England is expanding two AI tools intended to reduce the paperwork surrounding local planning decisions. Extract converts legacy documents into structured data, while Augmented Planning Decisions assists officers with application review, policy research and report drafting.
The target is administrative delay rather than statutory judgement. Routine household applications make up nearly 70% of submissions, consuming time that could be spent on housing and infrastructure.
Extract unlocks information trapped in planning PDFs
The Ministry of Housing, Communities and Local Government and the government’s Incubator for AI built Extract with Gemini models. Trials across more than 20 authorities have led to availability for every council in England.
The tool parses hundreds of pages of historical records and converts them into usable datasets within minutes. Trial estimates suggest it could remove around 255 hours of manual entry per council each year.
That benefit depends on verification. Old plans may contain poor scans, handwritten notes or inconsistent terminology. Authorities need sampling and exception queues so uncertain extraction does not silently become an official record.
Structured archives can support search and analysis beyond one application, but retention and access rules must remain attached. Making data easier to query should not make sensitive civic information easier to misuse.
APD prepares evidence while officers retain authority
The APD prototype consolidates incoming documents, flags missing information and identifies site data. It can find relevant national and local policies, summarise consultation letters and draft the basis of an evaluation report.
Planning officers review and amend every line before deciding whether to approve or reject. That division is important because local policy interpretation, precedent and public impact require accountable professional judgement.
Policy citations should link directly to their source and version. A fluent explanation can conceal an outdated rule, so officers need to distinguish extracted facts from generated reasoning.
The system records processing steps to support audit. Useful records include inputs, retrieval results, model version, edits and the officer’s final rationale without pretending that hidden model reasoning is a complete legal explanation.
Alpha testing in Barnet, Dorset and Camden gives the tool varied policies and datasets. The government aims to reach more than 300 authorities by 2027 after learning from those environments.
Public-sector scale must preserve local accountability
The government hopes to halve decision timelines and support a target of 1.5 million new homes by 2029. Faster administration can help, but planning outcomes also depend on staffing, consultation and infrastructure constraints outside the model.
Security controls on Google Cloud are intended to protect municipal data and block malicious inputs such as prompt injection. Authorities should still minimise information sent to models and test how poisoned documents affect summaries.
Performance needs breakdown by application type, council and applicant. Average time savings can hide new burdens when complex or poorly digitised cases generate repeated corrections.
Officers and applicants should help evaluate usability, not only technical teams. A summary that saves internal time but obscures a resident’s objection would weaken the legitimacy of the process. Public reporting can compare service time, correction and appeal before and after deployment.
Residents require a clear route to challenge factual errors and understand that an officer—not an algorithm—made the decision. Published governance can explain where AI assisted without exposing personal records.
The collaboration combines policy ownership in government with engineering from public and private teams. Long-term success requires councils to retain knowledge, data access and the ability to operate if a supplier changes.
These tools show a practical form of public-sector AI: converting unstructured records and preparing evidence for a professional. If they reduce clerical work while strengthening traceability, planning officers can spend more attention on the cases where public interest cannot be compressed into a template.
