AI-powered job hunting

Stripped of marketing, the candidate-facing toolkit falls into four buckets. They differ enormously in how much they change your outcome, and the ordering is not the one the tools themselves suggest.

The honest ranking, from most to least valuable, is preparation, then judgement, then tailoring, then volume — which is almost exactly the reverse of where most products put their emphasis.

The inversion is not a conspiracy; it follows from what is easy to build and easy to demonstrate. Sending more applications is a visible number, and rehearsing an interview well is not, so the feature that helps least is the one that markets best.

• Preparation — rehearsing answers, researching a company, pressure-testing your story
• Judgement — deciding which roles deserve your effort
• Tailoring — adapting your material to a specific posting
• Volume — sending more applications, faster

Interview preparation is where AI is most useful and least used. A model that has the job description and your background can generate the questions you will actually face, push back on a weak answer, and let you rehearse until it stops being the first time you have said something out loud.

This matters because the interview is where offers are decided, and because rehearsal is the one activity with a direct causal link to the outcome. It is also unglamorous, which is presumably why it is the least marketed.

The specific move worth adopting is adversarial rather than supportive. Give it your answer and ask what a sceptical interviewer would push on — the follow-ups it produces are usually the ones that would have caught you out, and answering them in advance is worth more than polishing the original.

Reading a posting and deciding whether it is worth applying to is slow, and most of the time the answer is no. Handing that first pass to a system that can reason about fit — rather than match keywords — recovers a large share of the total time a job search consumes.

Insist on a reason alongside any score. A number with no explanation cannot be corrected, and the point of the exercise is for the filter to get better at your particular taste rather than to be confidently wrong faster.

Check what it rejected, at least in the first week. A filter that quietly excludes a category you care about is worse than no filter at all, and it is invisible unless you deliberately look at the discard pile once or twice.

Adapting a CV to a posting is real and legitimate work: emphasising the relevant, matching the vocabulary, cutting what does not apply. AI does it quickly and well when it is restricted to rearranging things you actually did.

The limit is firm. A model asked to make a weak match look strong will invent, and an invented claim on a document with your name on it is your problem in the interview, not the tool’s. Read every tailored version before it leaves.

Keep the version you sent. Once you have tailored twenty applications, the document an interviewer is holding is one specific variant, and walking into a conversation unsure which achievements you led with is an avoidable disadvantage.

Everything above operates on a decision you have already made about what you are looking for. None of these tools can tell you whether you want a bigger company or a smaller one, whether you are optimising for money or for the work, or whether the thing making you unhappy is the job or the role.

That matters because a job search run without answering it produces a lot of activity and a strange offer. The tools make it easier than ever to move fast in a direction you have not chosen.

An hour spent writing down what you actually want — and what you would refuse — before configuring anything is the highest-leverage hour in the whole process, and the only one none of this software can do for you.

Every tool in this list needs your employment history, and several want your inbox or your job-board accounts. That is a larger transfer than it feels like, because a CV is one of the most complete personal records most people produce — and the fact that you are looking at all is confidential if you are currently employed.

The questions worth asking are ordinary ones a serious product answers quickly: what is retained, for how long, whether it is excluded from model training at every layer including sub-processors, and what deletion actually removes. Vagueness on any of them is more informative than the specific answer.

A few habits cost nothing and remove most of the exposure. Strip your home address, date of birth and any identity number before uploading — none affect an assessment of your suitability and all are what make a leaked CV worth stealing. Use a separate email address for applications, and keep the whole search off work devices and work accounts.

• What is retained, and for how long, including logs and backups
• Whether inputs are excluded from training at every layer
• What deletion removes, and by when
• Which sub-processors see your CV, by name
• Whether it will contact employers without you initiating it

Every tool in this space eventually offers to apply to more jobs. It is the easiest feature to sell and the least likely to help, because the number of applications you could send was never what was limiting you.

When applying is cheap for everyone, employers filter harder and the signals that survive are referrals, visible work and evident specificity. Spending recovered time on those beats spending it on a larger number of identical submissions.

There is also a cost people do not count: interviews you did not want. Time spent in a process for a role you would decline is time not spent on the ones you would accept, and high-volume automation produces exactly that.