7 beginner mistakes and how to avoid them
Almost everyone starting with AI trips over the same few things: treating it as a search engine, trusting numbers blindly, writing vague requests, quitting after the first answer. We gather the seven main mistakes and a short antidote to each.
Most first encounters with ChatGPT end after one evening, and over the same handful of mistakes. Here are seven of them, each with its antidote. This is the lesson that ties everything together.
Mistake 1. Treating the AI like a search engine
"berlin weather", "turkey holiday cheap" — that's how you talk to Google, and you'll get an averaged answer.
Antidote: talk in full sentences, like you would to an assistant (the first-dialogue lesson). Situation + constraints + the result you want, and the answer becomes yours rather than everyone's.
Mistake 2. Taking facts on trust
The model can produce a hallucination — a plausible invention: a Harvard study that doesn't exist, a made-up number, a fabricated link. It sounds confident, and the beginner forwards it to a client.
Antidote: facts, numbers, links and legal conclusions get checked (the 10-jobs lesson). The AI writes excellent drafts, but the facts are on you.
Mistake 3. A vague request
"Make it better", "write something about marketing" — the model has to guess, and it guesses wrong.
Antidote: swap fog for specifics. Not "better", but "cut it in half and strip out the corporate-speak". Not "about marketing", but "three post ideas for a coffee shop by the station, friendly tone".
Mistake 4. Stopping at the first answer
The first answer is a draft, and the beginner reads it as the finished thing and gets disappointed.
Antidote: iterate. "Too formal", "add examples", "give me another version". Remember from the first-dialogue lesson: the real skill isn't a perfect first request, it's fast rounds of editing.
Mistake 5. Dumping everything into one chat
The report, the recipe and the holiday plan in one conversation — the context tangles, the answers blur, and nothing is findable afterwards.
Antidote: new topic, new chat; rename the keepers; run repeating work in projects (the tidy-chats lesson).
Mistake 6. Pasting secrets
Passwords, full card numbers, other people's personal data, work secrets — none of it belongs in a public chat. The Samsung story from the settings lesson is what that does to a company.
Antidote: secrets stay out of the chat; sensitive things get anonymised.
Mistake 7. Expecting the AI to know everything, right now
Plenty of people picture the AI as an all-knowing oracle wired into everything. In reality a base model's knowledge stops at its cutoff date, and it only sees your personal context if you hand it over. Ask a plain chat "what's the dollar worth right now" or "what was in the news yesterday" and you'll get something stale or invented.
Antidote: for fresh facts, switch on web search mode (if the service has one) or check yourself; for personal context, supply it via custom instructions and files.
The one habit that closes half this list
Care to guess which?
Have a think before you read on. The answer: give context. Vague requests, the search-engine style, "the AI doesn't know me" — one thing cures all of it: describe the situation, the goal and the constraints. Almost any bad answer isn't a dumb AI — it's a request that was short of context.
The deliberate user's checklist
Before you get annoyed at an answer, run down this list:
- did I give context (who, why, what matters)?
- did I check the facts, if they matter?
- did I ask for another pass when it wasn't right?
- are there no secrets in the chat?
- is the topic in its own chat?
One "no" and there's your reason — and it's in your hands. Next up on the track is "Prompt engineering", where all of this becomes a systematic skill.
The eighth, invisible mistake: never starting
The most common mistake isn't on this list — it's not using AI every single day. People read about what it can do, nod, and close the tab, and the habit never forms. Remember the first-week method from the 10-jobs lesson: not everything at once — one job a day. And keep the tool one touch away: the first-screen icon was the apps lesson. A tool that's one step away gets reached for ten times a day, and within a month that changes the work.
What separates someone who opens an AI every day from someone who read about it and closed the tab is not talent or luck — it is a habit. And it has already started.
Do this now
Think of the worst AI answer you've had — during this course or ever. If you haven't done the tasks yet, borrow an example from the lesson: «make it better» will do fine. Work out which of the seven mistakes sits inside it and rewrite the request with the antidote. The gap between before and after is the step from beginner to confident user.
Short questions on the lesson — with an explanation for every answer.