Google DeepMind Deploys PlanAssist AI to 15 English Councils for Live Planning Decisions

Google DeepMind has moved its PlanAssist planning assistant from a three-council pilot into a 12-month beta across 15 English local authorities. The tool, built on the company's Gemini model, is designed to speed up the assessment of routine householder applications—loft conversions, conservatories, and extensions—while leaving the final decision with a qualified planning officer. DeepMind reports that early testing cut the average time officers spent on assessment tasks by 58%, though that figure measures a narrow slice of the process and comes from the company's own evaluation [1][2].

What PlanAssist does

PlanAssist was developed by Google DeepMind with the UK government, Google Cloud, and the AI firm Faculty. It targets the roughly 70% of planning officers' workload that comes from routine householder applications [1][8]. The system runs a sequence of checks: it validates that application forms and architectural drawings contain the required information; it reviews site constraints such as tree preservation orders and heritage designations; it assesses the proposal against local and national planning policies; and it drafts a fully cited assessment report that the officer can edit before reaching a decision [1][3]. The tool does not approve applications; officers retain statutory decision-making authority [1][2][8].

Pilot results: company-reported gains

The initial pilot ran from June across the London boroughs of Barnet and Camden and the largely rural Dorset Council. DeepMind says the average officer time on householder assessment tasks fell by 58%, and that more than 70% of the tool's generated reports were accepted by officers without edits [1][3][4]. Those reports covered checks such as policy compliance and submission completeness. The government press release states that early testing halved routine householder processing from eight weeks to four [8].

The figures describe officers' use of drafts within a supervised workflow. They do not establish a comparable cut in applicants' total waiting time, and the sources note they do not constitute independently verified accuracy [2][4]. DeepMind's blog post reports a 58% reduction in officer assessment time [1]; a company post on X phrased it as "up to 58%" [4]. The government press release cites a 58% reduction in "key planning tasks" [8]. Based on the available announcements, no independent evaluation has been published.

Beta expansion: 12 more councils, 12 months

The beta adds 12 authorities in three cohorts: Ashford, Lambeth, Newcastle, Nottingham, Bristol, Cumberland, North Lincolnshire, Rochdale, Redbridge, South Gloucestershire, Stoke-on-Trent, and Westminster [2][8]. Together with the original three, the 15 councils span metropolitan and rural districts, giving the system a broader test across different local plans, site constraints, and application volumes [1][3]. Officers will use PlanAssist alongside their normal processes during the trial [2][8].

The government intends to make the tool available to every council in England by the end of 2027, subject to successful testing. Adoption would not be mandatory [1][2][8]. The Department for Housing, Communities and Local Government is also mandating the digitisation and open-data publication of key planning registers by 2027, which would support other planning tools [8]. Over £14 million is being invested in planner recruitment and training this financial year [8].

What remains unproven

Several questions sit beyond the current evidence. The posts read do not say how many applications the pilot covered [4]. The 58% time saving measures officer assessment tasks only, not the end-to-end decision timeline that applicants experience. The "no edits" rate reflects first-pass usefulness of drafts under supervision; it is not an independent audit of correctness. The beta's staged rollout is designed to test whether the reported gains hold across higher volumes, varied local policies, and a wider range of applications [3]. No pricing or contractual terms for councils have been published.

Government context and conditions

PlanAssist is part of a broader push to meet the government's target of 1.5 million new homes by 2029 [1][3][8]. The same announcement introduced a National Scheme of Delegation to let expert officers decide straightforward applications without waiting for a planning committee slot, and confirmed 420 new planners recruited through government-backed programmes—exceeding a 2024 manifesto pledge of 300 [8]. AI Minister Kanishka Narayan said: "This is what putting AI to work for the public looks like in practice. By slashing the time spent on routine planning tasks by more than half, we're helping families get quicker answers while freeing up skilled planners to focus on the major developments that will get Britain building." [8] DeepMind's Owen Larter said the tool frees officers "to spend more time deciding on applications and less time on administrative paperwork" [8].