The Greater Cambridge Shared Planning (GCSP) service, a partnership between South Cambridgeshire District and Cambridge City councils, has embarked on a journey to integrate artificial intelligence (AI) into its planning consultation processes.
Background and Challenge
The Greater Cambridge Shared Planning (GCSP) service, a partnership between South Cambridgeshire District and Cambridge City councils, has embarked on a journey to integrate artificial intelligence (AI) into its planning consultation processes. With the growing complexity and volume of public representations during Local Plan consultations, GCSP encountered a significant administrative burden. On average, each consultation generated around 9,500 comments from approximately 3,000 individuals and organisations. Although summarising these comments did not require planning expertise, the task was both time-consuming and resource intensive.
GCSP aimed to explore how AI could reduce administrative burdens, improve the quality of data analysis, and ultimately enhance public engagement in the planning process. This ambition was made possible through support from the Ministry of Housing, Communities and Local Government’s (MHCLG) PropTech Innovation Fund. This is a government initiative designed to accelerate the adoption of digital tools in planning and streamline the plan-making process. By securing funding through this programme, GCSP gained essential financial backing and became part of a wider collaborative network of local authorities focused on digital innovation in planning.
AI Integration Approach
The project has evolved through multiple stages over the past 18 months. Initially, the focus was on exploring how improved data standards and back-office processes could accelerate the registration and processing of Local Plan representations. This began with a comprehensive mapping of the end-to-end consultation process and a data discovery phase, which led to enhancements in the Opus Consult platform, provided by JDi Solutions. Key improvements included transitioning from Excel-based workflows to a more integrated system for managing representations and enabling stakeholders to submit responses directly through the platform rather than via disparate channels such as email.
As the project progressed, the team identified further opportunities to increase efficiency through the use of AI. This led to a competitive procurement process, resulting in a partnership with the University of Liverpool to develop a bespoke large language model (LLM). The academic route was deliberately chosen to ensure the project was grounded in ethical best practices, transparency, and a research-led methodology tailored to the specific language and context of planning.
The LLM’s training and output
To ensure accuracy and rigour in the LLM’s development, the model was trained on 15 years of historical consultation data, including over 55,000 representations and 40,000 planning-specific terms and acronyms. The goal was to create a tool that could replicate the summarisation work of planning officers, while also generating richer insights.
The AI tool produces two key outputs: individual summaries and an executive report. The individual summaries provide a concise overview of each submission, capturing the core issues raised. The executive report is a comprehensive document that identifies key themes, categorises support and objections, links comments to specific policies or chapters, and includes geographic analysis of where responses originated. This allows the planning team to identify underrepresented areas and tailor future engagement strategies accordingly.
Initial impact
Over several months, the AI tool has undergone continuous refinement and iteration to enhance its performance and reliability. It was soft-launched during the winter of 2024–25 across three supplementary planning document (SPD) consultations. The results were highly encouraging: a summarising task that previously required 18.5 hours of officer time was completed by the AI model in just 16 minutes. This represented a time saving of over 60%.
Given that typical Local Plan consultations generate approximately 9,500 comments and can take up to 450 officer days to process, the potential impact is substantial. GCSP anticipates that the LLM will be able to complete the same task in under an hour, at a cost of just £1–£3 per run based on testing and the evaluation process.
While planning officers still read every representation or comment, the time saved by not having to produce a summary can significantly free up their time. This allows them to focus on higher-value tasks where their experience is better spent, such as analysing public feedback, engaging with stakeholders, and progressing plan-making activities.
Early findings also suggest that the use of AI may be contributing to improved public participation in consultations, by enabling faster feedback loops and more accessible reporting.
Public Engagement & Transparency
At the core of this project is a strong commitment to engaging with local residents, businesses, community organisations, and other key stakeholders. This collaborative approach ensures that the development plans for Greater Cambridge genuinely reflect the needs, values, and aspirations of the community.
As part of the soft launch, two focus groups were convened to ensure that the use of AI was clearly communicated and that any concerns could be surfaced and addressed early in the process. One group consisted of members of the public, while the other brought together key planning stakeholders.
Feedback from these sessions played a critical role in shaping both the development of the AI tool and how GCSP prepares for future engagement. Participants raised important questions around legal robustness, data privacy, and transparency. In response, GCSP developed a comprehensive and publicly accessible FAQ document , designed to explain how the AI tool works, how data is handled, and what safeguards are in place to ensure AI is used safely and responsibly.
The focus groups also inspired further enhancements to the tool itself. One key development currently underway is the introduction of an instant summarisation capability, which will allow respondents to generate and validate summaries of their own submissions in real time. This functionality is being developed by the University of Liverpool in collaboration with JDi Solutions and is scheduled for testing during the summer 2025 consultation on the Planning Obligations SPD.
Ethical and Environmental Considerations
Ethics have been at the heart of this project from the very beginning. The University of Liverpool contributed not only technical expertise but also a research-led approach that placed ethical considerations front and centre. From the outset, the team committed to exploring the ethical dimensions of AI use, ensuring transparency and fostering trust through active engagement.
They posed critical questions such as: How do people feel about AI summarising their comments? What are the implications for democratic accountability? How can transparency be upheld in an automated process? These questions informed the design of focus groups and continue to guide ongoing engagement efforts as the team works to build public confidence in the use of AI within the shared planning service.
To strengthen the ethical foundation of the project, the team partnered with ai@cam at the University of Cambridge and Anglia Ruskin University for peer review and expert guidance.
In addition to addressing ethical concerns and feedback from focus groups, the project also tackled questions about the environmental impact of AI. The team investigated the energy consumption of their LLM, finding that each document summarised by the AI generates approximately 4.32 grams of CO₂ emissions, roughly equivalent to sending two emails, growing three strawberries, streaming a 10-second video, or boiling half a cup of water.
The environmental footprint of training, operating, and maintaining large language models is significant, involving both electricity use and water consumption. For context, the training of OpenAI’s GPT-3 model was estimated to consume enough energy to power 1,000 households for an entire year.
Future Plans
Following the successful pilot, GCSP is preparing to fully deploy the AI tool during the upcoming consultation on the draft Greater Cambridge Local Plan. In parallel, the team is finalising the rollout of the instant summarisation feature, tested in summer 2025, which is designed to further improve transparency and give users greater control over how their input is processed.
Looking ahead, GCSP is actively engaging with government bodies to explore opportunities for scaling the tool across other local authorities. The goal is to develop a low-cost, flexible solution that can be easily integrated into diverse planning systems and contexts, delivering benefits across the wider local government sector.
Meanwhile, the University of Liverpool is investigating how the tool could be adapted to support development management, particularly in summarising public representations on planning applications. With GCSP handling approximately 7,000 applications each year, this extension has the potential to unlock significant efficiencies and further streamline planning processes.