What ChatGPT actually is (and what it isn't) and what i have learnt so far.
A few weeks ago I was working on a project. I needed to find social media accounts on a topic I wanted to research. I asked ChatGPT for 10 of them.
It gave me 10. Confident, detailed, exactly what I asked for.
Then I checked each one. Three were dead accounts. Two were completely inactive. And one, my personal favourite, was selling knives. Nothing to do with my topic. Not even close.
ChatGPT had no idea. It gave me 10 accounts that fit the pattern of what a list should look like. Once I understood why that happens, everything shifted. That is what I want to give you today.
What ChatGPT actually is
ChatGPT is not automatically a search engine. This is the biggest misunderstanding people have.
By default, it works entirely from what it learned during training.
Think of it like the "auto-complete" on your phone, but on steroids. It was trained on trillions of words from books, websites, and conversations.
From that, it learned how humans build sentences. When you ask it a question, it isn’t "looking up" the answer. It is calculating which word is statistically most likely to follow the one before it.
Because it’s so good at patterns, it can write a poem or a business plan in seconds. But because it’s just "guessing" the next word, it can be confidently wrong. In the AI world, we call this a hallucination. It isn't lying to you; it’s just following a pattern that doesn't match reality.
Genuinely impressive. Also the root of almost every problem beginners run into.
What ChatGPT is not
Not a fact-checker.
It generates language from patterns not verified information. So it can sound completely right and still be wrong.T here is no internal alarm. It does not know when it does not know something.
Worth noting though: ChatGPT now has multiple modes you can activate. Thinking mode checks its own logic. Deep Research digs deeper. Web Search pulls live information. We shall see how to find these in the coming weeks.
Not current.
ChatGPT's pretrained knowledge has a cutoff date, and it varies depending on which model you are on. As of mid-2026, most versions sit somewhere between late 2024 and mid-2025. Anything outside that window, it either does not know or fills in from old patterns.
A changed regulation, an updated statistic, it will not flag this for you. ChatGPT does have a live browsing feature that can pull current information, but it does not always trigger automatically.
Check which model you are using, and when in doubt, verify anything time-sensitive yourself.
Not a person.
It will say things like "I understand how you feel." But it has no internal feelings. Someone put it well: "all outside and no inside." It is just a machine predicting the next best word.
It produces the language of empathy with none of the experience behind it like a human.
Not your memory.
Each conversation starts completely fresh unless you turn Memory on in settings. Everything you told it last week, the context of your business, gone.
You are starting from zero every time, unless you are picking up from the same chat history.
Not a replacement for expertise.
It can explain, summarise, and help you think. It cannot give you advice a lawyer or accountant would stand behind. Verify anything important with a qualified human.
What it is genuinely good at
Drafting. Brainstorming. Summarising. Explaining complex things in plain language. Rewriting your own words so they land better.
Those things it does well. The skill is knowing where its confidence is earned.
This week's tip (carry this one with you)
ChatGPT does not look things up unless you switch on the web browsing. It only predicts. Treat every answer as a starting point, not a final answer. Anything that matters, verify.
One last thing. Everything above, the pattern prediction, the hallucinations, the knowledge cutoff, applies to every major AI tool. Claude, Gemini, all of them broadly work the same way at their core.
Whether you run a business, teach a class, use it everyday or are just curious about AI, this is your foundation. And it is what this whole series builds on.
Hit reply and tell me if this brings some deeper understanding to you. Am curious and excited at the same time.
Harriet
Founder, SaviteckX
If this was useful, forward it to a friend who uses AI and could use a little more clarity. That is the best thing you can do for this newsletter.