Using AI in planning: the rules, and how to stay on the right side of them
The RTPI has published its first Practice Advice Note on AI. The Planning Inspectorate now expects anyone submitting AI-assisted evidence to disclose the tool, its purpose, and the checks carried out — and warns that improper use can amount to unreasonable behaviour with costs consequences. Here is what the rules actually say, where AI goes wrong in planning work, and how to use it safely.
The regulatory landscape (as at mid-2026)
- RTPI Practice Advice Note on AI (2025) — the Institute's first formal advice: AI can genuinely help with large datasets, impact modelling, consultation analysis and drafting, but members are warned about inaccurate data, copyright and security. It deliberately does not prescribe — how, whether and when to use AI remains each chartered planner's professional judgement, and responsibility (and liability) stays with the individual.
- Planning Inspectorate guidance on AI in casework evidence (updated February 2026) — AI may support an appeal, application or examination only where its use is transparent: say that you used AI, name the tool, say what you used it for, and confirm the checks you carried out. The Inspectorate warns that AI can produce fake representations, evidence, text and images, and that improper use could amount to unreasonable behaviour and an award of costs.
- No single AI law — AI use in planning sits under existing law: UK GDPR (Articles 13–15 on automated decisions), the Data (Use and Access) Act, the Equality Act 2010, copyright, and planning law itself. The modern safeguard for automated decisions is a person's right to meaningful human review — to question, challenge and have a decision reviewed by a human.
- Government is adopting AI on the authority side — Extract is digitising legacy planning records, the Augmented Planning Decisions project is exploring AI-assisted determination of householder and similar applications (roughly 69% of all submissions), and consultation-summarising tools are in use. Applicant-side evidence standards will matter more, not less.
Where AI actually goes wrong in planning work
The consistent, documented failure is invented citations — and it clusters on exactly the material planners lean on most: court judgments, appeal references and policy numbers. A general-purpose chatbot answers from model memory; when memory runs out, it produces a reference that looks entirely convincing and does not exist.
The other named risks: sensitive material (viability figures, option agreements, client identities) uploaded to consumer AI platforms with unclear retention; bias inherited from historic spatial data; and deskilling, where junior planners never learn to do by hand the tasks they hand to a machine. All three are managed by the same discipline: know what the tool was used for, keep the evidence traceable, and keep a human responsible for the output.
How PlanningCore is built for these rules
PlanningCore takes the position that AI in planning is only useful when every material claim is traceable to a real source. Concretely:
- Library-grounded citations. The drafting tools are constrained to cite
from PlanningCore's indexed library of real documents — Planning Inspectorate appeal
decisions, the current National Planning Policy Framework, Planning Practice Guidance
and local plan documents — rather than from model memory. References the system cannot
confirm against that library are marked
[verify reference]instead of being presented as fact. - An AI Use Disclosure on every generated document. Every document the generator produces carries a built-in disclosure block naming the tool and model, what the AI was used for, the citation-provenance constraint, and the checks required before submission — the transparency the Inspectorate's guidance asks for, produced by default rather than remembered at the last minute.
- Measured data, stated sources. The intelligence pages (refusal patterns, officer records, appeal survival rates) compute from public planning records and state the corpus and sample size next to every figure.
- A human stays responsible. Every output is a draft for review by the applicant or their professional adviser. PlanningCore provides planning information and drafting support — not formal planning advice, and never an unreviewed submission.
A practical checklist — using any AI tool in planning work
- Disclose it. If AI touched material going to an LPA or the Inspectorate: name the tool, state what it was used for, and confirm the checks you carried out. Silence is now the risky option.
- Verify every citation against the original source. Case names, appeal references, policy numbers and NPPF references are where AI invents most confidently. If you cannot find the original, the citation does not go in.
- Watch what you upload. Client identities, viability figures and land assembly detail should not go to tools with unclear data retention. Check whether input is used for model training.
- Quote fees and statistics only from dated sources. Planning fees change every April; the NPPF was rewritten in August 2026 and old paragraph numbers are unreliable.
- Keep a human responsible. AI is support, not a substitute for a chartered planner's judgement. The person submitting the document owns it — professionally and, before the Inspectorate, on costs.
Sources
RTPI, An Introduction to Planning and AI (Practice Advice Note, 2025) · RTPI, Code of Professional Conduct · Planning Inspectorate guidance on the use of AI in casework evidence (updated February 2026, on gov.uk) · UK GDPR Articles 13–15 · Data (Use and Access) Act · Equality Act 2010 · Extract and the Augmented Planning Decisions project. Check the originals — the guidance is evolving.