September 15, 2026 · ~7 min read · #AI_Recruiting
Who Outsmarts Whom
How AI helped candidates tailor their profile to the job — and helped me filter them out.
Chapter 1 — Uploading Resumes
First, where I stand on AI: I'm not a fan, and not a skeptic either — a cautious enthusiast. I don't dive in headfirst ("let's just do AI interviews" — that's a rant for its own post on why AI interviews are trash), but I don't cling to the old ways either ("I've done this a thousand times, no AI is replacing me"). Delegating simple, mechanical tasks to AI is something everyone owes themselves — not out of love for AI, but out of love for themselves. What follows is exactly about those simple things…
Uploading resumes is the first thing you can and should automate. Let me walk you through my experience with HH.
Steps before (everything that could be simplified or automated — I simplified or automated):
1)Log in (say, login/password handled automatically)
2)3–4 clicks (from memory, but I stand by my memory) to reach all candidates for the vacancy
3)right-click the "…" on the candidate's card / download the resume (might need to pick a format)
Step 3 repeats once per candidate.
1)Go into the downloads folder and start opening each resume to check the name, then rename it (LASTNAME_FirstName_Position)
Step 4 repeats once per resume.
How it works now
1)Set up an AI agent in Slack
2)Give it API access to your HH account
3)Program it (plain language works fine) to pull new resumes for the vacancy every day
Looks like one fewer step, and the steps are heavier — BUT you do this once, instead of 200–300 times for steps 2 and 3, for every single vacancy.
Chapter 2 — Evaluating Candidates
Step two is evaluation. There's something everyone knows about, but candidates tend to underrate it, and recruiters tend to overrate it. HR doesn't deeply analyze experience from a resume… actually, let me correct that: HR doesn't analyze experience from a resume at all. But it also doesn't rely on some mysterious gut instinct (aka "the seasoned IT recruiter") that supposedly filters out half the best candidates. A recruiter (1) scans resumes for keywords, and (2) screens out candidates against criteria agreed in advance with the hiring manager, or with themselves — not necessarily published anywhere.
First — the criteria for entry (access pass): mainly hard skills (".NET," "Mercer evaluation," "Google Analytics") and, to a lesser extent, soft skills you can spot in a resume but will still need to verify at interview ("built a team of 4," "worked with US clients").
Second — the criteria for rejection (denial criteria): gender (let's not kid ourselves — for roles like personal assistant, employers usually want a woman), age (yes, ageism: past 40 we look closer, past 50 we filter out), and which companies show up on the resume — say, only domestic, or some international exposure too.
Since it's all this formalizable, it can be handed off to AI too. Spoiler: evaluating those same 400 resumes by hand would have meant far more errors on my part, including more critical ones — and it would have taken not 2–3 days, as it did with Claude plus checking, but 2–3 weeks.
Let me walk you through it using Claude — my current favorite. On the $20/month plan you get access to every model, including Opus, which works wonders, and the usage limit is roughly enough for daily heavy-duty work on stuff like this. So, step by step:
1)We feed resumes in batches of 10 (that's Claude's limit, and there's a real reason behind it — the more data you dump into the AI at once, the more errors you get).
2)We set a scoring scale (I used a 5-point one: 5 — a perfect match on the vacancy's keywords, 1 — nowhere close), and from there I only looked at anyone scoring 4 or above.
3)We spell out the blocking criteria separately — they're never in the job description. I really felt the urge here to say my piece about the fake fight against discrimination in our society (and honestly, probably most societies), and its hypocritical version written into law. Say none of this bothers you, and you'll happily interview a 30-year-old male candidate for a personal-assistant role, no problem. A blocking criterion can be anything, not just gender and age: no experience at certain companies (say, nothing Western), or, the opposite, a resume packed with gambling-industry experience.
4)We run the evaluation and go through the 4s and 5s.
Chapter 3 — The AI Check
As I said above, I'm not against AI — at the very least, I use it myself to evaluate candidates, so why shouldn't candidates use it to get noticed. I won't hand out extra points for that — over the last year or two it's become table stakes for every office worker — but I won't dock points for it either. What's always thrown me, though, with some candidates, is writing things in a resume they never actually did. It comes out eventually anyway (either at the interview, or worse for everyone, during the probation period), so the obvious question is: WHY? Though I get it, on a human level — you take any shot you can get, get a foot in the door, and figure out the rest later. And now HH, on some premium plan, is straight up offering to whip up a cover letter tailored to the vacancy for you! Half, if not 90%, of people don't even read what HH's AI writes for them: generate — send.
After tweaking the instructions for Claude — bluntly, because there's nothing to nuance here: the cover letter gets cut from the evaluation entirely — everything went smoothly from there.
The obvious next move on the candidates' side is customizing the resume itself to match the vacancy's keywords. On the HR side, it's filtering those out, or at least flagging them. That's just how we live now.
How to be like Oleg:
–Automate resume downloading and naming (LASTNAME_FirstName_Position) — it's pure mechanics.
–Split your criteria up front into access pass (hard + soft skills) and denial criteria (blocks).
–Set a scoring scale and a cutoff: review everything from the top score (your ideal candidate) down to one point below.
–Double-check the AI — aim for at least 2–3 clean runs before you're confident the model isn't making mistakes, or that you no longer need to adjust the instructions.
–Ignore the cover letter — it's increasingly AI-written and doesn't reflect real experience.
–Going forward, be ready to deal with customized resumes the same way you've had to deal with AI-generated cover letters.