Your first hour with AI, for people who have never used it
One task, one hour, no jargon: how to get a usable answer out of an AI chat app, and how to spot the moments when it is confidently wrong.
Three different people have told you to try this by now, and not one of them explained what you are supposed to do after it opens. It can be hard to know where to start. The screen is a box with a cursor blinking in it, everyone online has an opinion about what goes in the box, and none of that helps at eight in the evening when you have twenty minutes and a thing you need to get done.
So here is one hour, spent on one thing.
Say the shower in your flat has been broken for three weeks. You reported it twice, the letting agent answered once and then went quiet, and you have been putting off writing a proper letter because you cannot decide how firm to be and you do not want to lose an evening to it. That is the task for the next hour. Everything below happens to that letter.
Type it the way you would say it out loud
Open the app and type the sentence you would say to a friend. "I need to write to my letting agent about a shower that has been broken for three weeks and two ignored messages." Press enter.
Something starts writing back almost immediately, at roughly reading speed, and a few seconds later you have a complete letter with an address block, three paragraphs and a sign-off. No special wording was required to get that, and there is no command language to learn first. The message you typed is the whole interface. People call that message a prompt, which is a grand word for a thing you already know how to do.
Two things are worth noticing about what came back, and they pull in opposite directions. It is genuinely usable, better than the blank page you were staring at. And it is finished, fluent and sure of itself, which has nothing at all to do with whether it is correct.
It will be wrong sometimes, and it sounds exactly the same when it is
Read the letter it produced and you find a sentence like this sitting in the second paragraph, in the same even tone as the address block above it:
Under the relevant housing regulations, your landlord is required to complete repairs of this kind within 14 days of written notice, and you are entitled to compensation for every day beyond that deadline.
Treat that sentence as invented, because for the purposes of this article it is. There may be a fourteen day rule and a right to compensation where you live, or that may be a confident average of a few hundred articles written about a dozen different countries, none of them yours. This article is not telling you which, and neither did the letter. Nothing in the answer marks the difference. It does not slow down, hedge, or shift tone between the parts it has good grounds for and the parts it has assembled out of plausible pieces. This is the single hardest habit for a new user to build, because everything else you have used in your life sounds uncertain when it is uncertain.
There is a reliable tell, though, and it takes about a week to become automatic. Watch for the places where it gets specific in a way you did not ask for. You asked for a letter and received a number, a deadline, a named legal entitlement. That is the moment to stop. Numbers, dates, prices, names of people, medical specifics, anything that looks like a citation: those are the parts to check somewhere else before you rely on them, and a government or official page is where you check the ones in this letter.
What does not work is asking it whether it is sure. It will apologise very graciously and then give you a different confident number. Checking has to happen outside the conversation, which is annoying, and is also the entire price of admission.
Detail does the work
Most people's second instinct is to be nicer to it. Please, thank you, would you mind. That is fine, and it changes almost nothing about what comes back. The thing that changes the answer is how much of your own situation you are willing to type.
So try again, and this time hand over the facts only you have. You reported it on the 4th and again on the 18th. There is a toddler in the flat. You want the letter firm without being hostile. You want it under 200 words, because a long letter reads as a complaint and a short one reads as a deadline. You want one specific outcome, which is a repair date in writing within five working days. And you do not know the housing law where you live and do not want any legal claims in the letter at all.
That last instruction is the one that quietly removes the invented statute, and it is the shape of most good instructions: telling it what to leave out. The letter that comes back this time is about your actual situation, which the first one could not have been, because the first one only knew what a broken shower is in general. Everything that improved came from ordinary detail you already had in your head and had not thought to type.
What the first answer is for
Now the part that decides whether you ever use this again.
The second letter is close but still too long, and the closing line goes soft right when it should not. So you reply to it, in the same box, the way you would reply to a colleague who sent you a draft: cut it to 150 words, keep both dates, make the last line firmer, and delete the sentence about being a valued tenant. It rewrites in seconds. You read it again and reply again: the second paragraph now sounds sarcastic, take the edge off it. It does.
That loop is the difference between the people who find this genuinely useful and the people who tried it once and concluded it was overrated. Almost everyone who bounces off it typed one thing, read one answer, judged the answer and closed the tab. The first answer is a first draft from someone fast, eager and only half briefed. Nobody's first draft survives contact with the person who asked for it.
The same loop works when you are not asking it to write anything. Halfway through the letter you can ask what a letting agent's usual excuse for a three-week delay is and how to close that door in one line. You are still on the same task, and you get an answer worth arguing with, which you then argue with.
You can do all of this on the free plan
On MultiChats you can start typing without making an account at all. That gets you ten messages, and they do not refill, so a first-timer tends to burn them on experiments and then hit a wall mid-task. Signing in first is worth the thirty seconds: a free account gets 50 messages every 24 hours, which is several letters, several arguments about the letters, and change. It runs in a browser and as an iPhone and Android app on the same account, so the letter you started on the sofa is on your laptop. Paid plans are listed on the home page, and nothing in this first hour needs one.
There is one more thing behind the box that you should ignore today and will want in a fortnight: the app carries 60 models from 18 providers, which are different systems built by different companies, each with its own strengths. On your first day none of that matters. The app opens on one, and it will write your letter.
The suspicion you arrived with deserves a straight answer too. We do not train on your conversations. Some of the companies whose models we carry may do things with the text on their end, and where we know of a difference, that model's info card carries a notice, including one that states plainly that its maker may use your conversations to train its models. The practical version for your first hour: do not paste anything you would not put in an email to a stranger, which rules out passwords, card numbers and other people's medical details, and does not rule out a letter about your shower.
A verdict after one hour
The letter is sent. It took about twenty minutes rather than the evening you had budgeted, and the twenty minutes included reading a sentence about housing law carefully enough to delete it.
That is a fair picture of what this is worth. It will take the tedious part of a great many small jobs off your hands, and every so often it will hand you something invented, delivered in exactly the same even tone as everything it got right. Both of those are true at once, and the second one does not cancel the first. The people who eventually get burned are almost never the sceptics who came in like you did. They are the ones who were impressed on day one and never noticed that the writing does not change when the facts do. Keep checking even when the first result impresses you.