Are ai recruitment tools quietly filtering out older care workers — what families and managers should ask?

I recently started asking a lot of questions about how care homes and homecare agencies recruit frontline staff. Not because I wanted a job, but because I worry about the people at the sharp end of our social care system — and about whether the tools supposedly helping managers could be quietly shutting older, experienced care workers out of the running.

Why this matters now

We hear a lot about AI in recruitment as a time-saver: automated CV screening, video interview analysis, personality games and recommender systems that match candidates to roles. Vendors such as HireVue, Pymetrics, LinkedIn Talent Solutions and a host of Applicant Tracking Systems (ATS) like Bullhorn or Greenhouse are increasingly integrated into hiring workflows. For cash‑strapped care providers, anything that speeds up hiring is tempting.

But speed can come at a cost. The care sector already struggles with staffing shortages and a workforce that includes many older workers whose experience, reliability and local knowledge are invaluable. If AI tools end up favouring younger applicants because of proxies in data (like social media links, years since graduation, digital fluency), we risk losing that experience. As someone who follows tech and policy, I think families choosing care and managers running services should be asking hard, practical questions now — not later.

How AI tools can produce age bias

AI models learn from historical hiring data. If a provider in the past relied on younger recruits, or preferred applicants with certain online footprints, the algorithm can internalise those patterns and reproduce them. Age is often not explicit in a CV screening model, but it can be inferred indirectly from:

• Employment dates and career gaps
• Qualifications tied to graduation year
• Language and hobbies that correlate with generational cohorts
• Video analysis that interprets physical appearance or speech patterns
• Platform behaviour — for example, if an ATS scrapes LinkedIn activity and older workers are less active there

Video interview scoring systems have been criticised for picking up superficial cues like facial micro‑expressions or speaking tempo. If those systems weren’t tested for older users, they may systematically rate older candidates lower.

Questions I tell managers and families to ask vendors

You don’t need to be a data scientist to hold a vendor to account. These are direct, practical questions that cut to the heart of how a tool might disadvantage older applicants. I use them when I speak to providers and have shared them with carers and relatives who asked me for guidance.

• Does your system use age as an input or proxy? If not explicitly, ask for a list of fields and features used in decision-making.
• Have you run bias or fairness audits? Ask to see or receive a summary of audits covering age as a protected characteristic.
• What datasets trained the model? Know whether the training data came from similar care organisations or from broad web data that may not reflect the care sector’s demographics.
• Can you provide explainability? Ask for plain-English explanations of why a candidate was rejected or ranked lower.
• Is there human oversight? Who reviews flagged mismatches or edge cases? How easy is it to override automated decisions?
• How do you handle video interviews? What measures are in place to avoid penalising people for accents, expression, glasses, or mobility differences?
• What privacy protections and GDPR compliance do you have? Especially important for sensitive candidate data and biometric analysis.
• Can you share metrics for workforce diversity since adoption? Look for trends in age distribution before and after deployment.

Practical checks for managers using these tools

If you manage a care service and are considering (or already using) AI hiring tools, I recommend a few straightforward steps to spot and fix problems:

• Run regular age-disaggregated hiring reports. Compare application-to-hire rates across age bands. Sudden declines in hires for older brackets are a red flag.
• Keep humans in the loop. Make automated scores a suggestion, not the final decision. Ensure recruiters can see original CVs and add contextual judgement.
• Test with shadow applications. Try submitting matched CVs that differ only in age-related markers to see if outcomes change.
• Offer alternative assessment paths. If your process uses video or gamified tests, provide phone interviews or in-person assessments as an option.
• Train hiring teams to spot proxy discrimination. Awareness helps staff question odd patterns and challenge vendor claims.

What families should ask when choosing care

If you’re a relative or an older person looking for care, the technology behind recruitment may seem remote — but it has a direct effect on quality. Here are simple, actionable questions to ask a care provider:

• How do you recruit your carers? Ask about the process: online application, tests, interviews, and whether anything is automated.
• Do you welcome applications from people without digital profiles? Older, experienced carers may not have LinkedIn — that shouldn’t be a barrier.
• Are there alternative ways to apply? Phone, in-person, or paper applications should be available.
• How do you train and support carers who may be less tech‑confident? Ongoing training is a sign a provider values experience and invests in staff.
• What’s your staff age profile? If a provider has unintentionally narrowed recruitment, it might show up in a skewed age distribution.

Example due‑diligence table for managers

I’m not arguing we should reject technology wholesale. When designed and governed well, AI can free up managers’ time and help match the right people to the right roles. But I am arguing that care providers, families and regulators need to treat recruitment tech like any other safety‑critical system: with transparency, testing and clear routes for human override. Asking the right questions now can protect both older workers and the people who depend on them.