Now the specific thing you type into. What happens to your words on the way in, how the reply gets built one piece at a time, and why clarity beats clever wording every single time.
The first thing that happens to your message has nothing to do with meaning. ChatGPT breaks your text into small pieces called tokens before it does anything else with it. By OpenAI's own rule of thumb, one token works out to roughly four characters of English text — about three-quarters of a word.
So a short sentence might run a dozen or so tokens, and a long one considerably more. This is not a technicality — it is the actual unit the rest of the process runs on. Everything downstream, the guessing and the building of a reply, happens one token at a time, not one word or one sentence at a time.
Here is the part that surprises people: the model does not draft the whole reply and then show it to you. It looks at every token so far — your message, plus whatever it has already written — and repeatedly asks itself one question: what is the most likely next piece? Then it writes that piece, adds it to the pile, and asks again. Autocomplete on steroids, not a system that reasons the way a person does.
The pattern behind those guesses comes from training: months spent reading a huge slice of the internet — books, websites, code, conversations — while billions of internal numbers, called weights, get nudged so the next-word guesses get a little less wrong. As you already know from earlier in this course, that process is over by the time you are chatting with it.
One more step happens after that first pass: fine-tuning, which teaches the model to behave like a helpful assistant answering your question, rather than just continuing whatever random text it has been trained on. That is the difference between a model that finishes your sentence with an ad and one that actually answers you.
A prompt is simply what you type in — the question, instruction, or context you give it before it answers. There is no secret command language to learn. Plain English works.
What actually changes the answer is not clever wording. It is clarity: say what you want, who it is for, and what a good answer looks like — the same way you would brief a smart intern who has read everything but knows nothing about your specific situation. That is usually the entire gap between a vague reply and a genuinely useful one.
There is a deeper, more technical version of this — careful formatting, embedded examples, step-by-step instructions — that people building AI-powered products spend real time learning. It is a real skill. It is just not one you need for everyday use, and this course will not pretend otherwise.
Everything you type gets chopped into tokens first, and the reply gets built one token at a time from patterns laid down months earlier during training. The one lever you actually control is the prompt — and the whole trick there is clarity, not cleverness: say what you want, who it is for, and what good looks like.