Strengths and Limitations

Most people who give up on ChatGPT do it for one of two reasons. Either they asked it to do something it is bad at, got a bad result, and decided the whole thing was hype. Or they trusted it with something important, it got a detail wrong, and they got burned.

Both come from the same gap: not knowing where the line is. This lesson draws that line. You will learn which jobs ChatGPT handles well, where it reliably falls short, and a short set of habits that let you use the first without getting caught by the second.

Where ChatGPT is strong

Remember how the model works: it predicts likely text based on patterns in language. So it is strongest when the job is shaping words, and weakest when the job is knowing or checking facts.

Drafting and editing written content

This is the task ChatGPT is best suited for. First drafts of emails, letters, posts, reports, bios, and instructions. Rewriting something blunt into something polite, something long into something short, something technical into something plain. It can match nearly any tone you describe, and editing a decent draft is faster than facing a blank page.

Brainstorming

Need twenty names for a new program, ten angles for a newsletter, or five ways to explain your pricing? It produces options quickly. Most will be ordinary. You only need one good one, and a long list gets you there faster than staring at the wall.

Explaining and summarizing

It is good at taking something complicated and restating it for a specific reader: a contract clause for a first-time renter, a medical term for a worried parent, a long email thread for someone who missed it. Treat a summary as a map that shows you where to read closely, not a replacement for reading the parts that matter.

Organizing messy material

Scribbled notes, voice-memo transcripts, a pile of bullet points from a meeting. It can turn chaos into a clean list, a checklist, a schedule, or a table.

Working with code

It can write, explain, and help debug code in many programming languages. Even here, the code needs testing before you rely on it.

Where ChatGPT falls short

It can make things up

A hallucination is confident but incorrect information: a statistic nobody measured, a research paper that does not exist, a quote nobody said, a product feature that was never built. It happens because a fake citation looks just like a real one to a pattern-predictor.

The rule is simple. When ChatGPT gives you a statistic, a citation, a quote, a date, or a name, verify it independently before you use it. Look up the source yourself. If you cannot find it, treat the claim as unverified and leave it out.

It does not know recent events by default

The model's knowledge stops at its training cutoff. Some versions add a web search tool, and when it runs you can usually see the sources. Without that, it may not know about recent news, new laws, or changed prices, and it may guess instead of saying so. For anything current, check a reliable, up-to-date source.

It does not remember you across sessions

ChatGPT does not learn from your conversations, and it does not remember everything across sessions. Within one chat, it can refer back to what you said earlier. Start a new chat, and by default it starts fresh. Some apps offer a memory feature that saves short notes and adds them to later chats, but that is a stored reminder, not the model learning. If a detail matters, put it in the prompt again.

Math, counting, and precise logic

Because it predicts text rather than calculating, it can produce numbers that look tidy and are simply wrong. It can miscount letters in a word or items in a list, and it can lose track of steps in a long logic problem. Some versions can run actual code to do calculations, which helps, but for totals, quotes, budgets, and anything financial, check the numbers in a calculator or spreadsheet.

Bias and blind spots

It learned from human writing, and human writing contains biases. Be especially careful when using it for anything involving people: hiring, evaluations, descriptions of groups, or sensitive topics. You are responsible for the final judgment.

Facts about your life

It does not know your hours, your prices, your policies, or your family. If you do not supply those facts, it may fill the gap with something plausible. That is one of the fastest ways for AI to embarrass you.

Worked example: asking for accuracy

Here is a request where the limits matter.

Before:

Write a short article about why reading to kids matters, with
some statistics.

This invites the model to produce numbers that sound authoritative, and you have no way of knowing which are real.

After:

Write a 300-word article for [AUDIENCE, e.g. parents at our school]
about why reading aloud to kids matters.

Do not include any statistics, study results, or quotes.
Where a fact would strengthen a point, write [FACT NEEDED: short
description] so I can find a real source and fill it in myself.

Only include claims you are confident are widely accepted. If you
are unsure about something, say so rather than guessing.

The second prompt plays to the model's strength, clear and warm writing for a specific audience, and keeps it away from its weakness, inventing specifics. You end up with a solid draft and a short list of facts to look up, instead of a polished article with hidden landmines.

Habits that keep you safe

1. Use it for the draft, not the final word. Anything important, public, or carrying your name gets a human read before it goes out.
2. Verify facts independently. Statistics, citations, quotes, names, dates, and legal or medical claims all get checked against a real source.
3. Supply the facts only you know. Paste in your details instead of hoping it guesses right.
4. Break big tasks into steps. Several focused prompts beat one sprawling one.
5. Ask it to flag uncertainty. "If you're unsure, say so" does not guarantee honesty, but it helps.

Try this

Give yourself 15 minutes.

Step 1. Ask ChatGPT a factual question in an area you know well: your job, your neighborhood, a hobby. Include a request for a statistic or a source.

What are three facts about [TOPIC YOU KNOW WELL]? Include a
statistic and the source it comes from.

Step 2. Check every claim and every source. Mark each one: correct, partly correct, wrong, or could not verify.

Step 3. Now give it a drafting task on the same topic, using the "after" pattern from the worked example, with placeholders instead of statistics.

Step 4. Compare. Which answer would you be comfortable sending to someone? Write down one rule for yourself about when you will check its work.

Seeing it get something wrong in a subject you know well is the fastest way to calibrate how much to trust it everywhere else.

What to remember

• ChatGPT is best at drafting and editing written content, brainstorming, explaining, summarizing, and organizing.
• It can make up facts in a confident tone; verify every statistic, citation, and quote independently before using it.
• It does not learn from your conversations or remember everything across sessions; each new chat starts fresh by default.
• Check math, counts, and anything financial with a calculator or spreadsheet.
• A human reviews anything important before it leaves your hands.

These three lessons give you the foundation: how the model works, how to prompt it, and where to trust it. The paid modules build on that with Role-Based Prompting, Chain of Thought for harder reasoning tasks, working with Long Documents, ChatGPT for Business, and Automated Workflows that turn one-off prompts into repeatable routines.