Google Gemini tool cuts UK planning assessment time by 58%

Automating routine planning

Google DeepMind recently shared results from a pilot program involving PlanAssist, an AI tool built using the Gemini model. The tool was designed to help local planning authorities manage the high volume of householder applications, which include projects like conservatories and loft conversions. According to the company, these routine tasks account for up to 70% of the total workload for planning officers.

The pilot program took place across three UK councils: Barnet, Camden, and Dorset. During this period, the tool helped reduce the average time officers spent on householder assessment tasks by 58%. This means that tasks that previously took an hour could potentially be completed in about 25 minutes using the AI's assistance.

While the tool handles the heavy lifting of data processing, it is specifically designed to support human judgment. Planning officers retain the final say on every single application submitted to the council. The goal is to speed up the administrative side of the process without removing the human oversight required for local governance.

How PlanAssist processes data

The tool follows a structured four-step process to evaluate planning requests. First, it validates the submitted application documents and the architectural drawings provided by the applicant. It then checks the application against specific heritage constraints, such as tree preservation orders, as well as local and national planning policies.

Next, the system checks whether any objections have been raised by the public, such as those from local residents. Once these checks are complete, PlanAssist drafts a "fully cited report" that summarizes the findings. These reports are then presented to the planning officers, who can review, edit, and approve the documentation as needed.

The accuracy of the tool's drafting was a highlight of the initial pilot. The company reported that more than 70% of the reports generated by PlanAssist were approved by officers with no edits required. This suggests that the AI is becoming increasingly reliable at navigating the complex web of local regulations and policy requirements.

Expanding the beta phase

Following the success of the three-council pilot, PlanAssist is moving into a 12-month Beta phase. This next stage involves 12 additional local planning authorities across England, including both metropolitan and rural districts. These councils will be onboarded in three cohorts, with an average of four councils joining each group.

The expansion is part of a broader partnership between Google DeepMind, the UK government, Google Cloud, and Faculty. This collaboration is tied to a specific national objective: the UK government's goal to build 1.5 million new homes by 2029. By automating the routine aspects of planning, the government hopes to clear bottlenecks in the housing development pipeline.

This rollout follows an earlier Gemini-built tool called Extract, which was already made available to every council in England. PlanAssist represents a more specialized application of the technology, moving from simple data extraction to complex report drafting and policy validation for public services.

The path to nationwide use

If the 12-month Beta phase is successful, the government plans to make PlanAssist available to all councils in England by 2027. This would represent a significant scaling of AI in public infrastructure. Every assessment produced by the tool will still be reviewed and approved by a qualified planning officer before any final decision is reached.

The project highlights a shift toward using AI to handle the "routine end" of professional work. By automating the repetitive checks and drafting, the tool allows human officers to focus on more complex cases that require nuanced judgment. This reflects a growing trend in public sector workflows where AI manages the high-volume data while humans remain the final arbiters.

While the pilot results are promising, it is important to note that the figures come directly from Google DeepMind's own reporting. There has not yet been an independent evaluation of the 58% time-saving figure. However, the move toward a nationwide rollout suggests that the government sees the tool as a viable solution for meeting its ambitious housing targets.