How to Use AI in Your Job Search (Without It Backfiring)

What gives it away

I have spent about ten years in HR and recruitment, and I read applications every week. The ones written by AI are usually recognisable, though not for the reason most articles give.

The tells are mechanical rather than semantic. The structure and build: the same shape every time. Generic sentences that could sit inside anyone else's application. Language that reads slightly unnaturally, as though it has been translated. Long em-dashes. And a tidy list of three to five bullet points, which by now is a classic giveaway.

Does it actually count against you

Less than candidates fear, and differently than they expect. AI can genuinely support a good application, and used properly it strengthens a case you were already making.

It reads as wasted effort in two situations. The first is when there is no engagement behind it, when the letter arrives with nothing of the person left in it. The second is when it could have been sent to any company for any role. That second one is the same problem that makes me stop reading a CV, and it is worth being clear about: the issue is not that a machine helped. It is that nothing in the result was aimed at us.

Both ends of this have moved over the past couple of years. Some candidates have become genuinely good at using these tools. There are also far more weak, generic applications than there used to be. The ceiling has risen and the average has fallen at the same time.

Where it helps

Used well, AI tools remove friction from the mechanical parts of a search, not the parts requiring you to actually know your own experience.

• A first draft. Turning a rough list of achievements into a structured CV section or letter draft solves the blank-page problem well.
• Faster tailoring. Feeding a posting and your CV to an AI tool and asking which experiences to foreground can speed up the tailoring work covered in our application guide.
• Rehearsal. Practicing answers against an AI tool that asks realistic follow-ups is a low-stakes way to hear your own stories out loud.
• Fast research. A quick first pass on unfamiliar terminology or a company's recent direction. Treat it as a starting point to verify, not a source to cite.

Where it backfires

The failure mode is consistent: using AI output directly, unedited, in place of your own voice. Recruiters read high volumes of applications, and generic AI phrasing has become recognizable: a certain confident vagueness, achievements described in general terms with no specifics attached.

• Never let AI invent achievements, numbers, or experience you do not have. An interviewer's first follow-up exposes it immediately.
• Never submit an AI-drafted letter unedited. Structure can come from AI; substance has to come from you.
• Be sceptical of "resume optimizer" tools that promise to game applicant tracking systems through keyword stuffing. Whether those systems actively penalise it is not something I can tell you with any confidence. What is clear is that the more filler a document carries, the less weight any individual part of it can carry, so the tactic is unlikely to help you either way.

Employers use it too

This cuts both ways. A growing number of employers use AI somewhere in the pipeline: automated resume screening, chatbot-style initial interviews, AI-assisted scoring of open-ended answers. Not universal, and it varies by company and industry, but common enough to plan around.

• An automated screen is a reason to keep a CV clean and conventionally structured, though as the CV guide sets out, these systems tend to give indications rather than decide anything. The reader who sorts you is still a person working at speed.
• If told you will have a "video screening" or "initial assessment," ask directly whether it involves AI scoring. Recruiters will tell you if asked, and it changes nothing about preparation except speaking clearly into whatever format is given.
• Cognitive and personality assessments, which is what this site is built around, are frequently scored algorithmically rather than read individually. That does not change what a strong response looks like; consistency and following the instructions precisely both still matter.

Use it as a critic, not an author

Asked directly whether a candidate should use these tools, my answer is yes. The question worth more attention is what to use them for.

1. Fix the grammar and tighten the sentences. Uncontroversial, and it removes a distraction that has nothing to do with whether you can do the job.
2. Ask it to challenge you. Give it the job posting and your draft, then ask whether the application is genuinely targeted at this role and what you should be focusing on that you have not. This is the use that improves outcomes most, and it is the one almost everyone skips.
3. Keep the words your own. The moment the sentences stop being yours, the tells at the top of this guide start appearing.
4. Write your own rough draft first, even a bad one. Starting from your own words is what keeps the final version yours.
5. Use it for rehearsal, not for live answers. Leaning on it during a real interview is poor practice and, in most virtual formats now, detectable.

Prompts worth actually using

Vague prompts produce vague output. Specific ones, which hand the tool your actual material instead of asking it to invent something, produce something worth editing rather than something worth discarding.

• Turning achievements into bullets: "Here is a rough, unpolished list of what I did in this role: [paste notes]. Rewrite it as 3-4 CV bullets, each starting with an action verb and ending with a concrete result. Do not invent numbers or outcomes I have not given you."
• Tailoring to a posting: "Here is a job posting: [paste]. Here is my CV: [paste]. Which 3 of my existing bullets are most relevant to this posting, and how should I reorder my summary line to speak to it directly? Do not rewrite my experience, only reorder and reprioritize what is already true."
• Mock interview rehearsal: "Act as an interviewer for a [role] position. Ask me one behavioral question at a time, wait for my answer, then give me one specific piece of feedback on it before asking the next question."

The pattern across all three: give it your real material and a narrow task, not an open invitation to generate something from nothing. That is what keeps the output something you would actually want to submit.