Tutorials·beginner·7 min

Best AI for job seekers

The order to run a job search in, the four steps AI genuinely carries, and the one that decides your odds and has no AI in it at all.

Multi Chats Team
October 12, 2026

A job search can become disorganised in the first week, even with good applications. They go wrong because you open a job board, see forty things, and start applying. Six weeks later you have sent thirty applications, heard back from two, and cannot remember which version of your CV went to which company.

AI is genuinely useful here, and it carries four of the six steps below. The step that most changes your odds is not one of them. Here is the order to run a search in, with an honest rating at each stage.

1. Work out what you are actually looking for

Do this before you read a single ad. Ads are persuasive and they will quietly redefine what you want.

Open a chat and get interviewed about your last job instead of your next one. Ask it to question you for ten minutes on what you actually spent your time on, which parts you would happily do again, and which parts you dreaded on a Sunday night. Answer in full sentences. Then ask for the constraints that fall out of your answers: size of team, how much of the week you want to spend writing versus talking, what you will not accept again.

It works because the model is patient and has no stake in the answer. A friend gets bored at question four.

What it cannot do: tell you what you are worth. It has no view of your market, your city, or what someone with your experience gets paid this quarter. Salary questions are for people who have hired recently and for the ads themselves.

2. What the ad is really asking for

Job ads are written by committee and half of what is in them is aspiration. Paste one in and ask which requirements are load bearing, which are wish list, and what the ad conspicuously does not say.

That last one is the useful column. An ad for a senior role that never says who you report to tells you something worth walking into an interview knowing.

What it cannot do: tell you whether you are competitive. Ask a model whether you should apply and it will encourage you every time, because encouragement is the shape of its default answer. It has not seen the other applicants and cannot see the bar the hiring manager is holding. Apply anyway when you are close. Just be clear that the decision was yours and the encouragement was free.

It also does not know the company, only what was written before its training cut-off, which for a company of normal size is very little and sometimes wrong. Web search closes part of that gap, capped at two searches per 30 days on the free plan (the mobile app puts that button behind Pro), so spend them on something specific like recent funding or a leadership change rather than on "tell me about this company".

3. Write the application

Two jobs here, each its own piece of work. Tailoring your CV to the ad is a reordering exercise, moving the line that answers the ad from eleventh place to first. The cover letter is a separate document with a separate failure mode. Neither is improved by asking one model for one draft and sending it.

The rule that covers both: never send an application containing a fact you have not personally checked. Models invent numbers with total composure. A rewrite you did not read closely will hand a recruiter a percentage you never measured, and you will be asked about it in the room.

The second rule is about sameness. A recruiter is reading a stack of these in one sitting, and enough now open on the same note about being excited by the mission that the note has stopped meaning anything. Accept the first draft and you are in that pile. The way out is to run the same request on a second model, keep the strongest lines from each, then rewrite the two most important sentences yourself. Use the forking icon under the reply, which opens a new thread and leaves the first draft where it is. Do not go by the words: web calls that action Try Again, while on a phone Try again is the one that deletes your draft and replaces it.

4. Talk to somebody who works there

This is the highest-value step in the whole search and there is no AI in it.

One conversation with a person who does the job tells you what no amount of prompting will: whether the team is understaffed, what the last person in the role found hard, whether the salary band in the ad is real, and whether your experience reads as senior or as adjacent. Fifteen minutes of that changes what you write and how you interview.

The one thing a model helps with is the ask. Get it to draft a short message to a second-degree contact and cut it until it is four sentences with a specific question in it. Then send it yourself, from your own account, having read it out loud.

5. Rehearse before the first real one

Interview practice has its own guide. The short version: it works, because a model will run the drill past the point where a human helper starts being kind about it.

Watch for: the feedback skews generous by default. Ask instead for the three things a hiring manager would hold against the answer, and nothing else.

6. Weeks four to eight, and how to survive them

Nobody writes about this part, and it is what breaks people. Weeks four to eight are mostly silence, and silence is easy to read as a verdict on you.

Practical version. Start a new thread per application and give it the company name, so the sidebar becomes your log and thread search finds the one you need in a second. If you want them grouped, folders exist and the free plan holds up to 25, with one catch: on the web the button that creates your first folder currently only appears on a paid plan, while the mobile app lets any account create one. On free, make the first folder on your phone and it will be there when you open the web app.

Stop retyping who you are. The free Personal preferences field holds 5,000 characters and starts every new conversation with whatever is in it: under Account in web settings, under Personalization in your mobile profile. Target role, level, three strongest examples, written once. Memories, which it saves from a chat when you ask it to, are on the paid plans.

A search means pasting your CV and your reasons for leaving into a chat window, so be clear where that lands. We do not train on your conversations. What the provider behind a given model does varies, and only six of the 60 say so on their info card. The rest are quiet.

On staying sane: set a number of applications per week and stop at it. Thirty considered applications beat a hundred fired off, and a model will happily produce the hundred, because volume is what it is best at and what helps you least.

What the six steps add up to

AI carried steps one, two, three and five, doing the patient mechanical work: holding a long ad and a long CV in mind at once, asking the same question thirty times, producing a second version to compare against.

It did not decide what you want, and it did not tell you whether you were in the running. Step four, the fifteen minutes with a person, is the one that moves the odds, and people skip it because it is the only step that costs them something.

Sixty models from 18 providers, on web, iOS and Android. Twenty-one are open on the free plan, with 50 messages a day, which will carry a full search. Pro is $20.99 a month.