AI in Care Recruitment: Speed Over Substance Risks Quality

The telephone interview that Mollie Cole-Wilkin took for her care job lasted five minutes and felt, she says, like talking to another person. Her mum was in the next room. Neither realised she was speaking to an algorithm. The question worth asking is whether the person deciding someone is suitable to wash, medicate, and emotionally support vulnerable people should be human too.

Since August 2024, more than 14,600 people have had their first conversation with "Ami", an AI voice agent deployed by Cera, one of England's largest homecare providers. Over 1,000 of them have been hired. These carers will enter the homes of elderly and vulnerable people, providing the kind of intimate physical and emotional support that most of us assume requires a distinctly human touch.

Cera, which supports 2.5 million home visits monthly, argues its system addresses a genuine operational crisis. The company receives 500,000 applications annually. Traditional recruitment methods leave candidates waiting days or weeks for initial contact—enough time for many to accept other offers or lose interest entirely.

Enjoying this article?

Get stories like this in your inbox every week.

Join the Community

According to the company's own figures, Ami has halved the time from application to first interview and doubled job offers for the same recruitment spend. Those are impressive metrics. But they measure speed, not whether the right people are being hired.

The efficiency case for automation

The social care sector's workforce shortage is not abstract. Hospital beds remain occupied by patients well enough to be discharged but unable to go home without care packages in place. Lucy Kruyer, branch manager at Cera's Colchester office, frames AI recruitment as a direct response to this bottleneck.

People don't want to be lying in hospital waiting for care because they can't come home without it.

Ami uses a soft, calm female voice—an interesting choice given Cera acknowledges the evidence that female voices build trust is limited. The system asks standardised questions about experience, availability, and right to work. It scores applicants out of 100 based on their responses.

Those who pass move to an in-person interview with actual staff. For some candidates, the approach offers unexpected advantages. Cole-Wilkin, who occasionally stammers, found the AI less intimidating than facing a human interviewer.

The system responded with phrases like "I'm happy you shared that with me", which she described as rewarding. Cera argues this standardisation reduces bias, giving people who struggle with traditional interview formats a fairer chance. That claim deserves scrutiny.

Standardised questions may eliminate certain human biases, but large language models encode their own. They work through patterns and associations drawn from training data, which inevitably reflects the biases present in that data. Whether AI bias is preferable to human bias is a separate question from whether it exists at all.

What algorithms can't detect

Janet Beacham has spent more than 45 years in healthcare and nursing. Her objection to AI screening is straightforward: if they haven't got care in their heart, they're not going to be a good carer. That's not a sentimental statement.

Care workers enter clients' homes as guests, often dealing with people at their most vulnerable. They need the right personality and skills, Beacham argues, and only a person can judge that. The counterargument from Cera is that human recruiters remain in the loop.

Ami conducts initial screening; staff still run background checks, lead in-person training, and make final hiring decisions. Dr Ben Maruthappu, the company's founder and chief executive, insists his technology expands rather than replaces the workforce. The real question shouldn't be whether we use AI, he says—it should be how we use it to widen opportunity.

Empathy, warmth, genuine care—these qualities don't necessarily translate into the kind of responses a language model scores highly.

What's interesting here is that Maruthappu's framing positions speed as inherently democratising. By reaching candidates within seconds rather than days, Cera claims to capture applicants who would otherwise slip away. That's probably true.

But it sidesteps the question of whether initial screening by algorithm changes the type of person who makes it through to the next stage. Applicants who might excel in person but struggle to articulate their motivations to an algorithm in five minutes simply won't reach the next stage.

Beyond care: a template for mass recruitment

Cera isn't keeping this technology in-house. The company is now licensing Ami to employers in other sectors, including dentistry. That signals commercial ambition but also suggests this could become a template for recruitment across industries where workers interact with vulnerable populations.

This expansion is happening in what amounts to a regulatory vacuum. The government announced a "test and learn" approach to AI in public services in March 2025, emphasising innovation over restriction. But no specific legal framework exists yet for AI use in care recruitment.

Trade union Unison, through its head of social care Gavin Edwards, welcomes technology that frees up staff time but stresses that any use of AI must be transparent, fair, and compliant with equality and employment laws. The Local Government Association echoes this caution, noting that care is "fundamentally person-centred" and that AI must be co-designed with service users.

Both organisations emphasise keeping "a human in the loop" for final decisions. Whether that loop is wide enough remains an open question. Cera's process includes human oversight at later stages, but the initial filter—the moment where Ami decides whether someone's responses merit progression—operates autonomously.

The social care crisis demands urgent solutions, and nobody disputes that recruitment bottlenecks compound the sector's problems. But the pressure to hire faster shouldn't obscure whether we're hiring better.

As Ami expands into other industries and similar systems proliferate, the next 12 months will show whether this technology complements human judgment or quietly displaces it. Regulators who claim to be testing and learning would do well to define what success actually looks like before the template becomes the standard.

More articles by David Adams