AI recruiting software in 2026 is in an awkward middle. The category grew fast, the demos are slick, and a real chunk of what gets sold either does not work as shown or carries risk the buyer does not see until later. The honest state of play is that a few jobs are done well and a few are quietly dangerous.
This note is the candid read for a small agency or in-house team deciding what to buy, drawn from the testing in our recruiting reports.
What the tools do well
Sourcing and matching have genuinely improved. Given a role, the better tools surface candidates from a database faster and with more relevance than keyword search managed, and they keep learning from which profiles a recruiter actually contacts.
Coordination is the other strong area. Scheduling, reminders, and status updates run reliably and stop candidates falling into silence, which is where small teams lose people. This is unglamorous and it is where most of the safe time saving lives.
Drafting helps too. Job descriptions, outreach messages, interview question sets. The recruiter edits every one, but the blank page is gone, and for a small team that is real hours back.
Where the risk lives
Automated candidate ranking and rejection is the part to approach carefully. A model that scores applicants learns from historical patterns, and those patterns can carry bias that becomes your legal and reputational problem. The EU AI Act treats some recruitment uses as high-risk for exactly this reason, which brings obligations a small buyer may not expect.
The defensible setup keeps a human deciding who advances and uses the tool to summarise and surface, not to auto-decline. A tool that silently rejects people is an exposure you cannot see.
There is a quieter problem too. Candidates now use AI to write applications, so some screening tools are scoring one model's output with another model. The firms that handle this well lean harder on structured interviews and work samples and trust the automated paper screen less.
How to buy without getting burned
Separate the safe jobs from the risky ones and buy accordingly. Pay for sourcing, coordination, and drafting, where the tools earn their keep. Be cautious with anything that ranks or rejects, and where you use it, keep the recruiter as the decision-maker on the record and check the vendor's answer on bias testing.
Ask every vendor where the model was trained, whether you can audit its decisions, and how it handles the high-risk classification under the EU AI Act. A vendor that cannot answer those has not thought about the part that matters. The report behind this note ranks the tools by job and flags which screening products have been caught making decisions they should not.