the same level of risk.

They don’t.

Think in Three Risk Levels

Here’s a simple approach.

Low-Risk AI Use

These are tasks where a mistake is annoying but probably doesn’t hurt anyone.

Examples:

• brainstorming dinner ideas
• creating trivia questions
• suggesting vacation activities
• generating writing ideas
• organizing notes
• creating a packing list
• brainstorming birthday gifts
• suggesting hobbies
• helping outline a personal project

Experiment freely.

Check anything that seems odd.

Medium-Risk AI Use

Now we’re dealing with information that could affect a meaningful decision.

Examples:

• comparing expensive products
• researching a company
• planning a major trip
• researching career options
• evaluating business ideas
• reviewing financial concepts
• interpreting complicated documents
• researching home repairs

AI can be incredibly helpful here.

But don’t make the final decision based solely on one AI conversation.

Verify important facts.

High-Risk AI Use

This includes:

• medical decisions
• medication questions
• legal decisions
• taxes
• major financial decisions
• emergencies
• personal safety
• decisions that could seriously affect another person

AI can help you understand terminology, organize questions, summarize information, or prepare for a conversation with an appropriate professional.

It shouldn’t become your only source.

Use AI as a Thinking Partner

I use AI differently than someone who expects it to provide “the answer.”

I like questions such as:

“Give me three ways to think about this.”

Or:

“What am I missing?”

Or:

“What’s the strongest argument against my position?”

Or:

“What assumptions am I making?”

Or:

“What questions should I ask before deciding?”

Those prompts don’t surrender the decision.

They improve the decision-making process.

Ask AI to Criticize Its Own Answer

Here’s a habit I think everyone should develop.

After getting an important answer, ask:

“What parts of your answer might be wrong, uncertain, outdated, or dependent on assumptions?”

Then ask:

“What should I independently verify?”

Then:

“Give me primary sources where possible.”

You can even ask:

“Argue against your previous recommendation.”

You may discover information the first answer ignored.

This isn’t foolproof.

AI can be wrong while checking whether it was wrong.

But it changes your relationship with the tool.

You’re no longer passively accepting information.

You’re interrogating it.

Primary Sources Matter

Whenever possible, go to the original source.

If AI tells you something about Social Security, verify it with the Social Security Administration.

If it’s an IRS question, look at IRS information.

If it’s an FDA regulatory claim, look at the FDA.

If it’s a product specification, check the manufacturer’s documentation.

If it’s a company announcement, find the company’s announcement.

If it’s research, look for the actual research.

AI is excellent at helping you locate, understand, and organize information.

That doesn’t mean AI needs to be the final destination.

AI Can Also Be Outdated

Another thing to understand is that different AI systems have different access to current information.

Some can search the web.

Some rely primarily on their training data for certain responses.

Some tools use search selectively.

Features change frequently.

So if you’re asking:

“Who is the current CEO?”

“What does this product currently cost?”

“What law currently applies?”

“What happened today?”

fresh information matters.

Ask the AI whether it actually checked a current source.

Then inspect the source.

Dates matter.

What About Privacy?

This is where skepticism becomes especially useful.

You shouldn’t assume that every AI product handles information the same way.

Privacy policies differ.

Account types differ.

Settings differ.

Features change.

Before uploading sensitive information, find out what you’re agreeing to.

OpenAI, for example, provides consumer privacy controls for ChatGPT, including the ability to control whether future chats are used to improve models. It also offers Temporary Chat, which OpenAI says does not contribute to model training or memory and is automatically deleted according to its stated retention process. (OpenAI)

Anthropic likewise publishes information about how Claude consumer data is protected and the circumstances under which conversation information may be accessed. (Anthropic Privacy Center)

The point isn’t that one sentence summarizes every privacy policy.

It doesn’t.

The point is that you should know the controls exist and use them appropriately.

Things I Wouldn’t Casually Put Into AI

Regardless of which service I’m using, I would be extremely cautious about entering:

• passwords
• Social Security numbers
• authentication codes
• complete credit-card information
• banking credentials
• confidential company data
• private customer information
• information you’re contractually prohibited from sharing
• private information about another person

Ask yourself one simple question:

“Would I be comfortable putting this information into another online service?”

If the answer is no, stop and investigate first.

AI Scams Change the Trust Equation

AI isn’t only something you use.

It’s something scammers can use too.

That’s another reason AI literacy matters.

The Federal Trade Commission has warned about scammers using AI voice cloning in family-emergency scams.

A scammer may obtain a sample of someone’s voice from online content, create synthetic audio that sounds like that person, and use it to make an urgent request for money.

The FTC recommends independently contacting the family member using a telephone number you already know rather than trusting the incoming voice. (Consumer Advice)

That’s worth remembering.

Your ears are no longer perfect authentication.

Neither are your eyes.

Create a Family Verification Rule

Here’s one practical idea.

Establish a family rule for emergency requests.

If someone calls asking for money because of an emergency:

1. Don’t act immediately.
2. Hang up.
3. Call that person through a number you already have.
4. Contact another trusted family member if necessary.
5. Never allow urgency to prevent verification.

You could even establish a private family verification question or phrase.

The important thing is not to rely solely on a familiar-sounding voice.

AI changes what can be faked.

Our verification habits need to change with it.

Watch for Manufactured Urgency

Scammers have always used emotion.

AI just gives them more convincing tools.

Common pressure tactics include:

• “You must do this right now.”
• “Don’t tell anyone.”
• “Your account will be closed.”
• “Your grandchild is in trouble.”
• “You owe money immediately.”
• “Buy gift cards.”
• “Send cryptocurrency.”
• “Wire money now.”

The FTC repeatedly warns consumers that demands for payment through gift cards, cryptocurrency, or urgent wire transfers are major scam indicators. (Consumer Advice)

Slow down.

Urgency is often part of the attack.

AI Detection Tools Aren’t Magic Either

You may have seen software claiming it can tell whether text, audio, or an image was generated by AI.

Be careful.

Detection can be useful, but it shouldn’t be treated as perfect.

The FTC has discussed different approaches for combating harmful voice cloning while also acknowledging limitations in detection and watermarking approaches.

There isn’t one magical detector that solves every problem. (Federal Trade Commission)

Verification should involve context, trusted communication channels, and common sense.

Can AI Manipulate Us?

Potentially.

But the broader issue isn’t unique to artificial intelligence.

Marketing manipulates.

Political advertising manipulates.

Social media algorithms influence what we see.

Salespeople persuade.

Headlines are designed to get attention.

AI adds another layer because it can personalize communication and interact conversationally.

That makes awareness important.

Whenever you find yourself emotionally reacting to an AI conversation, remember:

This system generates responses.

It doesn’t become wiser merely because it sounds calm.

It doesn’t become your friend because it remembers what kind of fishing you enjoy.

It doesn’t become an expert merely because it sounds certain.

Don’t Confuse Helpful With Human

AI can be extraordinarily helpful.

It can also feel supportive.

You can brainstorm with it.

Practice an interview with it.

Talk through a problem.

Ask embarrassing questions you might not want to ask another person.

That’s useful.

But useful software is still software.

Maintaining that distinction helps you use the technology more intelligently.

Don’t Make the Opposite Mistake Either

There’s another extreme.

Some people hear that AI sometimes generates false information and conclude:

“Then it’s useless.”

That’s like discovering Wikipedia contains errors and deciding the internet has no value.

The right response to limitations isn’t necessarily rejection.

It’s learning how to work around them.

Cars can crash.

We still drive.

Email contains scams.

We still use email.

Online banking creates cybersecurity risks.

Millions of people still bank online.

We develop safety habits.

AI will be similar.

A Better Definition of Trust

Instead of asking:

“Do I trust AI?”

Ask:

“Do I trust this particular answer enough for this particular purpose?”

That’s a better question.

If I’m asking AI to give me ten podcast title ideas, my threshold is low.

If I’m asking it about a medication, my threshold is extremely high.

Context matters.

Five Questions to Ask About an Important AI Answer

Before acting on something important, ask:

1. Where did this information come from?

Look for sources.

1. How current is it?

Check dates.

1. What assumptions were made?

Ask the AI directly.

1. What’s missing?

No prompt contains every piece of context.

1. What happens if this is wrong?

This may be the most important question.

If the downside is trivial, proceed.

If the downside is serious, verify.

Skepticism Isn’t Fear

A skeptical user can be an excellent AI user.

The key is the difference between skepticism and automatic rejection.

Skepticism says:

“Show me.”

Fear says:

“I don’t want to understand it.”

Blind enthusiasm says:

“The machine said it, so it must be right.”

None of those extremes are particularly useful.

I prefer informed skepticism.

Experiment.

Question.

Verify.

Learn.

AI Shouldn’t Make Your Decisions

This is where I draw a fairly clear line.

AI can help me decide.

It doesn’t make the decision for me.

There’s a difference.

Suppose I’m considering a new business idea.

I might ask AI to:

• identify weaknesses
• estimate costs
• suggest competitors to research
• create questions
• challenge assumptions
• organize a plan

Then I make the decision.

Suppose I’m making a major purchase.

AI can compare.

I choose.

Suppose I have a health concern.

AI can help me organize symptoms and questions.

Qualified health professionals and I deal with the actual care decisions.

AI stays in its lane.

Your Judgment Still Matters

The better AI becomes, the more important human judgment becomes.

That may sound backwards.

But consider what happens when producing information becomes almost effortless.

We get more information.

More content.

More options.

More convincing text.

More convincing images.

More synthetic audio.

More opinions.

More noise.

The scarce resource becomes judgment.

What matters?

What’s accurate?

What’s relevant?

What’s nonsense?

What should I ignore?

Those are human questions.

Teach Your Kids This Too

AI literacy isn’t only important for people over 50.

Young people need exactly the same lesson.

Maybe more.

Knowing how to generate something with AI is not the same as knowing whether it’s correct.

Teaching kids:

• how to verify
• how to question
• how to research
• how to identify sources
• how to think independently

may become increasingly important as AI-generated information becomes commonplace.

The technology changes.

Critical thinking doesn’t go out of style.

My Rule for AI

Here’s the simple rule I keep coming back to:

Use AI to expand your thinking, not replace it.

Ask it questions.

Have it challenge you.

Use it for research.

Let it organize information.

Let it handle boring first drafts.

Let it suggest possibilities.

Then bring your own brain back into the process.

It’s still required.

So, Should You Trust AI?

Not completely.

And that’s fine.

You don’t need complete trust to get value from a tool.

You need to understand what the tool does well and where it can fail.

Trust AI with brainstorming before you trust it with medical decisions.

Trust it with organizing notes before you trust it with legal conclusions.

Trust it enough to experiment.

Distrust it enough to verify.

That’s the balance.

AI doesn’t need blind believers.

It needs informed users.

And if you’re over 50 wondering whether your natural skepticism makes you bad at artificial intelligence, I would argue almost the opposite.

That little voice in your head saying:

“Are we sure about this?”

Keep it.

You’re going to need it.

Sources and Further Reading

NIST’s Generative AI Risk Management Profile discusses confabulation and other generative AI risks. (NIST)

The Federal Trade Commission provides guidance on AI voice cloning, family emergency scams, impersonation scams, and fraudulent payment requests. (Consumer Advice)

OpenAI publishes information about ChatGPT consumer privacy and data controls. (OpenAI)

Anthropic publishes information about privacy and data handling for Claude users. (Anthropic Privacy Center)

AI for Ordinary People

At AIforOrdinaryPeople.com, the goal isn’t to convince you that artificial intelligence is perfect.

It isn’t.

The goal is to understand the technology well enough to use the useful parts, avoid some of the stupid parts, and keep your own judgment firmly involved.

Written by

Joe Foley

Joe Foley is the creator of AI for Ordinary People. He helps beginners, parents, creators, and small business owners use AI in simple, practical ways. Joe has been podcasting since 2013 and creates plain-English guides, prompts, and workflows for people who want useful help without tech jargon. His goal is to make AI feel less confusing and more useful for real life, real work, and everyday decisions.