Most AI Recruiting Tools Do Nothing. Here Is What Actually Moves the Numbers

The AI recruiting category has produced an enormous number of features and a small number of things that change a staffing firm's P&L. We build in this category, which makes the following list commercially awkward and more useful than the alternative. Four capabilities move a number. Most of the rest are demos.

Last reviewed August 2026. Assessment based on Sortinghat delivery data and customer observation.

What actually moves a number

Search over your own database. An audit of one firm's 3 million records found roughly 80 percent were real people with out-of-date records. Making them findable took portal spend from about $100 a role to $60. That is a line on a P&L, not a productivity claim.

Screening at volume. A recruiter manually manages around 40 calls a day and finds three or four people worth submitting. Filtering first produces roughly 10 submission-ready profiles in hours. That changes how many roles a recruiter can carry.

Criteria generated per role. Fewer wrong submissions, which shows up as a submission-to-interview ratio rather than as a feature.

Automatic record capture. Removes re-entry and, more importantly, means a desk survives a resignation.

What does not move a number

None of these are useless. They are just not what they are priced as.

AI-written outreach. Personalisation at scale became free, which is exactly why it stopped working. Covered in what changed about outbound.

Chatbots on a careers page. They deflect questions. They do not produce candidates.

Resume parsing improvements. Real and incremental. Parsing was rarely the bottleneck.

Sentiment analysis on candidate messages. A demo that has never changed a hiring decision we have seen.

Predictive attrition scoring. Interesting, rarely actionable, and frequently just tenure with extra steps.

Why so much of it does nothing

Three structural reasons rather than vendor dishonesty.

The bottleneck was misidentified. Most recruiting tools optimise sourcing, and on most desks sourcing was never the constraint. Qualification was.

The output lands on a human anyway. A tool that produces a transcript, a summary or a score somebody still has to read has moved work rather than removed it.

It was built for internal talent teams and sold to agencies. Different economics, different volumes, different definition of success. Much of the category does this.

The test to apply to any tool

One question: which number on your P&L or operating report will be different in ninety days, and by how much?

Acceptable answers are placements per recruiter, job portal spend, time from brief to shortlist, submission-to-interview ratio, or receivables ageing. Unacceptable answers are engagement, efficiency, insights and experience.

If a vendor cannot name a number, the honest reading is that they have not found one either. That is a fair question to ask us as much as anyone.

The uncomfortable part

The largest determinant of whether recruiting software produces a result is not the software.

Industry reporting consistently shows a wide gap between firms claiming AI usage and firms with it genuinely embedded in daily work. A capable platform nobody opened produces exactly the same outcome as no platform, and the money is spent either way.

That is a management problem and no vendor can solve it for you, which is why we publish a ninety-day adoption plan rather than claiming the product drives its own usage.

Frequently asked questions

The question to ask every vendor, including us

Which number will be different in ninety days, and by how much? Write the answer down before you buy, then check it at day ninety.

Most firms never do the second part, which is why the same disappointing purchase gets made repeatedly across the industry with different logos on it.