# What is an AI assistant, and which one should I use?

Level 1: AI basics · 8 min · Updated 19 November 2026 · Path: AI basics

By the end you'll know what an AI assistant is good and bad at, write a clear first request and choose one to start with.

Everyone around you seems to be using "AI" for something. A colleague drafts emails with it, your kid asks it about homework, your manager wants a summary of a 40-page report by Friday. You've maybe tried it once, got a so-so answer, and wondered what the fuss is about.

Here's the short version. An AI assistant is an app you talk to in plain language, such as ChatGPT, Claude or Gemini. Behind it is a large language model: a program trained on a huge amount of text that writes a likely, useful reply to whatever you send. That one trick turns out to cover a lot of everyday work.

## What it's good at

- **First drafts**: an email you've been putting off, an outline, a cover letter.
- **Summaries and explanations**: paste the 40-page report and ask for the five points that matter to you, or ask what a contract clause means in plain words.
- **Rewriting**: shorter, friendlier, more formal, for a different reader.
- **Thinking out loud**: pros and cons, the questions you haven't thought of, a plan for the week.
- **Files and other apps**: most can read a PDF or a spreadsheet you upload, many can search the web, and with [MCP](https://ai-powered.com/academy/glossary/mcp) some connect to your other apps.

## What it gets wrong

It can be confidently wrong. A model writes what *sounds* right; unless it searched the web and shows you its sources, it can invent a number, a name or a quote. Treat facts in an answer as a lead to check, not a fact.

It also doesn't know you, your company or your situation unless you tell it, and it may not know recent events unless it searches.

## How to ask well

Most disappointing answers come from short, vague requests. Anthropic's prompting guide puts being clear and direct first, and recommends giving context, examples and the format you want. The same habits work in every assistant.

*Diagram: What a good request contains.* A good request says what it’s for, gives your own material, shows an example, asks for a format, and is followed by feedback.

The fifth part is the one people skip: reply. The first answer is a draft. "Shorter, and mention the dinner" gets you closer in seconds.

**Try it**

Pick one real email you need to write this week. Ask for it with all four parts (goal, your material, an example, a format), then send one round of feedback. Compare it with what you would have written.

## Which one should I use?

All three of the best-known assistants have a free plan, so the honest answer is: try the same real task in two of them and keep the one whose answers you like better. As of 19 November 2026, from each provider's pricing page:

- **ChatGPT** (OpenAI): Free, Go, Plus and Pro plans for individuals, plus business plans. Prices weren't readable on the page when we checked; see the pricing page for your country.
- **Claude** (Anthropic): Free, Pro ($20 a month billed monthly, or $17 a month billed yearly), Max ($100) and Ultra ($200), plus Team and Enterprise.
- **Gemini** (Google): free with a Google account, with paid Google AI Plus, Pro and Ultra plans; prices vary by country.

What matters more than the logo:

- **Where you work.** If your company already pays for one, start there. It may come with data protections a personal account doesn't have.
- **What you'll use it for.** Writing, coding, research and images each favour different apps. The directory's [AI tools](https://ai-powered.com/tools) pages show what each one does.
- **Your data.** Read AI and your data before you paste anything private.

Popular AI tools in the directory, if you want to look beyond the big three:

Picks for ai models and llm tooling from the directory:

- [ragflow](https://ai-powered.com/tools/ragflow): Open-source RAG engine with a document pipeline, hybrid search and visual agent workflows
- [OmniRoute](https://ai-powered.com/tools/omniroute): Self-hosted AI gateway routing many providers through one local endpoint with fallback
- [cs249r_book](https://ai-powered.com/tools/cs249r-book): Open textbook and course materials on machine learning systems engineering
- [mcp-use](https://ai-powered.com/tools/mcp-use): Open-source SDK for building MCP apps and servers, with a hosted cloud for them
- [bifrost](https://ai-powered.com/tools/bifrost-3): Open-source AI gateway in Go that routes LLM calls with budgets, fallbacks and MCP

## Key takeaways

- An assistant writes likely, useful text from what you give it. Give it more and it does better.
- Check facts; it can be confidently wrong.
- Ask with a goal, your material, an example and a format, then reply with feedback.

## What to try next

- Read AI and your data before your first work task.
- Then try one workflow from your job: AI for marketing or AI for sales.
- Curious what happens when an assistant starts doing things, not just writing? That's an [AI agent](https://ai-powered.com/academy/glossary/ai-agent).

## Sources

Official documentation, checked on 19 November 2026:

- [Prompting best practices (Claude docs)](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices)
- [Claude plans and pricing (Anthropic)](https://claude.com/pricing)
- [ChatGPT pricing (OpenAI)](https://chatgpt.com/pricing)
- [Google AI plans (Google)](https://gemini.google/subscriptions/)

Next lesson: [What is MCP? The plug that lets AI use your apps](https://ai-powered.com/academy/ai-basics/what-is-mcp)
