ChatGPT · Beginner

The 3 mistakes beginners make with ChatGPT (that experts never do)

These aren't obscure. They're the same three mistakes, in the same order, every time.

Updated May 2026

By Mark Reeves — author of all 47 AI Field Guides. Written and tested in a working business.

AI-generated portrait. Pen name. Real work.

I want to start with something that might be slightly uncomfortable to say out loud: I made all three of these mistakes. Not briefly, either. I made them for long enough that they felt normal.

The reason I'm writing this is not to make you feel bad about your ChatGPT habits. It's the opposite. These mistakes are so consistent — so predictable, so universal — that I started thinking of them less as individual failures and more as a kind of natural learning path. Almost everyone passes through the same three stages, in the same order, before something clicks.

Knowing what those stages are means you can move through them faster. That's all this is.

Mistake 1: Treating it like a search engine

What it looks like

You type something like: "best way to negotiate a salary raise." Or "how to write a project brief." Or "symptoms of burnout at work."

Short. No context. No explanation of who you are or what you actually need. The same kind of query you'd drop into Google.

ChatGPT answers. It answers fluently, in complete sentences, with structure. It looks helpful. And it is — sort of. But what you get is the most generic version of an answer to that question. The version that would be equally true for a 22-year-old intern and a senior director. For someone in their first corporate job and someone running their own business.

It's not wrong. It's just not for you.

Why it happens

Google trained a decade of search habits into most of us. Short. Specific keywords. No sentences. That approach works because Google has enough context about you — your location, your history, the current date — to fill in the gaps.

ChatGPT has none of that. It starts every conversation completely cold. It doesn't know what you do, what you've already tried, what outcome you're looking for, or what format would actually be useful to you. When you give it a thin prompt, it has no choice but to fill in the gaps with statistical averages. The result is an answer that's plausible for everyone and actually useful for no one in particular.

The fix: context before questions

Before you ask your question, spend two sentences on who you are and what you need. Not an essay — just enough to give the tool something real to work with.

The difference between this:

best way to negotiate a salary raise

And this:

I'm a marketing manager with six years of experience at a mid-size agency.
I've been in my current role for two years and have a review coming up next month.
I want to ask for a 15% raise. What's the most effective way to frame that
conversation with a manager who responds well to data but is sensitive about
budget pressure right now?

The second prompt gives ChatGPT something to work with. The answer you get back will be specific to your situation — your role, your relationship, your timing — not a generic list of negotiation tips that could have come from a self-help book published in 2012.

The rule of thumb: before you ask the question, write one sentence about who you are in this context, and one sentence about what a useful answer actually looks like for you. That alone changes the output dramatically.

Mistake 2: Accepting the first answer

What it looks like

ChatGPT responds. The answer is coherent, well-organized, grammatically correct. It covers the main points. You read it, think "that's pretty good," and use it.

What you got was a first draft. What you treated it as was a finished product.

Why it happens

This one isn't laziness. I want to be clear about that because people assume it is, and it isn't.

ChatGPT writes with a tone of authority. It doesn't hedge the way a human draft might. There's no "I'm not sure, but..." or "you might want to rethink this part." It presents its output with the quiet confidence of something that has been thought through. And that confidence signals completeness — even when the answer is generic, too long, jargon-heavy, or missing the actual point you needed.

Pushing back on something that sounds confident feels unnecessary. It feels like you're second-guessing a result that seems fine. But "seems fine" is different from "is exactly what I need."

The first answer is always a starting point. That's its job. Your job is to tell it what to do next.

The fix: iteration instructions that get specific

The most effective follow-ups are concrete and directional. Not "make it better" — that's too vague. Not "can you try again" — that produces a variation on the same thing.

Here's what specific iteration looks like:

Make it shorter. The current version is 400 words — cut it to 150.

Cut the jargon. Replace "leverage," "synergy," and "scalable" with plain language.

Now rewrite the opening paragraph for someone who is skeptical of this idea.
They need to be persuaded, not informed.

The tone is too formal. Rewrite this as if you're explaining it to a smart
colleague over coffee — not a committee.

Each of these is a direction, not a wish. You're telling the tool exactly what to change and why. That specificity is what separates a useful second draft from a reshuffled version of the first one.

The habit to build: before you close a ChatGPT conversation, ask yourself whether you pushed once. Not a dozen times — once. One round of "make this shorter," or "rewrite this for a skeptical reader," or "cut everything that isn't essential." That single iteration usually produces something meaningfully better than what you started with.

Mistake 3: Not verifying factual claims

What it looks like

ChatGPT states a statistic. Something like: "studies show that 70% of employees report feeling disengaged at work." Or it quotes something attributed to a named person. Or it describes a regulation as if it's settled fact.

You read it. It sounds right. It feels like something you'd vaguely heard before. You use it — in an email, a presentation, a blog post, a client document.

The statistic doesn't exist in the form stated. Or the quote is misattributed. Or the regulation is from a different country, or has been updated, or was never what the summary described.

Why it happens

ChatGPT is not a search engine. It doesn't retrieve facts from a database of verified sources. It generates text by predicting what a plausible, coherent response looks like, based on patterns in its training data.

That means it can produce a statistic with a specific percentage, a quote with a name attached, and a regulatory summary with confident language — all in the same paragraph, all sounding equally authoritative — regardless of whether any of them are precisely accurate.

This isn't the model lying. It's the model doing exactly what it was built to do: produce fluent, plausible-sounding text. Specificity in the output is a stylistic feature, not a verification signal. A number with a percentage doesn't mean the number was checked. A name attached to a quote doesn't mean the person said it.

The categories where this causes the most damage: statistics cited in public writing, quotes attributed to real people, and regulatory or legal claims. These are the areas where a single error can undermine everything else around it — or, in professional contexts, create real liability.

The fix: the single-step verification habit

You don't need a fact-checking system. You need one habit: before you use any specific factual claim from ChatGPT in anything that matters, find the original source.

Not a summary of the source. The source itself.

The way to do it: take the specific claim — the statistic, the quote, the regulation — and search for it directly. Look for where it came from. Click through to the actual document, study, or transcript. Confirm that it says what you're about to repeat.

When you go looking, one of four things will happen: the source is real and accurate, the source is real but the number is different, the source exists but is outdated, or the source doesn't appear to exist at all. The first outcome is the only safe one. The other three are the ones that, without this habit, end up in your work.

The practical shortcut for high-stakes outputs: treat every specific claim from ChatGPT as a lead, not a source. Use it to know what to look for. Then go find the real thing. This takes two minutes. The alternative — trusting the output and being wrong publicly — takes much longer to recover from.

These aren't beginner failures. They're beginner defaults.

The difference between someone who uses ChatGPT well and someone who doesn't isn't intelligence. It's not even effort, necessarily. It's that the experienced user has learned — usually through a few specific bad experiences — exactly how the tool behaves.

They know that thin prompts produce generic answers, so they add context before they ask. They know that the first response is a starting point, so they push once before they accept it. They know that confident-sounding factual claims need checking, so they verify the ones that matter before they use them.

None of this is advanced. It's just knowledge of the tool's actual behavior. And the reason most beginners don't have it is simply that they haven't been told — they've been handed the tool and left to discover the edges on their own.

That's what these three mistakes have in common. They're not signs that someone is bad at this. They're signs that no one showed them how the thing actually works.