The honest explanation
When you send a message to Claude or ChatGPT, the AI doesn't look up the answer in a database. It doesn't 'think' the way you do. It generates a response by predicting — at each step — what word or phrase is most likely to follow everything that came before.
This prediction engine was trained on an enormous amount of text: books, articles, code, conversations, websites. It learned patterns. What words tend to follow other words. What a helpful answer to a question looks like. What code that does X usually looks like.
That's the core of it. Everything else — the different products, the safety features, the pricing — is built on top of this foundation.
Why this mental model matters
Understanding how AI works explains things that confuse many people:
**Why AI sometimes makes things up.** The model generates text that *sounds* right based on patterns — even when it's wrong. It's not lying. It's predicting. This is called 'hallucination' and every major AI system does it.
**Why AI is surprisingly good at writing and summarising.** These tasks have clear, rich patterns in text. The model has seen millions of examples.
**Why AI struggles with maths and current events.** Precise calculations and things that happened after the training cutoff don't have reliable patterns the model can draw on — so it improvises, and often gets it wrong.
**Why your prompt matters so much.** The model uses everything you give it as context. A vague question gets a vague answer. A detailed, specific prompt gets a much better result.
The four tools you'll hear about most
There are four AI tools that dominate business and professional use right now. They're all built on the same basic technology but made by different companies with different strengths.
What each tool is genuinely best at
**ChatGPT** (OpenAI) — The one that started the mainstream AI wave. Huge user base, good general capability, strong coding and analysis. The GPT-4o model handles images and voice natively. Best for: general tasks, research, drafting, image analysis.
**Claude** (Anthropic) — Known for being particularly good at long documents, nuanced writing, and following complex instructions carefully. Very strong at analysis and tends to be more cautious about making things up. Best for: long-form writing, document analysis, careful reasoning tasks.
**Microsoft Copilot** — Built directly into Microsoft 365. If your organisation uses Word, Excel, Teams, or Outlook, Copilot works *inside* those apps. It can summarise your email inbox, draft documents in your company's style, and analyse Excel data. Best for: people already in the Microsoft ecosystem.
**Google Gemini** — Google's AI, integrated into Google Workspace (Docs, Gmail, Sheets). Similar positioning to Copilot but for the Google ecosystem. Best for: organisations running on Google Workspace.
The difference between models and products
One thing that confuses people: 'GPT-4', 'Claude 3.5', 'Gemini 1.5' — these are the underlying models. ChatGPT, Claude.ai, and Gemini.google.com are the products (web interfaces) built on top of them.
Some products let you switch between models. The same underlying model can also power completely different products built by third-party developers.
For your purposes as a non-technical professional, the product matters more than the model. Focus on which product fits into your workflow, not which model version number is biggest.
AI doesn't 'know' things. It predicts what text should come next, based on patterns in billions of documents. This sounds reductive — but it explains almost everything about what AI can and can't do.