Artificial intelligence is moving from the technology sector into one of the capital’s most traditional public services: planning.
From analysing thousands of public responses to the draft London Plan to helping councils process routine planning applications, AI is increasingly being tested as a way to reduce paperwork and speed up decisions. But as the technology enters a system that determines how London’s homes, transport, workplaces and neighbourhoods develop, officials are also asking a difficult question: how much decision-making should be left to machines?
The issue came under fresh scrutiny at City Hall this month. The London Assembly’s Planning and Regeneration Committee held an investigation on 1 October 2026 into how AI is being used in planning and what role the Greater London Authority should play in guiding its adoption. The committee’s investigation comes as government-backed AI tools are expected to save planning authorities around 250,000 hours of manual work every year, while nine in ten local planning authorities have reported receiving planning submissions generated or assisted by AI.
London is already using AI to analyse planning responses
The Greater London Authority has already taken a practical step.
On 7 September 2026, the GLA approved spending of up to £25,000 to procure PlanAI, a digital planning tool developed by the University of Liverpool. The tool will help analyse representations submitted during consultation on the draft London Plan, whose statutory consultation runs for 13 weeks, from 16 July to 15 October 2026.
The scale of the task explains why AI is attractive to planners. The GLA expects a large volume of responses arriving through online forms, emails, PDFs and Word documents. PlanAI can group comments according to policy areas, produce summaries and identify whether representations are broadly positive, negative or neutral.
The technology is not being given the final say. According to the GLA’s decision document, planning officers will continue to read representations and exercise professional judgement. The tool is intended to reduce the time spent grouping, allocating and summarising responses rather than replacing planners.
PlanAI has also been trained using 15 years of historical consultation data, including more than 55,000 representations and 40,000 planning-specific terms and acronyms. Earlier consultations using the system generated an average of around 9,500 comments from approximately 3,000 individuals and organisations.
AI could also speed up planning applications
London is part of a wider national experiment.
In June, the UK Government announced an AI prototype designed to reduce the processing time for routine householder planning applications from an average of eight weeks to four weeks. Early testing is taking place with Barnet, Camden and Dorset councils.
The system can triage applications, summarise information and provide planning officers with an initial assessment. However, a qualified planning officer must review and approve every assessment before a decision is made. The government has said that, subject to successful testing, national rollout is planned from 2027.
Householder applications account for nearly 70% of planning applications each year, making them a significant part of the workload facing local planning authorities. Separately, the government’s Extract AI tool is available to councils across England and converts older planning documents and maps into usable digital data.
The government estimates that Extract could eliminate around 250,000 hours of manual document-checking work each year. Trials involving 20 local planning authorities found that the average council could save approximately 255 hours of manual work processing planning documents.
Officials say AI is meant to assist — not replace — planners
The government has repeatedly stressed that AI is being introduced as an assistant rather than an autonomous decision-maker.
Minister for Data and Modern Digital Government Ian Murray said the technology was intended to take administrative work away from planning officers so they could concentrate on professional tasks.
“This isn’t about replacing the expertise and judgement of planning professionals.”
Housing and Planning Minister Matthew Pennycook described the existing system as heavily dependent on paper-based processes and said AI could streamline planning applications and reduce delays.
The emphasis on human oversight is also reflected by the Planning Inspectorate. Its interim chief executive Graham Stallwood said its AI guidance supports “human control and oversight”, describing this as a “golden rule” for the use of AI in planning.
But AI-generated public objections are creating a new problem
The technology is not only being used by planning authorities. It is increasingly being used by the public.
The London Assembly has highlighted the growing use of AI in planning consultations, while national planning guidance warns that councils are receiving AI-generated representations that can be unnecessarily long, repetitive, irrelevant or contain fabricated references to planning policy or case law.
That creates a new challenge for planners. If hundreds of residents use AI to generate similar objections or supporting statements, the volume of submissions could rise without necessarily representing a corresponding increase in unique public views.
For London, this makes the question of AI more complicated than simply whether the technology can save time.
The bigger question: who controls London’s digital planning future?
London’s experiment with AI comes at a time when the capital faces pressure to build more homes while dealing with complex questions around transport, environmental policy, economic development and the character of local communities.
The GLA itself acknowledges that questions remain about transparency, accountability, skills, resources and public trust as AI becomes more involved in planning.
There are also practical safeguards. The GLA says personal information in consultation responses processed through PlanAI will undergo automated redaction followed by human checking. The resulting data will be handled under data-protection requirements and stored with restricted access.
For now, London’s approach is therefore not about handing planning decisions to an algorithm. It is about using AI to handle some of the work surrounding those decisions.
But as the technology becomes more capable, the distinction between assisting a planner and influencing a planning decision could become increasingly important.
London is now testing where that line should be drawn.
