1. Learning
For months, the model reads billions of pages (books, websites, code…). It does not store them like a library: it picks up patterns — grammar, common facts, ways of reasoning.
Claude, Gemini, ChatGPT… Everyone is talking about them, but how do they actually work? This 5-step guide takes you from your very first question to advanced uses, with no jargon.
Claude (made by Anthropic) and Gemini (made by Google) are conversational assistants. You type a question or a request in plain language and they answer within seconds: a text, a table, a summary, an email, some code, sometimes an image.
Behind these assistants is a Large Language Model (LLM). It is a program that has “read” a gigantic amount of text and learned one very specific skill from it: guessing the most likely continuation of a text.
The picture to rememberWhen you type a text message, your phone keyboard suggests the next word. A generative AI does the same thing, only vastly more powerful: it has learned so much text that by chaining “next words” it produces complete, structured answers tailored to your request.
The model computes the probability of every possible continuation, picks a word, then starts again. Word after word, the answer is built.
You don’t need to be a computer scientist to get the gist.
For months, the model reads billions of pages (books, websites, code…). It does not store them like a library: it picks up patterns — grammar, common facts, ways of reasoning.
Human teams then teach it to be useful: answer the question asked, follow instructions, admit when it does not know, refuse dangerous requests.
When you write your request (the “prompt”), the model reads it along with the whole conversation and generates its answer piece by piece. The more precise your request, the better the answer.
They make the difference between a bland answer and a truly useful one.
Who are you, who is the result for, what is the goal? “Write an email” gives a generic result; “Write a short, friendly email to a client who has been waiting a week for their quote” gives something you can use.
A list, a table, three paragraphs, 100 words max, a formal tone… Say it and the AI will stick to it.
One example of what you expect beats a long explanation.
The first answer is only a draft. Ask for “shorter”, “simpler”, “add an example”: it is a conversation.
Proofread, check figures, names and dates before reusing an answer, especially if it is going to a client.
Want to go faster? My free PromptLogic tool helps you build a well-structured request.
Try PromptLogicAvoid pasting sensitive information into a consumer AI: passwords, health data, bank details, customer files. In business plans (Google Workspace, Claude Team or Enterprise…), your content is not used to train models and stays under your organisation’s control. On personal plans, check your account’s privacy settings.
Follow the steps in order, or jump straight to what interests you.
You are here: what an AI is, how it works and the right habits.
Claude and Gemini in detail, other assistants, free and paid plans, and how to choose.
Prompt, token, hallucination, agent, MCP… the essential words, explained simply.
Teach the AI your methods once and for all: skills, in Claude and in Gemini alike.
Deep research, connectors, agents, automation with Apps Script and the API.
Yes, Claude and Gemini both have a free version, enough to explore and for occasional use. Paid plans (around $20 a month for the entry tier) unlock the most powerful models, higher usage limits and advanced features.
The one that fits your tools best. If you live in Gmail, Google Drive and Android, start with Gemini. If you write a lot, analyse documents or want a particularly rigorous assistant, try Claude. The best approach is to test both with the same request: step 2 of the guide helps you do that.
Not in the human sense. It has learned patterns from a huge amount of text, which lets it produce relevant answers. But it is not aware of what it says and cannot check on its own that it is right: your review remains essential.
Mostly it changes how you work: it handles repetitive tasks (first drafts, summaries, formatting) and leaves you more time for what needs judgement, relationships and expertise. People who use it well get a head start.
No. Everything is done by writing in plain language, like in a messaging app. Coding only comes in for advanced uses, and even then the AI can write the code for you.
Développeur freelance spécialisé dans l'optimisation des flux de travail professionnels, le développement d'add-ons sur-mesure (Google Sheets, Docs, Gmail) et l'intégration de modèles IA (Gemini).
Updated on