England is expanding trials of PlanAssist, an AI tool designed to help planning officers assemble evidence and assess routine applications. The next stage will test more than speed: councils will need to show that time savings survive careful review, that relevant constraints remain visible and that responsibility for decisions stays clear
Planning decisions bring together documents, policies, site information and professional judgement. A tool that reduces the effort of assembling that material could be useful, but its public value depends on what happens after the information is collected. Faster preparation needs to support a decision that can still be explained and examined.
In a 30 October 2026 announcement, the UK government says PlanAssist testing is expanding from Camden, Barnet and Dorset to 15 councils for 12 months. Officers will use it alongside their normal process, and the government intends to make it available across England by the end of 2027. The software will not be mandatory.
The announcement reports a 58 per cent reduction in time spent on key tasks during early testing. That is a government-reported result, not proof of an equivalent reduction in the full duration of every application. The expanded trial is an opportunity to examine that distinction at a broader operational scale.
Separate the tool’s contribution from the wider reform
The same announcement combines AI with recruitment and changes to delegation arrangements for straightforward applications. These are different interventions. A shorter wait could reflect several changes working together, so evaluation should avoid attributing every improvement to the software.
This is a familiar challenge in administrative reform. A new tool may arrive alongside revised responsibilities, additional staff or a simpler process. Measuring the combined outcome is useful, but it answers a different question from identifying the particular contribution of AI.
An informative evaluation would describe which activities changed and when. It would compare similar tasks where possible, acknowledge differences between councils and record other factors affecting performance. These are suggested evaluation practices, rather than a claim that the published trial already includes a particular research design.
An initial assessment is an aid to judgement
The earlier June announcement described the prototype, now named PlanAssist, as triaging applications, summarising information and providing an initial assessment for officers to consider. It also distinguished the prototype from Extract, a separate tool for converting historical planning documents into usable data.
These functions can assist different parts of a workflow. Extracting a fact, summarising a policy and suggesting an assessment are not interchangeable activities. Each needs a different kind of checking, even when they appear together on one screen.
Consider a hypothetical extension application. An assistant might correctly identify a relevant document but overlook a qualification within it. A reviewer needs to inspect the supporting passage and understand how the proposed assessment uses it. A polished summary is most useful when it makes that inspection easier, rather than encouraging acceptance through confident presentation.
Preserve the route from conclusion to evidence
A planning file can contain different versions of a drawing, material submitted at different times and information that applies only to part of a site. Bringing those sources into one place helps only if the relationships remain intelligible.
A reviewable process would make clear which document and version support a statement. It would distinguish an extracted fact from an inference and identify gaps that require further investigation. Where information conflicts, the officer should be able to see the conflict rather than receive an apparently settled answer.
Those are proposed design priorities, not verified descriptions of every PlanAssist feature. They matter because the administrative record must remain understandable after the immediate task is complete. An explanation should not depend entirely on what a particular user remembers seeing during an interaction with the tool.
Measure checking effort as part of the task
A quick draft can create work elsewhere. An officer may save time collecting material yet spend longer checking a summary or correcting an initial assessment. That does not make the tool ineffective, but it changes the calculation.
Useful measures would follow the completed task through preparation, verification, correction and approval. They could record substantive omissions as well as elapsed time. A small number of difficult errors might matter more than a large number of harmless wording changes.
Councils should also examine how review effort varies. Straightforward cases may be easy to check while unusual circumstances require close attention. An average time saving could conceal those differences. Reporting the boundaries of the tested work would make the findings more useful to authorities considering adoption.
Keep human responsibility connected to real working conditions
The October announcement says planning officers remain responsible for reviewing applications and making decisions. That allocation is important, but responsibility becomes effective through the organisation of work: access to evidence, relevant expertise and enough time to question an output.
If a team treats the AI assessment as the expected answer, disagreement may become harder even without a formal requirement to follow it. Staff should be able to record why they departed from a suggestion and raise recurring concerns without that being interpreted as failure to embrace innovation.
Training could use examples where the assistant is incomplete or uncertain, alongside examples where it performs well. This would help officers practise the judgement the workflow expects of them. These are practical recommendations, not reported findings about staff behaviour in the initial councils.
Make broader adoption a decision supported by evidence
Expansion to more councils can reveal conditions that a small trial does not capture. Authorities may organise records differently, have different support capacity or face different combinations of application types. Consistent software does not guarantee consistent administrative circumstances.
A council considering adoption would benefit from evidence about onboarding effort, support needs and the work required to maintain reliable source material. It should also understand the consequences of model or system changes and have a workable route for reporting defects.
The government’s intention to widen availability is therefore distinct from a conclusion that every council should deploy the tool in the same way. A non-mandatory offer creates room for local decisions, but those decisions become more credible when authorities can compare the relevant benefits, costs and limitations.
Public value includes the ability to explain a decision
Residents need more than a shorter processing time. They need a decision reached through a reliable process and an explanation they can understand. Those goals can support one another when technology reduces avoidable searching and leaves officers better able to focus on the substance of a case.
The next stage of PlanAssist should be read as a test of that relationship. Reported task savings are promising evidence to investigate, while broader service outcomes and the quality of review need their own assessment.
The wider lesson for public-sector AI is to evaluate assistance where it joins professional judgement. A successful system would make relevant information easier to inspect, preserve a clear account of responsibility and demonstrate benefits across the completed workflow. Speed becomes public value when those conditions hold together.