A tool called PlanAssist, which Google DeepMind built with Gemini for English councils, is moving from a trial in Barnet, Camden and Dorset to a 12-month test on live applications in 15 councils. It reads a householder application, checks it against local and national planning policy, and drafts a report that an officer edits and signs. The officer makes every decision. If the test goes well, the government plans to offer it to every council in England in 2027. Councils in Scotland, Wales and Northern Ireland are not part of it.
What the tool takes over
It does the work in four steps. It checks that the drawings and documents the council requires have all been sent. It checks the plan against constraints such as protected trees and heritage rules. It looks for objections already raised, for example by neighbours. Then it writes a cited report. In the trial, officers approved more than 70% of those reports without changing anything, and the time they spent on these tasks fell by up to 58%.
This is checking work, and it was always the bulk of the job. Most planning applications are granted, so for a typical extension the officer's question is rarely whether to say yes. It is whether everything has been checked properly. A tool that does the checking and shows its sources turns that into a review of a finished draft.
Saved hours are not saved weeks
A family's wait has several parts: the days before an officer picks the file up, the time neighbours have to comment, any wait for a committee meeting, and the officer's own work. The tool shrinks only the last one. The same weekend the trial widened, England's new national rules began sending straightforward applications to officers instead of to a committee, which removes another part of the wait. The government's own line goes further and talks of routine householder decisions falling from eight weeks to four. That follows only if the freed hours are spent on the queue.
The queue is where councils already differ most. Under the same law, Horsham decided 72% of householder applications within eight weeks without needing an extension in the year to June 2026, while Ealing managed 43%. I expect the tool to matter most in places like Ealing, where officer time is the thing running short, and least where the queue is short already. A council that uses the saved hours to cover vacancies will look busy and fair, and its applicants will wait as long as before.
What a family or a small builder can do now
The first thing the tool does is something an applicant can do before sending anything. Every submission is checked by hand for completeness, and one with a missing drawing or the wrong fee goes back to the sender, so the clock starts late. A homeowner or a small builder could give an AI assistant the council's published list of required documents and their own drawings, and ask what is missing or inconsistent. A mismatch between the floor plan and the elevation drawing, or a missing site plan, is the kind of thing it finds in minutes. The council still decides, but the application arrives complete the first time.
The same pattern fits any office where most cases are routine approvals, such as licences, building permits and benefit claims. AI turns checking into spare capacity, and the office then chooses what to buy with it.
So the number to watch in the 15 councils is how many householder applications are decided within the statutory eight weeks, not how many officer hours were saved. If that share rises, the hours went into the queue. If it stays flat, they went somewhere else.