DeepMind tests AI for UK planning files

Google DeepMind announced on June 16 an AI prototype built with the UK government to speed up so-called householder planning applications, the routine files submitted by residents for work such as loft conversions or home extensions. The stated goal is to cut decision times for these simpler applications by 50%, while keeping the final decision with public planning officers. Early trials involve Barnet, Camden and Dorset, with the government planning to make the tool available to councils nationally from 2027.

The useful signal is not simply that Gemini is entering another public-sector workflow. The source describes a specific administrative pipeline: extracting information from case files, highlighting missing data, identifying relevant national and local policies, summarizing consultation feedback and drafting an initial report that an officer must review and edit. Householder applications account for nearly 70% of planning applications each year in the UK, which makes them a practical target for reducing repetitive paperwork before applying AI to more complex or contentious projects.

DeepMind emphasizes one important constraint: the tool does not approve or reject applications. The planning officer reviews every line, keeps decision-making authority and can change the reasoning before anything is issued. The prototype is also designed to record its work, creating an audit trail for each case. That detail matters because administrative AI becomes risky when a drafting assistant quietly turns into an opaque decision-maker. In planning, residents need to understand why a file moves forward, stalls or receives conditions. The benchmark is therefore operational as much as technical: fewer hours spent chasing documents, without weakening accountability. For councils, the real value will depend on whether officers can trust the citations and correct the draft quickly.

The project builds on Extract, an earlier tool designed to turn legacy planning documents, often locked in PDFs, into usable data. DeepMind says Extract has already been trialed across more than 20 local planning authorities and is expected to save the average council about 255 hours of manual work each year. The practical takeaway is modest but important: this is not an AI system claiming to replace planners. It is an attempt to make a document-heavy process more searchable, checkable and consistent. If the UK rollout works, it could become a useful reference case for bounded public-sector AI, where the measure of success is not novelty but a faster, more transparent administrative path. That is a quieter story than a model launch, but probably a more revealing one for real deployments.