Scribe and prejudice? Exploring the use of AI transcription tools in social work
AI transcription tools could enable social workers to spend considerably more time with the people in their care. But if not reviewed properly, the outputs could lead to potentially serious consequences. Imogen Parker of the Ada Lovelace Institute argues that for these tools to have transformative power, robust regulation, testing and guidance is required
Last week, we published a new study on the use of AI transcription tools in social work. Transcription tools are seen as one of the most promising and potentially transformative applications of AI. These tools are being rapidly rolled out in critical frontline sectors with a high degree of information transfer, such as social care and healthcare.
Engaging with social workers and those involved in procuring and evaluating the tech across 17 different local authorities, our research explores how transcription tools are being used and assessed in practice, particularly amplifying the experiences and voices of frontline workers themselves.
Social work is a particularly significant use case. It’s high stakes because social workers are responsible for consequential decisions about vulnerable people. It’s also an early-adopting sector. Since the advent of more powerful, accurate tools (which draw on foundation models) we’ve seen a rapid roll-out, with one AI transcription tool in use in 85 local authorities last year. It’s also a sector where transcription could be highly transformative: social workers are responsible for extensive documentation about cases and records. This has led to a situation where social workers are spending only around 20% of their working week in direct face-to-face contact with people drawing on care.
So what did we find?
There is real potential for AI transcription to help the public sector
At the end of last year I asked: is transcription the use case that shows AI’s transformative power?
Our new research shows how AI – if effective and trusted – can quickly become highly valued by frontline professionals. While only a small-scale sample, the social workers we spoke to were overwhelmingly enthusiastic about AI transcription. In previous research, we have seen data-driven technologies in the public sector coming into conflict with aspects of professional expertise and judgment.
With transcription tools, many social workers reported a ‘freeing up’ of time, allowing them to focus on the meaningful core of their relational practice. Nobody gets into social work for the paperwork. Many social workers were happy to view and treat these tools and their outputs as something enhancing their existing professional practice, rather than disrupting or conflicting with it.
However, in this early stage of adoption and testing, social workers are being given a lot of flexibility and discretion about when and how to use transcription tools, and so the reality of their use in practice is very varied. For example, some are opting not to use AI transcription with some families or in some contexts, some are using them as an ‘on-the-fly’ notetaker when travelling between appointments and some have it rolling in internal meetings. Some said they only used them for transcription, while others used the summarisation capabilities as well.
Most social workers we spoke to felt that so far, the time-saving benefits were flowing directly to the social workers who can then decide how to use that saved time. One team leader did suggest that this could change in future, for example, through the use of service targets and mandates (e.g. increased caseloads, quicker appointments). This may affect social workers’ views about AI’s potential to augment professional practice, rather than automate it.
As part of the Innovation 2026 event programme, Global Government Forum is bringing together leaders from across the NHS and wider public sector to discuss how to achieve the 10 Year Health Plan. The conference and exhibition takes place on 24 and 25 March 2026 in London. Find out more
Social workers are being expected to manage risks arising ‘upstream’
Social workers assume full responsibility for AI transcription tools, and the time they are spending on reviewing transcriptions or summarisations for accuracy varies hugely. Some told us they spent a few minutes checking transcriptions, but others spent hours. It is unclear how much oversight is needed to review these outputs and how well the current approach to this ‘human in the loop’ review is working in practice.
There are two important implications. It’s important that ‘time savings’ from transcription tools are protected for social workers to undertake this additional time spent on reviewing, rather than being absorbed as efficiencies for the service. Any cost-benefit analysis needs to consider how social workers are using these tools safely in practice, including any additional time demands.
Second, we need more evaluation of current approaches to the ‘human in the loop’ review process. Social workers told us of serious inaccuracies they had caught in documentation, from ‘gibberish’ around names to false reports of suicidal ideation. Failing to catch these AI ‘hallucinations’ can have serious repercussions for people who draw on care, professional repercussions for social workers, or potential legal consequences if faulty evidence contributes to formal proceedings and decision-making in courts.
Given the UK’s lack of ‘upstream’ regulation of foundation models, which AI transcription tools rely on, frontline professionals may be taking on risks they can’t necessarily tackle or assess. Low-level but persistent gender bias in summarisation, for example, might be harder for a social worker to spot than directly inaccurate content.
Read more: Slick cities: How local authorities are using AI to tackle their most pressing problems
We need to address the structural gaps
Given the rapid rate of adoption, it’s welcome that many studies are looking at evaluating these tools. However, we found that resourcing challenges are incentivising local authorities to move quickly, through light-touch pilots and sometimes pared-down procurement processes, with knock-on implications for the quality and breadth of evidence gathering and evaluation.
Some dedicated providers are committing to co-producing a pilot and evaluation with every local authority using their tool, but ‘shadow use’ of generic tools with AI transcription capabilities is less likely to be subject to this type of approach.
This lack of grounded evidence or independent, systematic evaluations on different AI transcription tools means that public servants and policymakers are not yet able to properly understand how the use of AI transcription tools may impact social work, including the potential risks for people drawing on care.
Local authorities and procurers don’t have any detailed insights about how different tools compare on different aspects of performance, such as bias or accuracy.
And without clear sectoral guidance informed by regulators, professional bodies, social workers and people drawing on care, it remains unclear what legitimate or appropriate use should look like.
This approach to managing or mitigating the potential risks leaves sectoral adoption fragile: without adequate regulation, testing and guidance, a high-profile transcription error or incident could trigger a sector-wide retreat from the technology.
No public sector intervention is perfect, and it may be the case that, despite the potential errors, transcription tools used responsibly are able to produce outputs that are as good or better than a stretched social worker’s own notes. It’s certainly the case that AI transcription – if safe, fair and seen as legitimate – would be welcomed by the social workers we spoke to as a means of improving their practice.
That’s why we need structures – like a new What Works Centre for Public Sector AI – to assess tools in practice and in detail, to factor in risks and opportunities and to offer balanced guidance rather than leaving local authorities, or even individual social workers, navigating the responsible use of these under-scrutinised tools.
Read more: License to build: understanding what people think about public sector use of AI
Read more about AI in government and the public sector on Global Government Forum’s dedicated AI hub
About Imogen Parker
Imogen is Associate Director (Society, justice & public services) at the Ada Lovelace Institute. Imogen’s career has been at the intersection of social justice, technology and research. In her previous role as Head of the Nuffield Foundation’s programmes on Justice, Rights and Digital Society she worked in collaboration with the founding partner organisations to create the Institute. Prior to that she was acting Head of Policy Research for Citizens and Democracy at Citizens Advice, Research Fellow at the Institute for Public Policy Research (IPPR) and worked with Baroness Kidron to create the children’s digital rights charity 5Rights. She is a Policy Fellow at Cambridge University’s Centre for Science and Policy.
