A recruitment demo can make a crowded inbox look wonderfully manageable. Upload the applications, wait a moment and receive a tidy shortlist. Before buying, ask to see what happened to the people who didn't make that list.
That's where useful questions begin. What did the software assess? Which applications did it exclude? Could the hiring manager see the original evidence and change the result?
If you're choosing a tool for a small business, make the supplier demonstrate a complete application journey. A polished dashboard won't tell you enough about how a real candidate will be treated.
Pin down the job you want the tool to do
Write a short description before discussing features. For example, you might want help arranging interviews after a person has chosen the candidates. Or you might want software to score every application against the job description. Those are different uses and deserve different questions.
Our suggested starting document has four lines:
- The problem you're trying to solve, such as spending too long arranging appointments.
- The information the tool would receive.
- The recommendation or action it would produce.
- The person who could be affected if it got that wrong.
Take that document into the demonstration. Ask the supplier to use it as the scenario, including an application with missing information. See whether the software asks for clarification, makes an assumption or quietly reduces the candidate's score.
Check the EU connection and classification
Start with whether the EU AI Act applies to your business. A UK address doesn't settle the answer. The Act can reach businesses supplying systems into the EU and certain activities where the output is used there. Article 2 explains the scope.
For activities within scope, Annex III covers specified recruitment uses, including filtering applications and evaluating candidates. Article 6 provides a limited exception for some systems that don't present significant risks and meet particular conditions. Annex III systems that profile people remain high-risk. A person clicking an approval button doesn't automatically create an exemption. Check the current consolidated Act, especially Article 6 and Annex III.
Ask the supplier for its written classification and the intended use on which it depends. Then compare that with how your team plans to use the tool. If those differ, get the difference resolved before buying.
The Commission confirms that the main Annex III high-risk rules apply from 2 December 2027 following the July 2026 AI Omnibus changes. That future date doesn't suspend existing data protection duties or make an unsuitable hiring process acceptable now.
Test what the human reviewer can actually change
The ICO's 2026 recruitment work found that employers sometimes described their process as human-led even though reviewers effectively accepted automated results. Its guidance on meaningful human involvement focuses on real influence before the decision takes effect, supported by the authority and ability to change it.
During your demonstration, choose a fictional candidate the tool ranks poorly. Ask the supplier to show how a recruiter would open the original application, understand the concern and move the candidate forward. Check whether that is possible for every applicant, including those the system would otherwise hide.
Now consider the time available. If the proposed process leaves a manager a few seconds per application, its description of careful review may be unrealistic. Budget for the work you expect the reviewer to do.
Keep the legal questions separate. UK rules about solely automated decisions and the EU AI Act's high-risk classification have different tests. Getting one answer doesn't settle the other.
Ask for evidence that fits your vacancies
The ICO's procurement questions cover accuracy, bias, transparency and limiting personal information. Use them to challenge broad claims about fairness.
For your own comparison, ask each supplier to explain:
- Which types of roles and applicants it tested.
- What a poor result looks like and how it is detected.
- Whether recruiters can inspect the reasons behind a recommendation.
- What happens when the product or underlying model changes.
- Which features can be switched off.
For example, imagine you're hiring a part-time administrator. Ask how the tool handles a fictional application with a career break and experience gained outside conventional employment. You're looking for an explanation you can assess, not a promise that the software is unbiased.
Avoid uploading genuine applicant records into a trial until the data arrangements and necessary approvals are in place.
Follow the information beyond the demo
A data protection impact assessment examines how a proposed activity could affect people and how those risks will be addressed. The ICO's 2026 findings on recruitment DPIAs explain why this work needs to reflect the actual recruitment process.
Ask for a simple data-flow diagram. It should show what leaves your business, which organisations receive it, how long it stays there and how deletion works. Include CVs, recordings, assessments, generated scores and support tickets.
Work out who will answer a candidate who challenges a result. The hiring manager should know where the relevant records are and who can explain the tool's role. A supplier support address alone won't make that conversation easy.
Before signing, put unresolved questions into a short decision note with a named owner. It's reasonable to postpone the scoring feature while adopting a useful scheduling feature, provided the remaining setup has been properly assessed. Buy the help you can explain and manage.