Using AI in a job search that already has a brain

In the last two pieces I wrote about professional resilience as owning a structure for your thinking and about running my job search like a product system.

This one is about the bit everyone wants to skip straight to: where AI fits in.

The short version is that AI has helped my search a lot – but only because the underlying structure was already there. Without that, it's just another way to generate noise faster. It is not a thinking machine I hand decisions to. It is a tool that helps me collect, reshape, and inspect information so I can think more clearly myself.

What follows isn't a guide to prompts or a list of "top 10 ways to use AI". It's simply how I've been using it day-to-day, wired into the same structured setup I described previously (in my case, built in DEVONthink, but the principles don't depend on that).

Clear jobs for a blunt tool

AI is fast, tireless, and indifferent to boredom. It will happily produce pages of text on demand. It will also get things wrong, gloss over nuance, and sound strangely certain about guesses.

So I give it work that benefits from speed and pattern-spotting, but still needs my judgement on top:

• summarising long job descriptions
• highlighting patterns and gaps in my CV versus a role
• drafting first passes of application answers
• turning rough notes into something I can actually read later

In other words, I use AI to prepare the material for thinking, not to do the thinking. I do not ask it to choose which roles I go for, how I present myself, or what I'm "really" good at. Those are my calls. AI just helps me see the material more clearly and faster.

Starting from the structure, not the prompt

Because everything in my job search lives inside one system, I rarely paste raw job links into an AI chat and say "What do you think?"

Instead, I pull from the structure I already have:

• the job description as clean text
• my CV variant for that type of role
• any notes I've captured about the company, product, or people

Then I give AI explicit roles and constraints, along the lines of:

Because the input is already organised, the output is more useful. I'm not asking AI to discover who I am. I'm asking it to compare two concrete documents and talk about the differences, so that I can interpret what that means for my search.

Analysing roles: signal from the noise

A typical flow for a new role now looks like this:

1. Capture the job description into the system, tagged and grouped like any other role.
2. Drop the job text and the relevant CV into an AI chat.
3. Ask for a structured comparison: clear matches, partial matches, gaps, questions.

The useful part is not that AI "scores" me or tells me yes/no. It's that it forces the job into a clearer shape I can think about.

For example, it might surface that:

• the role talks a lot about stakeholder management but only glances at hands-on technical work
• they say "strategy" but most responsibilities read like delivery management
• their definition of "hybrid" in practice means three days a week in a city I'd be commuting to at rush hour

None of this is mystical insight. A careful human read would spot it. But when you're scanning a dozen roles a day, help with pattern-spotting is welcome.

Crucially, the judgement stays with me. AI can highlight that most of the clear matches are in areas I'm trying to move away from. Only I can decide whether that makes this a "no", a compromise, or something worth pursuing anyway.

Drafting applications without outsourcing my voice

The other obvious place AI helps is drafting.

My rule of thumb is that AI can produce a structured first draft, but the final wording has to sound like me. If I read it out loud and feel like I'm auditioning for a corporate training video, something has gone wrong.

A typical pattern:

• I give AI the job description, my CV variant, and any previous answers I've written for similar roles.
• I ask it to outline the key points an application should cover: relevant experience, results, context.
• I then ask for a draft answer in a specific style: short paragraphs, plain language, no buzzwords, British English.

The draft is never the final version. I treat it as raw material. I'll cut, rearrange, and rewrite until it sounds like something I would actually say to a human in a room.

Again, the value is in the shape of the answer: having the key points in one place, in some kind of order. AI helps me collate and arrange the information; the actual message – what I'm saying about myself – is still mine.

Keeping the system as the source of truth

One of the dangers of using external AI tools is that they can become a parallel universe. You have a neat setup, but all the "real" thinking ends up buried in chat histories.

To avoid that, I treat the AI layer as temporary. Anything useful it produces gets pulled back into the system I control:

• role comparisons saved as notes inside the role's space
• strong application paragraphs copied into templates or CV variants
• good questions for interviews added to a common list

If it isn't in the system, it doesn't exist. That sounds strict, but it prevents the quiet drift where your work gets scattered across tabs, apps, and chats again – exactly what the structure was meant to avoid. The thinking lives where I can see it, revisit it, and improve it; AI is just one of the ways material gets in there.

Guardrails: privacy, hallucinations, and laziness

There are a few guardrails I try to keep in mind.

First, privacy. I assume that anything I paste into a hosted AI service is leaving my machine. That doesn't mean "never use it", but it does mean I avoid including details I'd be unhappy to see leak. I anonymise names, strip out exact contact details, and keep anything genuinely sensitive out of the prompt.

Second, hallucinations. AI will invent details, especially about companies or industries, with a straight face. I do not use it as a research tool for factual claims unless I can quickly cross-check elsewhere. When it comes to my own career history, I already know the facts – I only need help with framing them.

Third, laziness. This is where, in my view, a lot of people go wrong. It is very easy to let AI do the thinking for you: accept the first plausible answer, paste it into an application, and call it done. Recruiters and hiring managers see the same generic phrasing again and again. If I catch myself accepting an answer wholesale because it "sounds smart", that's usually a sign I need to slow down, delete more than I keep, and write instead of just editing.

Tools, not talent

Underneath all of this is a simple belief: AI does not replace skilled people, it amplifies them.

If you already know how to evaluate roles, frame experience, and run a search, AI can help you do those things faster and with more breadth. If you don't, it won't invent that judgement for you. At best it will imitate other people's phrasing; at worst it will give you confident nonsense.

So I treat AI as an amplifier for existing skills, not as a shortcut to having them. The value comes when there is already a clear way of working, a sense of what "good" looks like, and a human who is willing to think. The machine helps collect, sort, and reshape information so that thinking is easier, not unnecessary.

Tools are not strategy

This loops back to the same idea as the first piece: don't let your ability to think clearly live entirely inside someone else's systems.

The structure gives me something I own. The product-style flow gives me a way to move work through that structure. AI now sits on top as an accelerator, not a replacement.

If everything disappeared tomorrow – the AI tools, the scripts, even the specific app I use to hold it all – I would still have a clearer sense of:

• how I evaluate roles
• how I explain my experience
• how I want to run my own search

The tools just help me do that work with a bit more speed and a bit less noise.

That, to me, is the real win: not "AI-powered job search", but a search where the thinking is still mine, and the machines are just helping me collect and arrange the pieces.