Your first conversation with an AI assistant: what to type, and what comes back

The screen is almost empty: a box, a blinking cursor and a line inviting you to ask anything. That blankness is what stops people. There is no menu to choose from and no hint of what a sensible first question looks like.

What you need

A phone or a computer, and a few minutes. Four of the best-known assistants, in alphabetical order, with what each maker’s own pages say at the time of writing:

• ChatGPT (OpenAI). The help centre says it “is free to use”, with paid subscriptions alongside. You reach it at chatgpt.com or through the phone apps.
• Claude (Anthropic). The help centre says you can use the free version by going to claude.ai and signing up with an email address. There are phone and desktop apps.
• Copilot (Microsoft). Microsoft’s help centre says Copilot is “available at no cost”, online at copilot.cloud.microsoft or through its phone and desktop apps. You can try it without signing in; signing in adds chat history and longer conversations.
• Gemini (Google). Google’s help page sends you to gemini.google.com. Some features work without signing in; the rest need a Google Account.

They do broadly the same job. Type the address yourself rather than following a link in an email or an advert.

It is a conversation, not a search

A search engine takes a few keywords and hands you a list of pages to read. An assistant takes whole sentences and writes you an answer.

The larger difference is that you can reply. Tell it the answer was too long, that it misunderstood, that you want a second attempt in a friendlier tone. Anthropic’s getting-started page puts it simply: speak to it “like you would a coworker or friend - naturally and conversationally”.

How to ask well

Think of briefing a capable stranger who knows nothing about you. Three things do most of the work.

Say who you are and what it is for. “I am 67 and writing to my energy supplier” produces a different answer from a bare “write a letter”.

Give the detail. Dates, places, what you have tried, what you cannot do. OpenAI’s prompting advice asks for prompts that are “clear, specific, and provide enough context”.

Say what form you want. A short list, a letter under 150 words, a table, three options.

Then answer back. OpenAI’s guidance describes exactly this loop: “Start with an initial prompt, review the response, and refine the prompt based on the output.” Useful replies include “shorter”, “explain that as if I had never heard of it” and “what did you leave out?”

Five first requests

Type these as they are, swapping in your own details.

I have written a letter of complaint to a furniture company about a sofa that arrived damaged six weeks ago. It sounds stiff and angry. Rewrite it so it is firm, polite and under 200 words, and keep the request for a refund. Here is my draft: [paste it in]

Here is a paragraph from my phone contract, with my name and account number taken out. Explain in plain English what it commits me to, and list anything I should ask the company about: [paste it in]

Plan three days in Lisbon in April for two people in their sixties. One of us cannot manage steep hills for long. We like food markets and small museums, and dislike queues. Give me a morning, afternoon and evening for each day.

I am learning Italian and I am a beginner. Be a patient conversation partner. Talk to me in simple Italian about ordering in a café, one sentence at a time, and correct my mistakes in English after each reply.

Below are my rambling notes before a meeting with a builder about a kitchen extension. Turn them into a numbered list of questions to ask, most important first: [paste them in]

The second one is an aid to reading, not a ruling. For anything with money or legal weight attached, treat the explanation as general information and take the document itself to someone qualified.

What to expect, and what not to

It will be fluent and quick. It will not always be right. OpenAI’s own help page says ChatGPT “will occasionally make up facts”, and Google’s privacy page says assistants like Gemini “can hallucinate and present inaccurate information as factual”. Treat every assistant the same way. A wrong answer arrives in the same confident voice as a right one, so anything that matters needs checking: When AI gets it wrong: how to check shows how.

Keep private things out. No passwords, no bank details, nothing about other people that they would not want shared. Google’s privacy page for Gemini is blunt, asking users not to enter “confidential information that you wouldn’t want a reviewer to see or Google to use to improve our services”.

You cannot break it

There is no wrong question, and nobody is marking your spelling or your typing.

If a conversation goes nowhere, start a new one. You can also remove one: OpenAI’s data controls page, for example, tells you to delete a chat when you want its content gone.

Learning with other people

• US: Senior Planet, from AARP’s charitable affiliate OATS, teaches technology classes. AARP says its interactive online classes “are free to anyone of any age”.
• UK: Age UK says many of its local branches offer computer training explained “in plain English”. For more information, the page gives the Age UK Advice Line: 0800 678 1602.
• Libraries: Gov.uk notes that local libraries in England and Wales offer free internet access and learning materials, and has a postcode search to find yours.

We could not confirm from those pages which of these currently teach AI specifically, so ask.

This week

Choose one assistant and one of the five requests. Type it, read the answer, and reply at least twice before you judge it. Then try a question of your own, on a subject you know well enough to spot a mistake.

For the wider picture, including privacy, scams and health questions, start with AI after 50: the complete guide.

Updated 2 November 2026: Copilot’s address and free access, from Microsoft’s current help page.