How to use AI for your job search in 2026

How to use AI for your job search in 2026

I've run a lot of 1:1 sessions lately where the brief is the same: "I'm job hunting. Help me use AI properly." Not "tell me about AI." Not "show me ChatGPT." Actually help me land a role.

So we build a stack together. Five pieces. By the end of the hour, the person has a working system that does most of the grunt work of a job search for them. The bit that used to eat entire weekends.

This is what we build. If you're applying for jobs right now and you're still copy-pasting your CV into ChatGPT and asking it to "make it better", you're leaving the best tools on the table.

Build a brief — let the AI interview you

Most people open ChatGPT, paste their CV, and ask it to help with their job search. That's backwards.

Before any of the tools further down the stack work properly, you need a brief. A specification. Who you are, what you do, what you want next, what you'd never do. Every other step in the stack feeds off this: the agent that searches for roles, the Gem that writes your cover letters, all of it. Garbage in, garbage out.

Here's the shift most people miss: you don't write the brief. You let the AI interview you for it.

Open ChatGPT in voice mode. Then go for a walk. Twenty minutes of talking gets you a richer brief than two hours of typing — you don't censor yourself out loud, and the model can chase the things that sound interesting when you say them. If voice mode isn't your thing, do it as a normal chat at your desk; the prompt is the same.

I'm planning my next job search. Interview me so we can build a brief I can reuse across other AI tools.

Ask me one question at a time, conversationally. Don't lecture. Listen, follow up where I'm vague, move on when I've answered. Cover:

- Who I am: background, current situation, why I'm moving on
- What I do: real skills, strengths, what I'm known for, what people come to me for
- What I want next: role title, seniority, salary band, location or remote, industry, must-haves, deal-breakers
- Companies I'd love to work for, and ones I'd never work for
- Anything else that would help another AI agent search the job boards effectively on my behalf

When you've got enough, write the brief as a clean markdown document I can save and paste into other tools. Use sensible headings. No fluff.

To give you a head start, here's my CV:

[paste your full CV here]

The CV in the thread does a lot of heavy lifting. The model has something concrete to push back on ("you led that £4m programme — is that the work you want to keep doing, or what you're trying to leave behind?") and you end up with a brief that sounds like you, not a job-board template.

Save the output as job-search-brief.md somewhere you'll find it again. This is the artefact every other step in the stack reads from.

Hand the search to a browser agent

This is where it stops being a chat and starts being a tool.

Most AI models can talk about jobs. A few can actually go and find them. I'm talking about browser agents — AI that drives a real browser on your machine. ChatGPT Atlas, ChatGPT's Agent mode, and Claude for Chrome will all do it.

Watch one for the first time and it's quite something. The agent opens a tab, clicks the LinkedIn search bar, types your query, dismisses the cookie banner, scrolls the results, opens the listings that look promising, reads each page, jumps over to Indeed, runs the same dance there, and comes back with a shortlist. It's not generating fake results. It's actually using the websites the way you would (pointing, clicking, navigating, dealing with cookie pop-ups), just faster and without complaining.

In a recent session we did the following. Gave the agent the brief from step one, then followed it with this:

Using my job search brief (above), open LinkedIn and Indeed and search for roles posted in the last 14 days that match it.

Return a list of 15 roles with:
- Job title and company
- Location and remote policy
- Salary if listed
- One line on why it fits the brief
- Direct link to the posting

Sort by best fit first. Skip anything older than 14 days or that doesn't match my must-haves. If you can't find 15, tell me what you tried and where you got stuck.

Then we hit go and almost made a cup of tea — but didn't, because you don't leave a browser agent running unattended on the open web. (More on why in a moment.) We sat next to it and watched it work.

Ten minutes later, we had a formatted shortlist. Not perfect: some matches were loose, and a few postings were three months old. But it had surfaced four roles the person hadn't seen in their own manual search. One of those felt strong enough to chase the same evening.

The shift here is that you stop being the search engine. You write the brief once, then you get a formatted document you can skim. You can do this every morning before coffee.

Why you supervise it. Browser agents are vulnerable to prompt injection. A page on the open web can contain hidden instructions: text in the HTML the model reads as commands, sometimes invisibly styled, sometimes buried in a comments section, sometimes just a heading written to manipulate it. Modern agents will, occasionally, follow them. The risk isn't logins or cookie banners; those they handle fine. The risk is the agent doing something on your behalf you never asked for: messaging a recruiter in your voice, downloading a file, clicking a link it shouldn't. (Worth half an hour reading up on prompt injection attacks before you let one drive your browser.)

Treat it like a junior assistant you trust with a defined task, not an autonomous worker you let loose on your machine. Sit nearby. Watch the tab. Especially the first few times.

Audit your LinkedIn and Indeed profile the way a recruiter sees it

The other job of a browser agent is to look at you.

Recruiters and in-house talent teams use LinkedIn and Indeed search in very specific ways. They filter by title, by keyword, by location, by years of experience. If your profile doesn't match the filters they're running, you don't exist to them. It doesn't matter how good your CV is — they never get that far.

So in the session we run a profile audit. Here's the prompt:

Open my LinkedIn profile at [URL].

Pretend you're a recruiter searching for [target role title] in [location]. You've got 100 profiles to get through today.

Answer these:
1. Would I come up in the searches you'd typically run for this role?
2. What's your first impression from the headline and the first 300 words?
3. What reads like filler?
4. What's missing that a recruiter would expect to see for this role?
5. Give me a prioritised list of edits that would take under 2 hours total.

Be blunt. Don't soften the feedback.

Same prompt, Indeed URL, run it again. The two platforms rank profiles differently, so the feedback diverges in useful ways.

Claude for Chrome is good for this because it genuinely reads the page the way a human would. It sees the layout, the order of things, the whitespace. Then it writes you a critique.

The critiques are often brutal in a useful way. "Your headline says 'passionate professional'. That's invisible to search. Your most recent role uses three lines of jargon before naming what you actually did. Your skills section has 50 tags, most of which dilute the strong ones." That kind of thing.

You end up with a checklist of profile edits. None of them take more than 20 minutes to make. All of them compound.

Build a CV and cover letter workflow with Gems or Projects

This is the part that usually saves people the most time.

The old way: you see a job ad, you open your CV, you read the ad, you try to remember which bits of your CV to emphasise, you rewrite the profile summary, you half-write a cover letter, you second-guess yourself, you move on. The whole thing takes 45 minutes per application. You apply to five a week and you're exhausted.

The new way is a workflow that sits inside Gemini Gems, ChatGPT Projects, or Claude Projects. Pick one. They all work. I usually use Gems with clients because it's free with a Google account, and Gemini is strong at instruction-following.

Quick aside on what these are, in case you've never used one. A Gem or a Project is a saved bundle of context (a system prompt and a set of reference files) that travels across every chat you start inside it. Normal chat threads are amnesiac: when you close one, everything you taught the model is gone. A Gem or Project is the opposite. You set it up once with your master CV, your voice guide, and your instructions, and every new thread inside it already knows all of that. You just paste the job description and go. That's the whole reason this part of the stack works: you stop re-explaining yourself every time.

Here's the system prompt I use inside the Gem or Project:

You are a CV and cover letter writer for [your name].

I'll paste you a job description. You return two things:

1. A tailored one-page CV, based on the master CV below, with bullets rewritten to emphasise the experience most relevant to this specific role.
2. A tailored cover letter, max 300 words, in the voice described in the voice guide below.

Rules:
- Never invent experience I don't have. If you would normally invent a detail (a metric, a company, a project) to fill a gap, flag it as [ADD DETAIL] instead.
- No corporate buzzwords. Follow the voice guide strictly.
- Make the cover letter specific to the company and role. If the role mentions a product, reference it. If the company has a public mission, reference it. Generic filler fails this brief.
- Don't open the cover letter with "I'm excited to apply". Ever.

Master CV:
[paste full master CV here]

Voice guide:
[paste voice guide here, see step 5]

Example cover letters I'm happy with:
[paste 2 to 3 examples here]

That "flag instead of invent" line matters. LLMs will hallucinate achievements if you don't stop them. Telling them to flag instead of invent is the difference between a CV you can send and a CV you'd be embarrassed by.

Once the Gem is set up, the workflow is: paste job description, get tailored CV and cover letter, edit lightly, send. The 45 minutes becomes five.

Train the AI to write in your voice

This is the last piece. It's also the one most people skip — and it's the one that decides whether your applications sound like you or like the other 200 in the pile.

Without a voice guide, everything an AI writes for you sounds like AI. You can feel it. Hiring managers can feel it. They read a hundred of these a week and they know the pattern: tidy three-clause sentences, every paragraph wrapped up with a neat summary, "in today's fast-paced world," "delve into," "it's important to note that," vague hedging, sudden bursts of corporate vocabulary that don't match the rest of the writing. Wikipedia maintains a whole article cataloguing the tells. Worth a skim before you build the voice guide, because half the job is teaching the AI what not to do.

So before we wire up the workflow in step four, we build a voice guide. Not a corporate one. A personal one that describes how you actually write — and lists the AI tics that should be stripped out of every draft.

Take three or four pieces of writing you already have. An old cover letter you liked. A LinkedIn post that got a reaction. A good email to a boss. A paragraph from a personal essay. Then run this prompt:

I'm going to paste 3 to 4 pieces of my own writing below. Analyse them and produce a structured voice guide I can give to another AI so it writes like me — not like AI pretending to be me.

Structure the guide in six sections:

1. Core principles
The 3 to 5 high-level rules that govern everything I write. One line each, with a one-line explanation. Examples: "Concrete over abstract." "Confidence without hedging." "Short sentences for emphasis after long analytical ones."

2. Hallmarks
The distinctive structural moves I make. How I open a piece, how I escalate, whether I use analogies, how I handle transitions. The shape of a typical paragraph. Whether I use bullets or only prose. My rhythm: short and punchy, long and analytical, or a mix.

3. Sentence-level patterns
Average sentence length, em-dash use, contractions, first or third person, how I open sentences, how I handle parenthetical asides, anything else distinctive at the line level.

4. What to avoid
The things I never do, pulled from what's missing in my samples. Be specific. "Listicles." "Marketing CTAs in the body." "Philosophical tangents without a payoff."

5. Banned phrases
A concrete list of words and phrases that should never appear in anything written for me. Build it from two sources:
(a) Corporate-speak that's the opposite of how I write: "leverage," "synergies," "unlock value," "best-in-class," "cutting-edge," "robust," "seamless," "world-class," "thought leader," "digital transformation," "empower."
(b) Common AI-writing tells: "delve into," "it's important to note that," "in today's fast-paced world," "navigating the landscape of," "tapestry," "ever-evolving," "stands as a testament," excessive em-dashes, three-clause parallelism, summary sentences at the end of every paragraph, hedging without commitment, the rule of three applied mechanically. Use the Wikipedia article "Signs of AI writing" (https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) as a reference and pull anything from there that's the opposite of my voice.

6. Quality checklist
Five yes/no questions a writer (human or AI) can use to verify a draft is in my voice before sending. Examples: "Would I actually say this out loud?" "Does it pass the 'this could appear on 10,000 other websites' test?" "Are there any banned phrases?" "Are there summary sentences I'd cut?" "Does it open the way I'd open it?"

Don't flatter me. Be specific. If my writing has a quirk, name it. If a sample has a weakness, point at it.

Writing samples:
[paste 3 to 4 pieces here]

The output is usually a one-pager: punchy rules, real examples, a banned list you can run any draft against. That doc goes into the Gem from step four, and into anything else you ask AI to write.

The best bit: the voice guide improves every other piece of the stack. Paste it into step one when you're being interviewed for the brief. Hand it to Claude for Chrome when you're auditing your profile. Use it when you reply to recruiter messages. One artefact, used everywhere.

And in a job search where half the applications are written by an AI that didn't get a voice guide, sounding like a human is a real edge.

Putting it together

The full stack looks like this:

1. A job search brief you built by being interviewed by ChatGPT, saved as markdown
2. A browser agent that brings you a shortlist every morning
3. A profile audit from that same browser agent, run every couple of weeks
4. A Gem or Project that tailors your CV and cover letter to any job description in five minutes
5. A voice guide that trains every AI you use to sound like you — and lists the AI tells they should never produce

Set up properly, this takes one 60-minute session. After that, you run it yourself, every day, at about 10% of the effort a manual job search takes. None of it replaces the human bit: the conversations, the follow-ups, the networking. It just frees up the time to do that bit well.

Want the stack built with you?

The job search system is one of the five fixed tasks in the Power Hour. £199, 60 minutes, on Google Meet. We set it up together using your real CV, your real target roles, your voice — and you leave with it running on your machine. That's the right call if you've read this post and you want this exact stack.

If you don't know yet what you want AI to do for you (career, business, side project, no clear brief), that's a different conversation. The 90-minute Deep-Dive at £699 is the session where we figure that out together. Different product, different purpose. Don't book it just to get more time on the job search stack. The Power Hour already does the whole stack.

Not sure which one fits? Message me. I'll tell you honestly.

If you're weighing up a course instead, I wrote about the difference. Short version: courses teach theory. This builds your actual system.