Best AI Tools for HR: Safe Wins and Legal Landmines

Most roundups in this category list twelve tools and mention the legal position in one sentence near the end. That ordering is backwards. In hiring, the regulatory question determines whether you can use a tool at all, which makes it the first filter rather than a footnote.

This is written for small companies and founders doing their own hiring rather than for enterprise HR departments with legal teams — which is precisely the group most likely to adopt a screening tool without realising what comes attached.

The two groups

	Administrative AI	Decision AI
Examples	Scheduling, notes, drafting, onboarding, FAQ	CV screening, ranking, video assessment, performance scoring
Who it affects	Your time	Someone’s employment
Failure cost	Mild annoyance	Discrimination claim
Regulatory load	Ordinary data protection	Specific AI hiring rules in several jurisdictions
Verdict	Adopt freely	Only with audit, disclosure and human review

Administrative tools worth using

• Interview note-taking. Otter, Fireflies, Granola and the recording features in Zoom, Teams and Meet. Genuinely transformative for interview quality, because you can actually listen instead of typing. Get consent, every time, and say what you are recording.
• Job description drafting. Any general model. Paste your best previous posting and the role requirements, ask for a draft in that voice. Then ask it to flag language that might narrow your applicant pool — it is genuinely decent at spotting that.
• Scheduling coordination. Calendly-class tools with AI routing. Unglamorous and it removes the single most tedious part of running a hiring process.
• Policy and handbook drafting. Excellent first drafts of the documents nobody wants to write. Have anything with legal weight reviewed by someone qualified in your jurisdiction.
• Internal FAQ and onboarding answers. A model grounded in your own handbook answering “how much holiday do I have left” style questions. High value, low risk.
• Summarising employee survey free text. Reading four hundred open comments is a real job. Clustering them into themes is exactly what these tools do well.

Decision tools, and what comes attached

CV screeners, candidate rankers and automated assessment tools promise to cut a hundred applications to ten. Here is what a small company usually does not know before buying one.

• New York City requires bias audits. Local Law 144 requires employers using automated employment decision tools for candidates in NYC to have an independent bias audit conducted, publish a summary of the results, and notify candidates. That is an ongoing obligation with a real cost, and it applies to you as the employer rather than to the vendor.
• The EU AI Act classifies employment AI as high risk. Systems used for recruitment, selection, promotion and termination decisions fall into the high-risk category, carrying obligations around risk management, data governance, human oversight, transparency and record-keeping.
• Existing discrimination law already applies. This is the part people miss. You do not need an AI-specific statute to be liable. If your process produces disparate outcomes, ordinary equality legislation covers it, and “the software decided” is not a defence anywhere.
• Data protection rules constrain automated decisions. Under GDPR, decisions with legal or similarly significant effects made solely by automated processing carry specific requirements including a right to human intervention.

Why screening tools inherit bias

The mechanism is worth understanding rather than taking on trust, because it explains why “we removed names and photos” does not solve it.

These systems learn from historical hiring data — who was hired, who succeeded. If past hiring favoured a group, the pattern the model learns is that group’s characteristics. It does not need a protected attribute to do this. University attended, postcode, gender-coded language in self-description, employment gaps, sports played: all correlate with protected characteristics and all survive anonymisation.

The well-documented case is Amazon, which abandoned an experimental recruiting tool after finding it penalised CVs containing the word “women’s”, having learned from a decade of male-dominated applications. Amazon had world-class engineering resources and could not fix it. A small company buying a vendor tool has considerably less visibility into what it learned.

If you use one anyway

There are legitimate reasons to — genuinely high application volumes make some filtering necessary. Six conditions I would insist on.

1. It ranks or shortlists; it never rejects. A human reviews every rejection. This alone addresses much of the legal exposure and most of the moral objection.
2. Ask the vendor what it was trained on and what bias testing exists. A vague answer is a decision.
3. Audit your own outcomes. Track demographics of applicants versus those advancing at each stage. If the funnel narrows unevenly, you have a problem regardless of what the vendor says.
4. Tell candidates. Required in several jurisdictions and correct everywhere. Say what is automated and offer a route to human review.
5. Keep records. Which tool, which version, what criteria, what it output, who reviewed it. You will need this if challenged.
6. Sanity check it periodically. Take twenty CVs it rejected and have someone read them properly. If good candidates are in there, you have your answer.

The candidate side, which cuts both ways

Worth noting because it undermines the premise. Candidates now use AI to write CVs and cover letters at scale, which is why application volumes have risen sharply, which is why employers reach for screening tools, which drives candidates to optimise harder.

Both sides are now running language models at each other, and the signal in a written application is worth less than it was. My honest conclusion: the response is not a better filter. It is to weight earlier stages differently — a short structured task, a brief real conversation — which is expensive per candidate and actually tells you something.

My verdict

Adopt the administrative tools immediately. Note-taking, drafting, scheduling, policy documents, survey analysis. Real time saved, minimal risk, and in the case of interview notes it makes you better at the job rather than just faster.

Be sceptical of decision tools. Not because they never work, but because the obligations land on you, the bias is hard to see from outside, and the thing they optimise — filtering applications faster — is a symptom of a signal problem they also make worse.

If you are a small company drowning in applications, I would spend the money on a better-written job posting and a short paid task before I spent it on a ranking algorithm. Both improve the pool rather than sorting a bad one more quickly.

For the underlying tools, see the best AI content creation tools and what ChatGPT is and how it works. For automating the administrative side properly, workflow automation covers the rule-based approach.

Frequently asked questions