Part of the AI Regulation News hub.
The Planning Inspectorate has updated its non-binding guidance on declaring AI use in casework evidence, asking each expert who intentionally used AI for their own declaration and telling local planning authorities that AI summaries do not replace original documents
The costs warning is not new; it was already on the page in the February version. What changed is who has to declare, what they have to point to, and what they no longer have to name.
Bottom line: Non-binding guidance. The Planning Inspectorate updated it on 23 September 2026; it is not legislation and sets no commencement date or deadline, though it warns that improper AI use could be treated as unreasonable behaviour open to an award of costs.
Who this affects: Planning agents and consultants, planning solicitors and counsel, local planning authority officers who prepare appeal questionnaires, expert witnesses such as chartered surveyors and archaeologists, and consenting teams on Nationally Significant Infrastructure Projects.
Issue date: Updated 23 September 2026. The guidance was first published 6 September 2024 and previously updated 20 February 2026.
What changed: Against an archived copy of the February text, the update limits disclosure to intentional AI use, drops the request to name the tool, adds paragraph or section locators, and adds separate text for experts, LPAs and applicants.
Analysis: On our reading the update moves the declaration away from which product was used and toward where AI touched the evidence and who stands behind each expert document. That is our characterisation; the page does not describe its own changes.
Primary sources: Use of artificial intelligence in casework evidence (Planning Inspectorate, GOV.UK)
- Instrument (EN)
- Use of artificial intelligence in casework evidence
- Authority
- Planning Inspectorate
- Jurisdiction
- England and Wales, per the page's own banner. It covers appeals, applications and examinations dealt with by the Planning Inspectorate
- Status
- Guidance. First published 6 September 2024, updated 20 February 2026, most recently updated 23 September 2026
- Bindingness
- Non-binding guidance. It is not legislation. It warns that improper AI use could be found to be unreasonable behaviour open to an award of costs
- Issue date / next deadline
- Updated 23 September 2026. No commencement date, transition period or compliance deadline is stated
- Document
- GOV.UK guidance page. The update history entry for 23 September 2026 reads: Update to guidance on how to declare the use of AI in casework submissions
- Primary source
- https://www.gov.uk/guidance/use-of-artificial-intelligence-in-casework-evidence
What does the updated guidance ask for?
This is guidance, not legislation, and the page says so in its own classification. Within that frame it asks anyone sending material to the Inspectorate to say so if they have "intentionally used AI" to draft or substantially rewrite text, produce a summary or analysis, generate or alter images or videos, or create content that goes beyond straightforward formatting or presentation.
The declaration is meant to be short. The page says it can sit in a covering email or letter, or form part of a statement or proof of evidence, and that it should state that AI was used, what it was used for and how, and what checks were made, with an acceptance of responsibility for accuracy.
Where AI was used for a specific part of a document, the guidance says this should be made clear "such as by reference to paragraph numbers, sections or appendices". Those are offered as examples of how to point to the affected part. They are not a prescribed format.
Routine tools stay outside it. The guidance lists spellchecking, grammar suggestions, formatting tools and accessibility functions, and says the Inspectorate is seeking information on "deliberate and substantive use of AI tools to create or materially alter the content of a submission".
What is different from the February 2026 text?
The page's update history does not list the changes, so we compared the live page with an Internet Archive capture dated 19 August 2026, which showed the page as last updated 20 February 2026. Every difference below is drawn from that comparison, and the current text remains guidance rather than law.
The February text asked people to state "Which tool you used", giving CoPilot, chatGPT and midjourney as examples. That request is absent from the 23 September version, and the old interested-party example that named Microsoft CoPilot has been replaced.
The trigger has narrowed. February asked people to tell the Inspectorate if they had "used AI" for the listed tasks; the current page adds the word intentionally. The routine-tools exclusion was already there, but it now carries an explanation that AI features are embedded in everyday software and that users often cannot tell where they have operated.
Several passages have no counterpart in the archived capture: a statement that all parties, not only professional parties, are expected to take responsibility for accuracy and lawfulness; the multi-expert paragraph; the sections for LPAs and for appellants and applicants; a caution against using AI to create cartoons or exaggerated illustrations of a proposal; a note that an Inspector may seek clarification; and a model declaration from a planning agent.
The data protection line also changed. February said: "Do not enter sensitive or personal information into public AI tools." The current page asks users to check that they have met their data protection responsibilities before entering sensitive or personal information into AI tools. On our reading that is less absolute than the earlier wording.
Experts, agents and multi-author cases
The new professional-parties text addresses cases where several experts produce separate documents, giving a chartered surveyor's viability study and a qualified archaeologist's field survey as examples. In that situation the guidance says "you must produce your own declaration when AI has been intentionally used".
Two limits sit inside that sentence. It is conditional on intentional AI use, so it does not ask an expert who used none to declare anything. And the page says a catch all declaration is "unlikely to provide the level of information required", which falls short of a prohibition.
The planning-agent example is worth reading closely because it shows what the Inspectorate thinks a full declaration looks like. It lists the tasks AI assisted with, states that AI "was not used to formulate professional planning judgements on the merits of the proposal", and confirms that quotes are genuine, unedited extracts. It is an example. The guidance does not make any of its wording compulsory.
Local planning authorities and AI summaries
This section is new against the archived text, and like the rest of the page it is guidance. It says that if an LPA has used AI tools this needs to be clearly signposted, and that an LPA using AI as a summarisation tool should still send PINS the original documents.
The page gives a concrete instance: providing only a summary of representations as part of the appeal questionnaire "is not acceptable". It explains that summaries alone "may not comply with the relevant Regulations" and risk depriving an Inspector of material considerations omitted from the summary. The page does not name the Regulations in that passage.
The same advice is extended to appellants and applicants who have used AI tools to summarise formal consultation responses, with the examination of a National Infrastructure project given as an example.
The costs warning was already there
The sentence most likely to be quoted is not new. Both the archived February text and the current page say that improper use "could be at risk of being found as an example of unreasonable behaviour, and open to an award of costs".
That is a conditional warning in guidance. It does not create a costs power, it does not say that a missing declaration will attract costs, and the page cites no case in which costs have been awarded for AI use. We found none in the material we read and did not search for one.
What we did not verify
What we opened: the live GOV.UK page, fetched on 24 September 2026 UTC and read end to end, including its update history, and an Internet Archive capture of the same URL timestamped 19 August 2026 that displays "Last updated 20 February 2026". The comparison in this piece is between those two texts.
What we did not open: the Welsh language version, the procedural guides for planning appeals, local plan examinations and National Infrastructure applications that the page links to, and the Regulations the LPA section refers to without naming. We did not review any Inspector decision or costs decision.
What we refuse to claim: that every difference we list was made on 23 September 2026, because the page could have changed between the 19 August capture and the 23 September update without a separate history entry; that the guidance binds anyone or creates a new costs power; that any expert is expected to declare without having used AI; and that a catch all declaration is prohibited. Quotations are reproduced with ASCII punctuation in place of the page's typographic quotation marks.
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
If your firm's AI declaration template still names the tool and says little else, it was written for the February page. The 23 September guidance, which remains guidance, points instead to intentional use, the specific paragraphs or sections affected, the checks made, and a separate declaration from each expert who used AI.
Source File
https://www.gov.uk/guidance/use-of-artificial-intelligence-in-casework-evidence
Open the GOV.UK page and check three things: the 23 September 2026 entry under Updates to this page, the multi-expert paragraph under The role of parties, Professional parties, and the LPA paragraph on AI summarisation and original documents.
In such cases, you must produce your own declaration when AI has been intentionally used. ยท Use of artificial intelligence in casework evidence, The role of parties, Professional parties, 23 September 2026
FAQ
Is the Planning Inspectorate's AI guidance legally binding?
No. It is guidance published on GOV.UK. It sets no deadline and creates no new legal duty, although it warns that improper AI use could be found to be unreasonable behaviour open to an award of costs.
Do I still have to name the AI tool I used?
The current page no longer asks for it. The February 2026 text, as archived on 19 August 2026, asked people to state which tool they used; the 23 September version asks what AI was used for, how, which parts it affected, and what checks were made.
Does every expert on an appeal need an AI declaration?
Only where AI has been intentionally used. The guidance asks each such expert for their own declaration and says a catch all declaration is unlikely to give enough information.
Is using spellcheck or grammar tools covered?
No. The guidance excludes spellchecking, grammar suggestions, formatting tools and accessibility functions, and says it is concerned with deliberate and substantive use of AI to create or materially alter content.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.