AI register and risk register for RICS compliance
Sections 3.2 and 3.3 of the RICS AI Standard say that RICS-regulated firms using material-impact AI must keep two written records: an AI register of the systems in use and an AI risk register covering risks, controls, ownership and status. Here is what each record needs to contain, with a worked example you can adapt.
The short answer
The RICS Professional Standard Responsible use of artificial intelligence in surveying practice (1st edition, September 2025, in force from 9 March 2026) puts the responsibility for AI use on the regulated member or firm, not on the software vendor. Where AI has a material impact on your service, sections 3.2 and 3.3 say that a RICS-regulated firm must maintain a small set of written records. Two are central to that file:
- An AI register — the material-impact AI systems your firm uses, their purpose, first-use date and next review date.
- An AI risk register — the risks, likelihood and impact, controls, appetite, status and RAG rating, reviewed at least quarterly by responsible staff.
RICS does not prescribe a page count or a particular template. The record should be detailed enough to describe the firm's actual use and decisions, and it must stay current. The rest of this guide explains the required fields without pretending a template completes the work for you.
AI register vs AI risk register — the difference
They are two different documents and it's worth being clear which is which, because the names get used interchangeably:
- The AI register answers “what material-impact systems are we using?” — the system, provider, purpose, first-use date and next-review date.
- The risk register answers “what could go wrong, and what are we doing about it?” — the risk, likelihood and impact, controls, appetite, owner, progress and RAG or similar status.
You keep both because they do different jobs: the register proves you know what's in use; the risk register proves you've thought about the consequences and put controls in place.
What goes in an AI register
A working AI register for a surveying firm is just a short table. For a firm using SurveyorSuite it might read like this:
| AI system | Provider | Purpose | First used | Next review |
|---|---|---|---|---|
| Current generative drafting system | See supplier reference card | Drafting narrative report content from site notes | [date] | [date] |
| Fallback generative drafting system | See sub-processor list | Continuity for optional report drafting, when enabled | [date] | [date] |
A few practical points: name the provider, not just the tool, so an auditor can trace the data path; record a first-used date so you can show when each capability entered the firm; and add a row whenever you adopt a new AI system. SurveyorSuite dictation is transcription, not AI. It writes down the surveyor's own words; it does not generate, rewrite, recommend or decide report content, so it does not belong in this AI-register example.
What goes in an AI risk register
The RICS Standard specifies what each risk-register entry must cover: a description, likelihood and likely impact, a mitigation or management plan, the firm's risk appetite, status or progress, and a RAG or similar indicator. The worked prompts below are a starting point only; replace every placeholder with the firm's own assessment, owner and evidence.
What the R/A/G rating means. The Standard requires a RAG or similar status indicator. R/A/G is a simple red–amber–green traffic-light score for how serious each risk still is after your controls are taken into account — its “residual” risk. You pick one colour per row:
- Green — low / acceptable. Your controls bring the risk down to a level you're comfortable with. Routine monitoring is enough — no action needed.
- Amber — medium / watch. Controls are in place but there's a gap, or the risk needs active management. Record a planned action and a date against it.
- Red — high / act now. Current controls aren't enough on their own. Give it an owner and an immediate action before the risk is acceptable.
RAG does not replace the separate likelihood and impact assessment. Anything rated amber or red should carry a named owner, dated action and current status rather than just a description.
| Risk | Likelihood / impact | Controls and actions | Appetite / owner / status | RAG |
|---|---|---|---|---|
| AI invents a factual detail that is missed on review | [record both] | Named-surveyor review of every AI-drafted section; compare against inspection evidence; periodic QA sampling | [appetite] · [owner] · [action/status] | R / A / G |
| Private or confidential information is sent to the AI without suitable preparation | [record both] | Data-handling policy and staff training; anonymise where practical; record consent and risk assessment where required | [appetite] · [owner] · [action/status] | R / A / G |
| The firm proceeds despite a client's request about generative drafting | [record both] | Prominent warning; surveyor acknowledgement that the matter was resolved; internal record of request, acknowledgement and use; truthful report disclosure | [appetite] · [owner] · [action/status] | R / A / G |
| AI output is biased or discriminatory | [record both] | Reviewer training; check language and conclusions for unsupported assumptions; defined route for reporting and learning from problematic output | [appetite] · [owner] · [action/status] | R / A / G |
| The AI provider is unavailable during working hours | [record both] | Manual drafting remains available; document the fallback process; communicate service interruption where necessary | [appetite] · [owner] · [action/status] | R / A / G |
| AI output reproduces distinctive or attributed text without a reliable source | [record both] | Check quotations, citations and distinctive wording; remove or verify unsupported source claims; record the provider's contractual position | [appetite] · [owner] · [action/status] | R / A / G |
| A provider or model change alters output behaviour | [record both] | Monitor supplier notices; repeat practical tests after material changes; pause use if output is no longer suitable | [appetite] · [owner] · [action/status] | R / A / G |
| A client challenges the use of AI after delivery | [record both] | Retain evidence under the firm's documented schedule; keep an explainability and complaints process; answer from the actual project record | [appetite] · [owner] · [action/status] | R / A / G |
| Staff become over-reliant and stop applying professional scepticism | [record both] | Annual training; peer review of a sample; management action where review quality falls below policy | [appetite] · [owner] · [action/status] | R / A / G |
Add any firm-specific risks below those — for example a report type where you've decided not to use AI, or a data category your firm prohibits from going near it. The register earns its keep at review time. RICS requires it to be reviewed and updated at least quarterly by staff responsible for decisions about the firm's use of AI.
Download the free Excel template
A starter workbook with an AI register, a risk register and a checklist for the wider RICS record. It is a structure to complete with your firm's facts, owners, decisions and evidence — not a completed assessment and not compliance advice.
Download the AI compliance register (.xlsx)The rest of the RICS AI compliance file
The two registers sit inside a wider set of records required where AI use is material. The detail should match the firm's actual systems, services and risk profile:
- Materiality determination — a written assessment of whether each AI use has a material impact on the service. Generative report drafting will ordinarily be material. SurveyorSuite dictation is transcription rather than AI drafting and sits outside this AI compliance workflow.
- Procurement and due-diligence record — the information requested from the supplier, the documents reviewed, the practical testing carried out, the firm's conclusion, and any information gaps transferred into the risk register. Revisit it when a provider, purpose or other material fact changes, and on the firm's documented review cadence.
- Client disclosure in Terms of Engagement — telling clients generative drafting or rewriting may be used, its scope, your PI position, and how they can record a request or preference, contest the use and seek redress. A request is not an automatic product-enforced ban: if the firm proceeds, it must resolve the matter with the client and accept responsibility.
- Reliability-decision process — the named surveyor reviews every AI-drafted section against their own site notes and professional knowledge, corrects what falls short, and records their decision. The exported report's attestation is supporting evidence, not the whole written reliability decision.
- Annual staff training record — who was trained, when, and on what, for everyone who uses the AI features.
Read the standard itself. The RICS Professional Standard Responsible use of artificial intelligence in surveying practice (1st edition, in force 9 March 2026, ISBN 978 1 78321 555 3) is free to download from RICS. It's short and worth reading once — it's the authority your registers answer to, and its current wording and section numbering govern over any summary, including this page.
How often to review
Section 3.3 says responsible staff at RICS-regulated firms must review and update the risk register at least quarterly. The AI register must record the date on which each system's use and appropriateness will next be reviewed, but the Standard does not set one universal interval for that register or for the whole file. Set a documented cadence, bring reviews forward after material changes, and date each update.
How SurveyorSuite fits this
SurveyorSuite supports your RICS AI compliance process by helping you record and evidence the decisions that remain yours. Every AI generation is logged to an internal audit trail; where a client request is active, the request, acknowledgement and subsequent generative use are recorded. Every AI-assisted report carries an AI disclosure block with the surveyor's reliability attestation. The attestation supports, but does not complete, the firm's written reliability decision. We publish the current provider position and primary-source links, and maintain a supplier due-diligence pack and an AI supplier reference card for your procurement file. For the full picture of what the AI does and doesn't do — and the primary sources behind every claim — see our AI and compliance page.
The internal audit trail is not client-facing. Actual generative-AI use is disclosed in the delivered report. SurveyorSuite dictation is transcription, not part of the generative-AI drafting and rewriting workflow or the related client request.
Frequently asked questions
Does the RICS AI Standard actually require an AI register and risk register?
What is the difference between an AI register and an AI risk register?
How often should the AI risk register be reviewed?
Does voice-to-text dictation need to go in the AI register?
Is there an AI risk register template I can download?
Important — general information, not compliance advice
This page is a general, plain-English summary written for surveyors, based on the RICS Professional Standard Responsible use of artificial intelligence in surveying practice as published. It is not legal, regulatory, or compliance advice and must not be relied on as a definitive statement of your obligations. The standard's exact requirements, section numbering, and any later editions govern — always read the current RICS standard, apply your own professional judgement, and take your own advice where the position matters. SurveyorSuite Ltd is a software provider, is not a regulator, and is not affiliated with or endorsed by the RICS.
Last reviewed: · Checked against the RICS Professional Standard on the responsible use of AI current at the date of review. We update this page when the standard or our products change.