Google DeepMind's PlanAssist goes to 15 UK councils after a 58% time cut

Google DeepMind's PlanAssist, a planning tool built with Gemini, is moving out of its pilot. It was developed with the UK government, Google Cloud and Faculty, and it is now heading to 12 more councils in England, for 15 in total, in a 12-month beta on live applications.

What it does. PlanAssist works on routine householder applications: loft conversions, conservatories and extensions. It checks that the application documents and architectural drawings include the required information. It checks the plans against heritage constraints, such as tree preservation orders, and against local and national planning policies. It looks at whether issues or objections were raised before, for example by residents. Then it drafts a fully cited report that the planning officer can edit. Officers keep the final say on every application.

The pilot numbers. The tool has been piloted since June at Barnet, Camden and Dorset councils. According to Google DeepMind and the government, average officer time on householder assessment tasks fell by 58% (DeepMind's headline says "up to 58%"), and officers approved more than 70% of the tool's reports with no edits.

Why householder applications. DeepMind says routine householder applications make up to 70% of planning officers' workloads. GOV.UK puts it as nearly 70% of applications each year.

What happens next. The 12 new councils will join in three cohorts and include metropolitan and rural districts. Subject to successful testing, the government plans to make PlanAssist available to all councils in England by the end of 2027. The government also says the tool is helping halve processing times for routine householder applications, from eight weeks to four. That is the government's claim. The wider context is the UK target of 1.5 million new homes by 2029.

Who it's for. Council planning officers in England. It is not for homeowners or the public, and it is not a product you can sign up for.

The catch. All of these figures come from the makers and the government, based on a three-council pilot, not an independent study. The 58% measures officer time on assessment tasks, not the time it takes to reach a decision. The 12 new councils have not been named. MODEL has not tested PlanAssist hands-on.

Our take

This is a narrow, sensible job for an AI model: checking paperwork against rules and drafting a cited report that a person signs off. Keeping officers in charge of every decision matters. The beta on live applications at 15 councils is the real test, and the number to watch is whether decision times actually fall, not just time spent on assessment.

This summary is based on the Google DeepMind blog post and the GOV.UK announcement, read October 31, 2026. Not sponsored. No affiliate links.