A larger test of assistance, not automatic permission

England is expanding tests of PlanAssist, an AI tool for routine householder planning applications such as extensions and loft conversions. The government’s October 30 announcement takes the program from Camden, Barnet and Dorset to fifteen councils.

Google DeepMind’s October 31 account describes a twelve-month beta using live applications. Twelve additional councils will join in three cohorts, starting with a subset. That is a staged test, not confirmation that every participating council is already using the tool in the same way.

What the tool is intended to do

Developed with Google DeepMind, Google Cloud, Faculty and government partners, PlanAssist brings application material, policies and site constraints together for planning officers.

DeepMind says the Gemini-based tool checks whether documents and drawings contain required information, examines constraints and policies, and drafts cited reports. Officers can edit those assessments and retain responsibility for final decisions.

The government says officers will use it alongside their normal process to test performance at scale. It also says PlanAssist will not be mandatory software.

The headline saving needs a careful denominator

DeepMind reports a 58% reduction in the average time officers spent on householder assessment tasks during early testing in the original three councils. This is a result reported by the developer, not a measurement reproduced by Model Current.

The government release also describes routine processing moving from eight weeks to four. That broader claim sits alongside announcements about recruitment and changes to decision-making arrangements; the release does not provide a controlled comparison isolating AI’s contribution to end-to-end waiting times.

Time spent doing a task and elapsed time from submission to decision are different measures. Neither claim guarantees the same saving for every application or local authority.

A wider release remains conditional

The government intends to make the tool available to every council in England by the end of 2027. DeepMind makes that wider availability subject to successful testing.

The next useful evidence would include results across different council workloads, time needed to correct reports and whether officers can reliably trace an assessment to its underlying material. A larger pilot can help answer those questions; it does not settle them in advance.

Sources

  1. Google DeepMind: PlanAssist pilot and expansionDisplayed October 31 primary developer account: a claimed 58% reduction in average officer assessment-task time, the Gemini workflow, a staged three-cohort beta and qualified-officer review. The feed carried an October 30 timestamp; the displayed article date is retained. No independently reproduced benchmark.
  2. GOV.UK: planning-tool testing expansionOctober 30 government release: 12 additional councils, 15 in total, twelve months of testing alongside normal processes, non-mandatory software, and intended England-wide availability by the end of 2027. Its broader eight-to-four-week processing claim is attributed and distinguished from the task-time metric and other reforms. No blanket automatic-approval claim.