PlanAssist is moving into a 12-month beta involving 15 councils in England. Google DeepMind’s October 31, 2026 announcement describes an expansion from Barnet, Camden and Dorset to 12 additional authorities, with planning officers retaining the final decision on every application.
An expansion of an existing pilot
This is a new testing phase for a system piloted since June, rather than the first announcement of planning AI. DeepMind’s update says the additional councils will join in three cohorts, beginning with a subset. Fifteen participating authorities therefore does not mean every authority began using the system on the announcement date.
The government’s parallel release is marked October 30, while referring to delegation rules taking effect Saturday, October 31. Those rules, the AI beta and additional planner recruitment are distinct parts of the wider announcement. They should not be collapsed into one AI result.
What the assistant contributes
Built with Gemini and developed with the government, Google Cloud and Faculty, PlanAssist brings application material and planning context together for officers. DeepMind describes checks on submitted documents and drawings, relevant constraints and policies, and earlier objections, followed by cited draft reports that officers can edit.
The government says the beta uses the tool alongside normal processes, and the software will not be mandated. Wider availability across England in 2027 remains an intention subject to successful testing, rather than a current nationwide service.
Read the early results at their stated scope
DeepMind reports a 58% reduction in time spent on householder assessment tasks in early pilots. It also says officers approved more than 70% of generated reports without edits. These are partner-reported pilot findings; Nerova has not independently evaluated PlanAssist.
The distinction between assessment time and end-to-end turnaround matters. An application can wait for missing material, consultation or scheduling even when an officer completes a particular task faster. A task-time percentage cannot be directly applied to every applicant’s calendar wait.
Similarly, accepting a report without edits is not an independent measure of its correctness. Understanding that result requires the review criteria, application mix and checks on omissions. The beta is an opportunity to examine those questions across a broader set of councils.
What a useful scaled evaluation should establish
Our assessment is that the next evidence should separate document completeness checks, policy retrieval, report drafting and officer review. Each stage has different failure modes. A missing drawing is different from citing an outdated policy or overlooking a location-specific constraint.
For a comparable public-sector workflow, useful measures would include reviewer time, missed issues, corrections after approval and the proportion of cases sent back for more information. Measure results across straightforward and difficult cases, rather than relying only on an average.
Officers also need a clear route from an assessment statement to the underlying application document or policy. A citation is useful only when it supports the relevant claim and remains accessible to the person making the decision. Human review requires evidence that can actually be checked.
The beta’s significance is the move from a small pilot toward live use across different authorities while preserving officer responsibility. The next milestone is evidence that those controls and the reported time savings hold across local conditions, alongside a clear account of where the system still needs correction.