AI for Recruiters: what actually works in 2026

What AI does well

Boolean search generation. LinkedIn Recruiter AI turns a plain-language role brief into a working Boolean string in seconds. I timed this against a senior recruiter building the same string by hand: the AI version took 8 seconds, the manual version took 4 minutes. The catch: it over-includes adjacent titles, so you still trim the result list yourself.

First-draft outreach messages. ChatGPT and Claude both write a decent first InMail draft from a job description and a candidate’s headline. Response rates on AI-drafted openers matched hand-written ones in my own A/B test across 40 sends. The caveat: unedited drafts read generic, and candidates notice the pattern after the third templated line.

Job description rewrites. Claude is stronger than ChatGPT at stripping biased language and jargon from a job post without changing the actual requirements. It also holds tone consistency across a batch of 10+ postings, which matters when a hiring manager reviews them together.

Interview note summarization. HireVue’s AI compresses a 30-minute structured interview into a five-bullet summary a hiring manager can read in under a minute. This saves real time in a high-volume funnel, but the summary drops nuance, so a borderline candidate can read worse on paper than they sounded live.

Where it fails

HireVue scoring without a visible reason. I asked HireVue’s assessment tool to score two candidates with near-identical answers to the same competency question. One scored a 7, the other a 5, with no explanation of the gap in the output. The recruiter using the tool had to re-watch both clips manually to justify the ranking to the hiring manager. That defeats the time savings the tool is sold on.

No memory across a candidate’s full journey. Neither ChatGPT nor Claude remembers what you told it about a candidate in last week’s session. Every follow-up message, screening note, or scheduling email starts from a blank context window, so the recruiter re-pastes the same candidate history over and over across a multi-week req.

Screening summaries that miss disqualifiers. In one live req, an AI resume screen flagged a candidate as “strong match” while missing an explicit work-authorization mismatch stated in the resume’s second line. The recruiter caught it only because of a manual final pass. A team that skips that pass would have scheduled an interview that couldn’t lead to an offer.

My take

My take (August 2026, Bernat Sampera)

Recruiting AI earns its keep on the repetitive front end: search strings, first-draft messages, note cleanup. It does not earn trust on judgment calls. HireVue’s scoring and resume-screen tools produce a number or a label without the reasoning behind it, which means every recruiter I’ve watched use these tools still does the manual check that the tool was supposed to remove. That’s not a small gap. It’s the difference between a tool that saves time and a tool that saves time until it costs you a discrimination complaint or a missed disqualifier. If you compare model choice for the writing tasks, ChatGPT vs Claude is worth reading before you standardize a team on one. My verdict holds until a vendor ships a screening tool that shows its reasoning, not just its score.

What I would build

The unsolved problem in recruiting AI is candidate memory across a hiring funnel that can run six to ten weeks. A single candidate touches a sourcing message, a screen, two interview rounds, and an offer negotiation, each written by a different tool or a different recruiter, none of which remembers what came before. A real fix would carry candidate context (what’s been asked, what’s been promised, what’s been flagged) as a persistent thread attached to the requisition, not the chat session, so a recruiter picking up a candidate mid-funnel doesn’t start from zero and doesn’t repeat a question the candidate already answered.