ADD2869 Use of Plan AI to support London Plan consultation analysis

1.1.    The Mayor is required to prepare, publish and keep under review the Spatial Development Strategy for Greater London (the London Plan); and to carry out appropriate consultations as part of that process. The statutory consultation on the draft London Plan will run for 13 weeks from 16 July to 15 October 2026, as approved by Mayoral Decision MD3515. 

1.2.    The Town and Country Planning (London Spatial Development Strategy) Regulations 2000 require representations to be made in writing and be addressed to the Mayor. Once the consultation closes, all representations must be made available for public inspection at the principal office of the GLA. Once the Secretary of State has appointed the examinational panel, a copy of all representations should be sent to the panel (as soon as practicable). All representations are expected to be published on the examination website as part of the examination library.

1.3.    A high volume of representations is expected from a wide range of stakeholders. These representations will come via the online survey form and also as emails, PDF documents and Word documents. Responses must be processed in accordance with UK data protection legislation and the GLA’s privacy notice. The frequently asked questions, published alongside the consultation, state that the GLA may use digital tools, including AI-assisted analysis, to help process, organise and analyse consultation responses. It also states that any use of such tools will be subject to appropriate safeguards and contractual arrangements, including data protection requirements.

1.4.    Plan AI is a large language model-based tool developed at the University of Liverpool, to support local authorities summarise and analyse responses to public consultations for planning documents. It was initially developed in partnership with the Greater Cambridge Shared Planning Service (GCSP) and was supported by funding from the Ministry of Housing, Communities and Local Government (MHCLG) Prop Tech Innovation Fund. On average, GCSP Local Plan consultations generated around 9,500 comments from approximately 3,000 individuals and organisations. The model was trained on 15 years of historical consultation data, including over 55,000 representations and 40,000 planning-specific terms and acronyms. GCSP continues to use the tool.

1.5.    MHCLG provided further funding to stress-test the tool with five more local planning authorities, including the Royal Borough of Greenwich which used the tool for its Regulation 18 Local Plan consultation. The tool works by taking each consultation submission and recognising which comments (or parts of comments) relate to different policy areas. It then reviews all the representations together and can produce summaries by chapter, policy or site allocation, drawing out relevant insights. The tool produces two reports:

•    A catalogue of summaries for all representations received
•    A separate analysis document that provides an insight into the overall character of the consultation.

1.6.    The tool handles the time-consuming task of grouping and summarising consultation comments. This means officers can spend more time reading and analysing feedback. While planning officers still read every representation and comment, use of the PlanAI tool saves a significant amount of officer time on grouping, allocating and summarising representations. It also supports a consistent, evidence-based assessment of the key issues raised. The planning officers remain responsible for reviewing all representations and exercising professional judgment. 

1.7.    This decision form seeks approval to spend up to £25,000 on procuring University of Liverpool to support the initial consultation analysis of representations received on the draft London Plan consultation using their Plan AI digital planning tool. The PlanAI tool was developed specifically by the University of Liverpool for use by the public sector and it is the only provider in the market offering a bespoke digital planning tool of this nature. 
 

2.1.    Use of the PlanAI tool will support efficient and consistent initial analysis of representations received through the Draft London Plan consultation. Its use is expected to save a significant amount of time and ensure best use of officer resource. 

2.2.    This work will provide the following outputs:

•    Data assembly and redaction: University of Liverpool will work collaboratively with GLA officers to assemble the full dataset of submissions, with each respondent and response assigned a unique identifier. Automated redactions will be applied to text-based responses where personally identifiable details are included in the response, with human checking of redaction before responses are published.

•    Catalogue of summaries by response and by policy: A summary of every response made to the consultation will be produced. PlanAI will also split every response into constituent ‘representations’ (the individual salient points that every respondent makes) and associate each of these with the corresponding part of the plan. This will produce a summary of representations for each policy of the Draft London Plan. 

•    Consultation analysis: PlanAI can also produce some analysis of the consultation as a whole. This would give an overview of the character of responses made at aggregate level, relative to an agreed and AI- appropriate structure. It would also summarise respondents’ views on parts of the policy document where representations have been made and explore the sentiment of those representations.  For example, PlanAI can identify those responses which are positive, negative or neutral in tone. 

2.3.    The consultation analysis would primarily be used for internal purposes in the first instance. Although the Regulations do not require a consultation report, the examination panel is likely to request a summary of the main issues raised in representations. Depending on the nature/format of summaries requested by the panel, the PlanAI tool may also be used to support this. 
 

3.1.    Under s149 of the Equality Act 2010), as a public authority the Mayor/GLA must have due regard to the need to eliminate unlawful discrimination, harassment and victimisation, and to advance equality of opportunity and foster good relations between people who share a protected characteristic and those who do not. Protected characteristics under the Equality Act are age, disability, gender re-assignment, pregnancy and maternity, race, religion or belief, sex, sexual orientation, and marriage or civil partnership status (the duty in respect of this last characteristic is to eliminate unlawful discrimination only).  

3.2.    No equality impacts have been identified arising from the use of the PlanAI tool to support consultation analysis. 
 

Key risks and issues

4.1.    The PlanAI tool was developed primarily to support Local Plan consultations. It has not yet been tested on an SDS such as the London Plan. However, it was initially developed by the GCSP which covers two local authority areas, and has subsequently been stress-tested by the Liverpool City Region Combined Authority, which covers six local authority areas. The evaluation and testing carried out by University of Liverpool, to date, has applied PlanAI to various contexts to ensure it is robust and fit for purpose at different stages and scales of plan-making. As set out above, planning officers also remain responsible for reviewing all representations and exercising professional judgement. 

4.2.    The extensive testing and refinement of the PlanAI tool – combined with the University of Liverpool’s established procedures and protocols for project management, quality assurance and ethics – minimise the risk around using the tool at SDS scale for the first time. In accordance with the University of Liverpool’s ethics procedures, data generated through the project would be stored in password-protected files and locked storage facilities that can only be accessed by relevant members of the project team.  

Links to Mayoral delivery plans and other strategies and priorities

4.3.    The London Plan is the overarching document that integrates the spatial aspects of the other statutory Mayoral strategies. In developing the Draft London Plan for consultation, GLA officers have had regard to the Mayor’s statutory strategies, to ensure alignment and consistency across related areas. These include: 

•    transport
•    housing
•    environment
•    economic development
•    health priorities. 

4.4.    The draft London Plan is a core workstream in the Making Best Use of Land Delivery Plan. The draft London Plan also reflects the priorities set out in the London Growth Plan. This was jointly developed by the GLA and London Councils, with respect to its second priority (after housing delivery) of supporting economic growth priority. 

Consultations and impact assessment

4.5.    Consultation on the draft London Plan was approved by MD3515. An Integrated Impact Assessment (including the Equalities Impact Assessment) and Habitats Regulation Assessment were carried out for the draft London Plan and were published as part of the consultation. 

Conflicts of Interest

4.6.    There are no conflicts of interest to declare from those involved in the drafting and/or clearance of this decision form. 
 

5.1.    Approval is requested for expenditure of up to £25,000 to support the initial analysis of representations received on the draft London Plan. This expenditure will occur in financial year 2026-27. The proposed spending is financed from reserves (revenue grants unapplied). This is a planned use of reserves as set out in the GLA Mayor budget for the 2026-27 financial year.

6.1.    Subject to approval of the required budget, the University of Liverpool would be appointed in September 2026, following the procedures in the Contracts and Funding Code for procurement under £25,000 (excluding VAT). Data assembly and redaction are anticipated to take place in October and early November 2026, with initial outputs related to summaries and analysis are expected November / early December 2026. The project is expected to complete with all final outputs provided by the end of December 2026.