Treat AI as a drafting and research aid inside existing CUSPAP duties: keep confidential client information out of unvetted tools, stay competent in what you use, take responsibility for every word you sign, and keep a work file you can produce. Privacy law also covers personal information in photos and files. This is not legal advice.
The most useful discovery is how little the standards say about AI by name, and how much they say about the underlying duties. In the 2024 edition I read, there is no generative AI rule. What exists is a confidentiality section, a competence rule, a responsibility-on-signing rule, work-file retention and a short section on automated valuation models. Those rules are enough to tell you where AI fits. Prompts for report narratives are a separate job, covered in AI prompts for appraisers; the guardrails come first, because they decide what a prompt may contain.
What CUSPAP 2024 actually says that bears on AI
The Appraisal Institute of Canada publishes the Canadian Uniform Standards of Professional Appraisal Practice, and compliance is mandatory for members on all types of professional services assignments. The standards state that legislation supersedes CUSPAP, so provincial licensing law sits above it. I read the 2024 edition directly, and these are the provisions that matter.
| Provision | What it says, in brief | What it means with AI |
|---|---|---|
| 5.1.1 | Perform assignments ethically, objectively, competently and without bias. | A tool's bias or blind spot becomes yours. |
| 5.2.1 and 5.2.2 | It is unethical to use a report you know, or ought to know, is defective, erroneous or misleading, including from many small errors in aggregate. | A confident AI error you did not catch is your error. |
| 5.8.2 and 5.8.3 | Do not disclose analyses, opinions or conclusions, or information a client provided in confidence, except to authorised parties, as legally required or to an Institute committee. | Pasting client material into a tool is a disclosure question. |
| 5.8.5 and 5.8.6 | Photographing an occupant's dwelling is collecting personal information, so obtain meaningful consent that states the purposes. | Uploading photos to an AI tool is a further use to cover. |
| 5.7.2 to 5.7.5 | Keep a work file with true copies, drafts, a signed certification and supporting data, for at least seven years or two years after any proceeding, whichever is later. | Decide what AI material belongs in the file. |
| 5.10.1 | Have the knowledge and experience to complete the assignment competently, or disclose the gap in writing first. | You must understand a tool before relying on it. |
| 7.12.1 | By signing, you take responsibility for the report, including any professional assistance. | Your signature covers anything an AI helped write. |
| 9.8.9 and 9.8.10 | When using an AVM, decide whether it is appropriate, make sure it does not exclude relevant data, disclose assumptions and limiting conditions, and identify it. The output of an AVM is not a value. | The closest existing rule for model-driven output. |
Two definitions help. An automated valuation model is defined as a computer program that analyses data in an automated process and may include regression, adaptive estimation, neural network, expert reasoning or AI programs. And competence means having sufficient knowledge, skill and experience for the specific assignment. If your report uses any model-driven output as evidence, 9.8.9 is your checklist, and 9.8.10 is the reminder that the output is not a value until you apply judgement.
The standards also distinguish clerical assistance, such as collating reports, preparing appendices and spelling and grammar checking, from professional assistance that leads to the analyses, opinions and conclusions. AI drafting grammar is clerical. AI selecting comparables or shaping a conclusion starts to look like professional assistance, and you cannot delegate responsibility for either. The definitions were written before generative AI, so treat this reading as a sensible analogy rather than a ruling, and check with the Institute if a case is borderline.
I found no AI-specific guidance for appraisers in the Institute's published standards. It may exist in member communications, so check the Institute's member area before drawing firm lines. For how another appraisal body frames the first checks, see the tutorial on the Appraisal Institute's guidance.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowPrivacy law on top of the standards
The federal Personal Information Protection and Electronic Documents Act applies to private-sector organisations that collect, use or disclose personal information in the course of commercial activity. Alberta, British Columbia and Quebec have their own private-sector laws deemed substantially similar. For an appraiser, personal information includes borrower names and incomes, owners' details, photos of identifiable people or of a home's interior, and notes on occupants' circumstances.
Three of PIPEDA's principles do the work with AI tools:
- Accountability. The commissioner's guidance says you remain responsible for personal information you transfer to a third party for processing, and should use contracts or other means to secure comparable protection. An AI vendor that processes your files is that third party.
- Consent and purposes. People who allowed a photograph for an appraisal did not necessarily agree to it being processed by an AI service. CUSPAP 5.8.6 asks for meaningful consent that states purposes in a way the occupant can reasonably understand.
- Safeguards. Financial and household data call for strong access controls. A shared login to a consumer chatbot does not provide them.
Quebec firms have further duties under its private-sector Act, including impact assessments before certain transfers of personal information outside the province, which an AI service hosted elsewhere can trigger. The accountants' tutorial, AI for accountants in Canada, walks through those provisions, and the same reasoning applies to appraisers.
The simplest rule: sort inputs into three bins
Most of the risk disappears when you decide what may go into a prompt before you are mid-assignment.
| Bin | What goes in it | Rule |
|---|---|---|
| Open | Public data and your own generic material: published zoning bylaws, public market statistics, your template wording, a description of a method | Any approved tool, any plan. |
| Cleaned | Subject facts with identifiers stripped: "a three-bedroom bungalow, 1,480 square feet, built 1978, in a suburban cul-de-sac" with no address, no owner name, no client name | Approved business-plan tool only, and still reviewed. |
| Closed | Anything a client gave in confidence, borrower or owner names, addresses tied to people, interior photos, lender instructions, draft conclusions for an undisclosed client | Not in an AI tool unless the client has authorised it and the tool's terms and your assessment support it. |
Notice that a property address alone can tie facts to a person. Treat "cleaned" inputs with suspicion: strip the address and anything unusual enough to identify the house, such as a unique renovation or a named heritage feature.
Practical uses that never touch client data
Seven uses fit the open bin and are worth trying this week.
1. Plain-language drafts of your standard wording
Ask for three alternative phrasings of your limiting-conditions paragraph or your scope-of-work explanation, then replace your standing text only if a version is clearer and still accurate. Your Institute rules and any lender requirements govern the content, so use the tool for readability, not substance.
2. Market-area descriptions from public sources
Give it public statistics and the municipality's published planning summary, and ask for a neutral 150-word description of the area's land uses and trends. Illustrative before and after: the draft says "a thriving neighbourhood with strong demand". You replace it with sourced facts: "planning documents show light-industrial redevelopment along the northern edge; the published statistics show a 3 per cent rise in the median sale price over twelve months". The model supplied structure and the evidence is yours.
3. Spreadsheet formulas on dummy data
Build a table of invented sales and ask for a formula that computes price per square foot, a time adjustment at a monthly rate or a flag for sales more than 12 months old. Test it against dummy numbers you can verify by hand, then move it into your real workbook offline.
4. Engagement letters and client emails from templates
Draft the structure of an engagement letter, a delay notice or a request for documents with placeholders such as [client name] and [property address]. Fill them in your own system.
5. Inspection checklists
Ask for a room-by-room inspection checklist for a house of a stated age and construction type, then trim it against your experience. Hallucinated items are obvious to a practitioner and harmless to remove.
6. Explaining your method to a client
Turn your own bullet points on the sales comparison approach into a plain page a nervous homeowner can read. You supply the content and the model supplies the plain words.
7. Studying CUSPAP itself
Paste a section of the standards (public text) and ask for a summary and five test questions. Compare against the original, because a model can paraphrase a "must" into a "should", and that difference matters.
A prompt and a realistic output for the market-area paragraph
A prompt that keeps the model on a short lead:
Below are my bullet points about a neighbourhood. Write one neutral paragraph of
about 150 words for an appraisal report. Use only facts in my bullets. Do not
add statistics, adjectives such as "thriving", or any prediction. If a bullet is
vague, keep it vague.
- Mostly detached homes built 1960 to 1990
- Planning summary shows light-industrial redevelopment on the northern edge
- Median sale price up 3% over 12 months (public statistics, table 4)
- Two elementary schools within 1.5 kmIllustrative output: a tidy paragraph with all four facts. The reviewer's fix list is usually short but real: the model wrote "stable demand", which no bullet supports, and rounded 1.5 km to "nearby". Delete the first, restore the number in the second. This is the whole discipline in miniature: the facts and judgement stay yours, the model supplies word order, and your review removes whatever it added.
Uses that need a review step
When the work touches the report itself, tighten the process.
- Drafting narrative from your own analysis. Write your adjustments and reasoning in bullets, strip identifiers, and ask for a paragraph. The reasoning is yours, and the model's job is wording. Read the result against your bullets and delete anything not in them. The usual failure is an added justification you never made.
- Proofreading. Ask for typos, repeated words and inconsistent units, not for rewrites. Compare the changes before accepting.
- Comparable summaries. If a tool summarises public listing text, check every figure against the source record, because summaries drop or merge numbers.
- AVM-style outputs. If you use any model's value indication as a check, follow 9.8.9: decide it is appropriate, confirm it does not exclude relevant data, disclose the assumptions and identify it. It is a check on your value, never the value.
Keep the conclusion, adjustments and reconciliation in your own hands. Section 5.2.2's warning that many small errors can add up to a misleading report is the strongest argument for reading every sentence.
Work files: what to keep when AI helped
CUSPAP requires a work file with true copies of reports including drafts, a signed and dated certification, and all other data needed to support your opinions and show compliance. It also says an electronic copy sent to the client counts as a true copy, with a PDF stored on a privately accessible drive or server as the industry norm, and that storage only on an online appraisal software platform does not satisfy the requirement. Apply the same logic to AI tools: the vendor's chat history is not your work file.
A workable habit for the file:
- Save the final text you used and the evidence behind each figure.
- If an AI-assisted draft shaped a significant passage, save the prompt (with identifiers removed) and the output you started from.
- Add a two-line note: tool, task, what you changed, date.
- Keep it for at least seven years, or two years after any proceeding, whichever is later.
AI use note (work file, assignment 24-0388)
Tool / plan: [tool], business plan
Task: Rewrote 180-word neighbourhood paragraph from my bullets
Inputs: My bullets and public planning summary; no names or address
Changes: Removed "thriving"; replaced two unsourced claims with cited figures
Reviewed by: [name], AACI, [date]The note takes a minute, and if a client or committee ever asks how a paragraph came to be, you can answer. The work-file checklist tutorial covers filing in more depth, though it is written for another country's standards.
Photos and inspections: the easy mistake
Appraisers collect more images of private homes than almost any profession. CUSPAP records the federal privacy commissioner's position that photographing an individual's dwelling is a collection of personal information and a violation of statute without the occupant's consent, and rule 5.8.6 asks for meaningful consent that states its purposes in a way the occupant can reasonably understand. Uploading interior photos to an AI tool for condition notes adds a new use and a new recipient. Before you do it, check three things: whether your consent wording covers it, where the tool processes and keeps the images, and whether a text description you write yourself would do the same job. For most assignments the text route is the safer one.
A worked example: a solo appraiser sets up in an afternoon
Say a solo AACI completes about 14 residential reports a month and spends roughly 35 minutes on each drafting the neighbourhood and market sections. She wants that cut without risk to her designation.
- Read the rules (45 minutes). She reads sections 5.7, 5.8, 5.10, 7.12 and 9.8 of CUSPAP and writes down the five that bear on AI in her own words.
- Pick one tool and check it (60 minutes). She chooses a business-plan chat tool, reads its data terms, confirms it does not train on business content by default and notes the retention period. She checks current plan pricing on the vendor page before buying and keeps a dated note.
- Write the three-bin rule (20 minutes). Open, cleaned, closed, with examples taken from her own files.
- Build one prompt (30 minutes). Her prompt takes her own bullets, forbids invented figures and asks for 150 words in neutral language.
- Trial on five reports. Each draft is checked against her bullets. She finds that the tool inserts an unsourced phrase in three of five drafts, which she deletes. Drafting time falls from 35 minutes to about 20, including the check, a saving of roughly 15 minutes on 14 reports, or about three and a half hours a month.
- Add the note. She adds the two-line AI use note to each work file.
The gain is modest and honest. The bigger win is knowing exactly why each sentence is in the report.
Telling clients how you use AI
CUSPAP's confidentiality section is built on the client relationship, so a short statement in your engagement letter is good practice even where no rule forces it. Something along these lines, adjusted to your practice and checked with the Institute or your adviser, does the job: "I may use software with AI features for drafting and proofreading. I do not enter information you give me in confidence, or details that identify you or the property, into AI tools unless you agree in writing. I review all text before it appears in my report, and I remain responsible for it." Lenders and appraisal management companies sometimes set their own rules about technology, and where their instructions conflict with your preferred workflow, read the assignment conditions first and ask before proceeding. A client who prohibits AI use outright is easy to honour if you have written the rule down in advance and tagged the file.
What would change the picture
Three developments would change what you do, and each is worth a quarterly check. Appraisal-body guidance may arrive: the Institute or your provincial regulator could publish AI-specific rules, which would override any analogy drawn here. Privacy regulators may issue sector guidance, and Quebec's rules in particular affect where a tool may process data. And vendors change terms, so a plan that did not train on your content when you checked may differ next year. Put a dated review of tool terms in your calendar twice a year.
Four things to do before your next assignment
- Read the CUSPAP sections above and underline the five rules.
- Write your three-bin rule with examples from real files.
- Try two open-bin uses on dummy data and measure the time.
- Add the AI use note to your work-file template.
If you would rather map this with someone, that is what my AI implementation consultation covers. For prompts once the guardrails are set, the comps preparation tutorial is a good next read.
Questions appraisers ask about AI and CUSPAP
Does CUSPAP mention artificial intelligence?
In the 2024 edition I read, AI appears in the definition of an automated valuation model, which can include regression, neural network, expert reasoning and AI programs. There are no separate generative AI rules, so existing rules on ethics, competence, confidentiality and work files apply. Check the Appraisal Institute of Canada for newer guidance.
Can I put a client's report details into ChatGPT?
CUSPAP says you must not disclose information a client gave you in confidence except to the people the client authorises, where the law compels it, or to an Institute committee. Pasting it into a tool you have not vetted risks breaching that. Keep client data out unless the client has authorised it and the tool's terms protect it.
Is an AI-written paragraph in my report a problem?
Not in itself, but you sign the report and accept responsibility for its content, analyses and conclusions. Any text must be your own reasoned analysis, checked against the facts in your file, and not misleading. Treat generated text as a draft you must verify and rewrite.
Where should I store AI drafts and outputs?
CUSPAP asks for work files kept for at least seven years, or two years after a legal proceeding, whichever is later, and says storage only on an online platform does not satisfy the true-copy requirement. Keep your own copies of the final report and the support you relied on, including the AI output you used.
Do photos of a property create privacy problems with AI?
They can. CUSPAP notes the federal privacy commissioner treats photographing an individual's dwelling as collecting personal information, so you need meaningful consent from the occupant. Uploading those photos to an AI tool is a further use, so check what the consent covers and the tool's terms first.
Further reads
- AI-Ready Appraisal Software: What UAD 3.6 Means for Your Practice — What AI-ready appraisal software changes in practice.
- UAD 3.6 Checklist for Solo Appraisers: Templates, Photos and AI Tools — A readiness checklist for templates, photos and tools.
- Best AI for Lawyers in Canada: Privacy, Tools and a Safe First Task — How another profession handles the same Canadian privacy rules.
- Can AI Help Draft a Reconsideration of Value Response? — Whether AI can help draft a reconsideration response.
- ChatGPT for Appraisers: Drafting Market-Conditions Sections — Drafting market-conditions text with verification.
- AI for Commercial Appraisers: Lease Abstracts and Income Prep — Lease abstracts and income prep for commercial work.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Appraisal Institute of Canada, CUSPAP 2024 (definitions of automated valuation model, clerical assistance, professional assistance and competence; rules 5.1, 5.2, 5.7, 5.8, 5.10, 7.12, 9.8.9, 9.8.10); Office of the Privacy Commissioner of Canada, PIPEDA requirements in brief and the accountability principle.