Google DeepMind’s PlanAssist is expanding from a three-council pilot to a 12-month beta involving 15 English councils. The AI tool brings together application documents, planning policies and site constraints, then prepares cited draft assessments for planning officers to review.
The UK government’s announcement, published on October 30, 2026, reports a 58% reduction in time spent on key planning tasks during early testing in Camden, Barnet and Dorset. A Google DeepMind post dated October 31 identifies those tasks more specifically as householder application assessments.
That distinction matters. Less officer time spent assessing an application is a useful operational gain, but it is a different measurement from the number of weeks an applicant waits for a decision. The larger beta will test the tool on live applications across more councils, with officers retaining the final say.
How PlanAssist Turns Documents Into Draft Assessments
PlanAssist was developed by Google DeepMind with the UK government, Google Cloud, Faculty and local authorities. Google says it is built with Gemini and supports the processing of routine householder applications, including proposals such as extensions, conservatories and loft conversions.
According to DeepMind’s workflow description, the tool works through four main stages:
- Application validation: It checks documents and architectural drawings to identify whether the required information has been submitted.
- Policy and constraint checks: It examines the proposal against local and national planning policies and relevant site constraints, including protections such as tree preservation orders.
- Previous issues and objections: It checks whether concerns have already been raised, including objections from local residents.
- Draft reporting: It produces cited assessment reports that planning officers can edit before reaching a decision.
The practical benefit is bringing several kinds of evidence into one review process. An officer assessing an extension may need to connect the submitted drawings with the applicable policies, site protections and consultation responses. PlanAssist is intended to reduce the time spent finding and assembling that material.
The citations are consequential. They give an officer a route back to the evidence behind a statement in the draft. But a cited report still needs scrutiny: a source may support a sentence while the assessment misses another relevant document or constraint.
Google describes the intended approval boundary as “every assessment reviewed and approved by a qualified planning officer before any decision is made.” That puts the system in an assistance role, with the assessment and decision still subject to professional judgment.
What the 58% Figure Actually Measures
The announcements contain three different performance claims. They should be read separately because they measure different parts of the planning process.
| Reported result | Source and scope | How to interpret it |
|---|---|---|
| 58% less time on key planning tasks | The government’s description of early testing; DeepMind specifies average officer time spent on householder assessment tasks | A reported reduction in assessment work, not a 58% reduction in the full application timetable |
| More than 70% of generated reports approved without edits | DeepMind’s account of the three-council pilot | An indication of how often officers accepted drafts unchanged, not an independently measured accuracy rate |
| Routine householder processing reduced from eight weeks to four | The government says early testing helped achieve this reduction | A broader processing-time claim, with insufficient published detail to isolate PlanAssist’s contribution |
The government release and DeepMind post do not provide sample sizes, a detailed task breakdown, error rates or an independent audit. They also do not provide enough comparative detail to establish how much of the reported eight-to-four-week change came from the AI tool. These are early results reported by project participants.
DeepMind’s wording also varies slightly: its introductory summary says “up to 58%,” while the body describes an average reduction of 58% in householder assessment task time. The body gives the more specific scope, but a published evaluation would help clarify the calculation and variation between cases.
I’d treat the no-edit rate as a workflow measure. It suggests many drafts required no rewriting by the reviewing officers. It cannot, by itself, show how often the system missed relevant information or whether an unchanged report was correct.
The timing distinction is equally important. An officer may finish an assessment faster while an application still waits for consultation responses, amendments or another procedural step. Planning Geek’s analysis highlights those other sources of delay when examining the government’s claims.
The useful question for the beta is therefore whether reduced assessment work also shortens the full application process, and under which conditions.
The 15-Council Beta Is a Planned Expansion
The initial pilot has run in Barnet, Camden and Dorset since June, according to DeepMind. The next phase involves live applications over 12 months, with 12 additional councils being onboarded in three cohorts.
That means the announcement describes a phased expansion to 15 authorities, not evidence that all 15 have already completed testing. DeepMind says the participating councils include metropolitan and rural districts, giving the project a broader range of local planning settings.
That variety is useful because an assessment tool must work with each authority’s policies and records. A time saving in one council’s workflow may not transfer unchanged to another.
The government intends to make PlanAssist available to every council in England by the end of 2027. DeepMind explicitly makes wider availability subject to successful testing. Planning Geek also reports that adoption will not be mandatory.
Availability, adoption and demonstrated improvement are separate milestones. A national offer would let councils consider the tool; it would not mean every authority had deployed it or achieved the pilot’s reported savings.
Why Traceability Is the More Demanding Test
The larger beta creates an opportunity to assess whether the time savings survive careful review. A quick draft has limited value if checking and correcting it consumes the time saved elsewhere.
For planning officers, the practical checks follow directly from the workflow:
- Policy currency: Is the draft using the applicable policy version?
- Constraint coverage: Has it identified the relevant protections affecting the site?
- Has it accounted for the submitted drawings and supporting material?
Frequently Asked Questions
3 questions
1Does PlanAssist Decide Planning Applications?
No. It checks application material and drafts cited assessments. Qualified planning officers must review the assessments and retain responsibility for decisions.
