30-second briefing
The review process is the product decision
- DHSC published its adult social care AI guidance on 29 September.
- The guidance identifies limited evidence for truly personalised AI-generated care plans and requires staff review for accuracy.
- Care Circle’s assessment: measure correction effort and the quality of the final record alongside any time saved.
A care record is an account of a person’s needs, wishes and support. When software helps write it, the central question is who can explain and stand behind the final account. Fluent wording can make a draft easier to read without making its contents more reliable.
The Department of Health and Social Care’s guidance, published on 29 September, puts AI-assisted care planning alongside monitoring, administrative work and audit tools. It also states that evidence for genuinely personalised plans remains limited. That qualification deserves as much attention as the efficiency claims.
What has changed
The new guidance gives providers a common starting point for deciding where AI belongs. It supports human review, clear accountability, staff skills, data protection and measurable outcomes. It is guidance about responsible adoption, not an endorsement of a particular system or proof that every use improves care.
The practical consequence is that a technology discussion now needs a care-quality owner. A procurement team can examine price and contractual terms; a registered manager or relevant practitioner must help define what an acceptable record looks like. Those are different decisions that need to meet before rollout.
The risk sits between the draft and the record
Consider an illustrative handover note in which a person declines an activity. A summary might collapse that account into a general statement about behaviour, lose the person’s explanation, or omit the support offered. The concern is the change in meaning. Checking spelling would not catch it.
A meaningful review should therefore compare the draft with the underlying account, including chronology, uncertainty, preferences and the difference between observation and interpretation. The reviewer needs access to the original material and enough time to challenge the output. A tick box labelled human reviewed does not describe that work.
Test a narrow task before widening the programme
Start with a clearly bounded task and an agreed baseline. Record how long the existing process takes, the quality problems already present and the people involved. Test the same measures after introducing assistance, including time spent correcting errors and supporting colleagues.
Use realistic cases reflecting the service, including complex communication and incomplete information. An attractive demonstration with clean input tells a provider little about difficult everyday records. Define which outputs may be drafted, which decisions remain with staff, and what triggers stopping the pilot.
The supplier discussion needs to reach the whole workflow
Ask where information is processed, who can access it, how long it is retained and how the provider can export or delete it. Have the appropriate information-governance lead examine those answers rather than relying on a generic assurance slide.
Changes to the model, prompt or software can change the output. A provider should know how changes are communicated and when its own review process must be retested. The adoption cost includes training, supervision, checking and ongoing ownership, as well as the licence.
What we will examine next
The next useful evidence will show the difference between gross drafting time saved and net time released after checking, and whether the quality of the final account is maintained. Providers can contribute anonymised process measures and lessons without sharing resident records.
This develops our earlier coverage of digital assurance: a completed digital workflow is a starting point for scrutiny. The important evidence concerns what people do when the system produces an incomplete or misleading account.
Questions leaders should ask now
- 01
Who signs off the final account?Name the responsible reviewer for each use case and explain what they check.
- 02
What is the net time saving?Include correction, training and supervision in the comparison.
- 03
Can the person’s voice be recovered?Check that preferences, uncertainty and context survive summarisation.
- 04
When would the pilot stop?Agree thresholds for material errors and escalation before introducing the tool.
The Care Circle view
Start with the record you need to trust
A strong AI programme begins with a defined care task and a defensible review process. Staff need the authority to reject an output, not a target that rewards accepting it quickly.
The clearest evidence of value will be a better or equally reliable final account, with a measured reduction in workload and an explanation of where responsibility sits.
Earlier evidence
The digital assurance divide.
Read the August provider-scale analysis. Its original evidence period is retained; this feature makes no fresh measurement of the gap.
How the story develops
Continue from the earlier evidence.
This feature develops a continuing leadership question. Earlier publication dates and evidence periods remain visible.
Beyond DSPT: can you prove your cyber controls actually work? · 9 August 2026
A digital reset needs to prove that care can continue · 5 August 2026
Digital care records reached 83.6% of locations. Assurance is the next question · 9 October 2026
AI inside the care workflow: govern the decision, not the product label · 9 July 2026
Develop the analysis
Read the connected flagship reports.
Digital continuity: can the care service depend on its systems?
Operational assurance: suppliers, equipment and resident voice
Sources, method & limitations
How to read this analysis
Care Circle examined the primary publications linked below on 9 October 2026. Statements about official policy or guidance are attributed to their source. Operational examples, suggested questions and the Care Circle view are editorial analysis, not findings from a provider survey.
- No provider-specific assessment, eligibility decision or prediction of regulatory outcomes is made.
- Source publication dates and this edition’s review date are separate. Local arrangements and later updates may change the position.
- Suggested management actions support discussion with appropriately qualified advisers; they do not replace individual care planning or professional judgement.