An assistant is easiest explained in a vendor's own words

Start here

Explaining an assistant to someone who has never used one goes better with the vendors' own plain-language explainers than with jargon borrowed from a technical paper. All three describe roughly the same thing in different words: a program that generates text in reply to a prompt, good at some tasks, prone to specific kinds of mistakes, and worth checking rather than trusting outright.

What the documents say

OpenAI's help center FAQ, titled plainly 'What is ChatGPT,' explains how it was built and trained, and separately cautions that its outputs 'may be inaccurate, untruthful, and otherwise misleading at times,' recommending a reader flag wrong answers with a feedback button rather than assume correctness. Google's Gemini Apps help page frames the product around a task rather than a mechanism, describing it as a way to 'supercharge your ideas' for outlines, summaries and brainstorming, while stating plainly that 'Gemini Apps can make mistakes' and advising a person to 'double-check responses' rather than rely on them for professional advice. Anthropic's own consumer page describes Claude as 'an artificial intelligence, trained by Anthropic...to be safe, accurate, and secure — the trusted assistant for you to do your best work,' listing example tasks such as processing large amounts of information, brainstorming and simplifying routine work. None of the three uses more than a sentence or two of technical language; each pairs a description of the tool with a caution about its limits in the same breath.

Check this

This is an editorial guide for a first explanation. Describe the tool by what it does — generates text or an image in reply to a written request — rather than what it is in a deep sense, since even the vendors avoid that question. Suggest a first task that is easy to check, such as summarizing a short document the person already has, since a summary can be compared against the original directly. Then show the check itself: point out one place the summary added or dropped something, using the vendors' own cautions about mistakes as the reason to look.

What holds and what fails

This approach holds for a first, low-stakes task where the output is easy to verify against something the person already knows. It fails as an introduction to harder uses — coding, research synthesis, anything the person cannot independently check — where the same caution applies but is harder to demonstrate in a first sitting. This is an editorial synthesis of three vendor explainers, not a claim that any explanation replaces hands-on use.

• Pick one easy, checkable first task, like summarizing a short document.
• Show the comparison between the summary and the original side by side.
• Repeat the vendor's own caution about mistakes as part of the introduction, not an afterthought.

The simplest accurate description available is usually the one a vendor already wrote for its own newcomers.