Appraisal Institute on AI: What Appraisers Should Check First

I found no standalone Appraisal Institute rulebook on AI. Its seminar material, including the "AI on AI" series, frames AI as a decision-support tool whose outputs must be independently reviewed and verified. The enforceable standard is USPAP, and Advisory Opinion 41, adopted 23 April 2026, applies it to AI tools. Check competency, confidentiality, accuracy and workfile records first.

A note on sources. The Institute material I could read in November 2026 is educational: the "AI on AI" webinar series, chapter seminars that work through FAQ #110, "Artificial Intelligence (AI)", described as sitting in the Guide Note materials, and a conference post saying the responsibility for credibility, transparency and independence remains with the appraiser. I have not reproduced the FAQ text, and the advisory opinion is summarised here from published summaries. Read both in the original before relying on them, and use the checks below, which hold up under either.

Follow me on Instagram@sagnikteaches

Where the guidance actually lives

Appraisers often search for "the Appraisal Institute's AI rules" and expect one page. In practice, guidance on AI sits in four layers, and it helps to know which carries weight.

Connect on LinkedInSagnik Bhattacharya

• USPAP and its advisory opinions. The Uniform Standards of Professional Appraisal Practice are the enforceable standards. Advisory Opinion 41, "Use of Technology in an Appraisal or Appraisal Review Assignment", was adopted by the Appraisal Standards Board at its 23 April 2026 public meeting and took effect immediately, after a first exposure draft dated 2 January 2026. Summaries describe it as covering tools such as automated valuation models, regression and statistical software, and generative AI. An Institute seminar description ties it to the Ethics Rule, the Competency Rule and the nondiscrimination provisions, and one summary notes that it retires two older opinions, AO-18 and AO-37. Advisory opinions illustrate how the standards apply; they do not replace reading the Rules themselves.
• The Appraisal Institute's own material. Its education catalogue carries the "AI on AI" live webinar series, whose second season runs monthly from July to December 2026 on topics such as data privacy and confidentiality, AI-assisted adjustment support, and model validation and QC checks. The series description asks participants to approach AI as a decision-support tool that can enhance, but not replace, the appraiser's expertise, and says all AI outputs should be independently reviewed, verified and applied with professional judgement. Chapter seminars add FAQ #110 from the Guide Note materials. Read that FAQ directly, because I cannot reproduce it here.
• Your licensing board. Boards license and discipline appraisers, and some publish their own notes or consult on guidance. Check yours.
• Your clients' engagement terms. Lenders, management companies and other clients can write rules on technology and data into engagement letters and appraiser guidelines. They bind you by contract even where USPAP says nothing.

The useful reading order is: your engagement terms, then the advisory opinion, then the Institute's FAQ, then your board. The last three tell you what is expected; the first tells you what you have already agreed to.

Subscribe on YouTube@sagnikteaches

One trap when you search: "AI" is also the Appraisal Institute's own abbreviation. Results mix artificial intelligence with ordinary Institute pages, the ai.appraisalinstitute.org address is the Institute's member sign-in rather than an AI resource, and the Institute's own seminar series on the subject is called "AI on AI". Search for "artificial intelligence" in full, or for "Advisory Opinion 41", and you will reach the material that matters much faster.

The principle that runs through all of it

Summaries of Advisory Opinion 41 stress that tools do not do appraisals, appraisers do: the appraiser remains responsible for deciding whether a tool's output is accurate, and a tool cannot itself comply with USPAP. Whether the opinion states it in those exact words, the practical rule is clear and consistent with the Competency and Ethics Rules: you must understand what the tool did well enough to stand behind the result, and you must not let it substitute for your own judgement.

Seen that way, an appraiser's AI policy is a list of checks you perform before accepting anything a tool produces, and a record that proves you performed them. For the practical side of using AI on comparables and drafts, comps prep and draft report checks covers the tasks themselves.

Six checks to run before you use any AI tool

1. Competency: do you understand what it does?

The Competency Rule asks whether you have the knowledge and experience to complete an assignment competently, and that includes tools you use. Before relying on a tool, write one paragraph answering: what is it, what does it take in, what does it produce, and where is it known to fail? If you cannot write that paragraph for a chatbot, an automated valuation model or a photo-analysis feature, you are not ready to put its output in a report. A short test: could you explain to a reviewer, without the vendor's brochure, how it arrived at the number or sentence?

2. Confidentiality: what are you sending and where does it go?

Appraisal assignments carry confidentiality duties to the client, and client details can include borrower names, loan information and property addresses. Pasting those into a public chatbot raises the question of where the data is stored and whether it can be used to train a model. Consumer plans of the main chat assistants let you switch off use of your chats for training, and business plans typically do not train on business content by default, but the settings and the terms change, so read the current ones. The safest routine is to strip names, addresses and lender details before any text leaves your machine, and to ask the client before using any tool on a confidential file when their terms are silent.

3. Accuracy: can you verify every fact it gives?

Generative tools produce fluent text that can contain invented facts: a wrong zoning description, a school that does not exist, a sale that never happened. Treat every factual statement in a draft as unverified until you have matched it to a source in your workfile: the MLS record, the public record, the inspection, your own notes. A good habit is to colour-code a draft, with facts in one colour once verified and opinions in another once you have adopted them as your own.

4. Bias and discrimination: could the output carry it?

The nondiscrimination provisions mean that language and data used in a report must not reflect prohibited bias. Neighbourhood descriptions generated by AI can lean on stereotypes or demographic assumptions. Read any generated description of an area for subjective or loaded wording, and replace it with observable, sourced facts about housing, amenities and market behaviour.

5. Workfile: can you show what was done?

Standards require records that support the work. When AI is used, keep the tool name and version where you can see it, the date, the inputs you provided (anonymised if necessary), the output and what you changed. For the broader workfile question, the USPAP workfile checklist and how AI can file it goes through what to keep.

6. Client and board rules: is it allowed?

Check the engagement letter, the client's appraiser guidelines and your board's notes for any restrictions on technology or disclosure. If they say nothing, record that you checked. If they are ambiguous, ask in writing.

A solo appraiser's market-conditions section, checked

Illustration only. A solo residential appraiser wants help writing the market-conditions narrative for a report on a three-bedroom house, and has pulled a table of sales for the past twelve months.

Walking the six checks:

• Competency. The appraiser uses a general chatbot to rewrite and tighten text, not to produce market analysis. The paragraph on file says: it predicts plausible wording, does not know the local market, and may invent numbers.
• Confidentiality. The table is reduced to month, number of sales, median price and median days on market. No addresses, no borrower, no lender. The tool is on a paid plan with training switched off.
• Accuracy. The appraiser calculates the figures in a spreadsheet and gives the chatbot the results. After drafting, every number in the narrative is ticked off against the sheet. One sentence says inventory "rose sharply"; the sheet shows listings up 9 per cent, so the appraiser rewrites it to the figure.
• Bias. The draft calls the area "up and coming", which is vague and loaded. It is replaced with two observable facts: median days on market fell from 34 to 27, and the share of sales above list price rose from 22 to 31 per cent.
• Workfile. The appraiser saves the anonymised table, the prompt, the first draft and the final text in the assignment folder with a one-line note: "Tool used for wording only; figures from my spreadsheet; edited to match."
• Client rules. The engagement letter is silent on technology; the appraiser notes that they checked.

Time: roughly 25 minutes for the narrative with AI against 40 without, plus 10 minutes of verification and filing. The saving is small, around 5 minutes, but the narrative is cleaner and the file is defensible. For prompts that suit this kind of task, see ChatGPT for market-conditions sections.

A prompt that keeps the tool in its lane

You are helping me tighten wording in an appraisal narrative.
Use ONLY the figures below. Do not add numbers, locations,
comparable sales, school names or zoning details that are not in
the data. If something is missing, say "not provided".

Data (monthly, last 12 months): [paste anonymised table]

Task: write a 120-word market-conditions paragraph in plain,
neutral language. No subjective descriptions of neighbourhood
character. List any claim you made that is an inference, not
a figure from the table.

The last instruction is the useful one. It forces the tool to flag its own inferences, which you then either verify or delete. A realistic sample reply would give a clean paragraph and a short list such as "Inference: 'balanced market' is my reading of the days-on-market trend". You decide whether that reading is yours too.

Where the sections of the opinion meet daily work

Because I am not reproducing the opinion, use it as a reading list with these questions in mind, and write down your answers in your own words.

A second example: a commercial lease summary that was half right

Illustration only. A commercial appraiser asks a chatbot to summarise a 40-page lease for a small office unit and extract rent, term, escalation and break rights. The reply is neat:

Rent: $42,000 a year, reviewed annually by 3%.
Term: 5 years from commencement.
Break: tenant may break after year 3 on 6 months' notice.

Checked against the lease, the rent and term are right. The escalation is a fixed 3 per cent only from year three, with a flat rent for the first two years, and the break right needs a penalty payment equal to three months' rent. Each of those errors changes the income analysis. The appraiser corrects the abstract, saves the lease pages that support each line, and notes in the AI-use record that the tool was used to produce a first-pass summary only, with every line verified to the source. This is a typical failure: the summary reads as complete, so the gaps are invisible unless you check against the document. The rule that follows is simple. A summary is a map to the source, and you must go to the source for anything you will rely on.

A third example: reviewing a generated neighbourhood paragraph

A draft says: "The area is a desirable family neighbourhood with excellent schools and low crime." Three problems show up on a read-through. "Desirable" and "family neighbourhood" are subjective. "Excellent schools" and "low crime" are factual claims with no source. And a phrase such as "family neighbourhood" can carry assumptions about who lives or should live there. The appraiser replaces it with observable, sourced content: dominant property type and age range, distance to the nearest primary school and the rating source if one is relied on, recent permit and sales activity, and facts from the public record. The result is longer, drier and defensible. If you do this once on a bad paragraph you will recognise the pattern in every later draft.

How much time the checks cost

Appraisers rightly ask whether the checks eat the saving. For a typical residential assignment where AI helps only with wording, the extra work is about ten minutes: stripping identifiers (2), verifying figures (5) and writing the log (3). If AI saves 15 to 20 minutes of drafting, you are slightly ahead and much safer. If a task saves less than the checks cost, such as a one-paragraph tweak, leave AI out. That is a perfectly good policy: use it where the time saved is clear, and do not feel obliged to use it elsewhere.

Five ways this goes wrong in a real practice

The invented comparable

A chatbot asked for "similar sales" produces plausible addresses and prices. If there is no data source behind them, they are fiction. Never ask a general chatbot to find comparables. Give it records you pulled and ask it only to format or describe them.

The quiet confidentiality leak

An appraiser pastes a full engagement email, including the borrower's name and loan amount, into a free chatbot to "summarise the scope". The summary is fine, but the data has left the controlled environment. A rule such as "scope text goes in only after removing names and numbers" prevents it.

The confident wrong adjustment rationale

The tool explains why a $15,000 adjustment for a garage is appropriate, citing paired sales that do not exist. The appraiser's own paired-sales analysis, which you should see in the workfile, supports the adjustment or it does not. Ask the tool to help phrase your analysis, not to supply it.

The tool nobody recorded

Six months later a reviewer asks how a paragraph was produced, and the appraiser cannot say which tool or version was used. The fix is a one-line log per assignment.

The assumption that the software vendor has checked compliance

Vendors market tools as "USPAP-ready". That phrase is a sales claim, not a standards finding. The responsibility is yours, so ask the vendor what the tool does, what it does not do, and where data is stored. AI-ready appraisal software and UAD 3.6 compares what vendors say with what you still need to check yourself.

A short AI-use note for your workfile

AI use note (assignment ref: [ref])
Date: [date]    Appraiser: [initials]

Tool and version: [name, version, plan]
Purpose: [wording only / data formatting / summary]
Data supplied: anonymised table; no names, addresses or lender data
Settings: training off; chat history [on/off]
Output used: [paragraph 3 of market conditions, edited]
Verification: all figures checked to [spreadsheet/MLS record]
Bias review: loaded descriptors removed
Client terms checked: yes, no restriction found
Conclusions reached by: me

Ten lines, two minutes. It is the cheapest protection an appraiser has. Adapt it to your own workfile style, and keep it with the assignment for the retention period your standards require.

What to put in a one-page practice policy

1. Which tools are approved, on which plans, with which privacy settings.
2. What data may never be pasted (names, addresses, loan details, client identifiers).
3. Which tasks AI may help with (wording, formatting, summarising your own notes) and which it may not (selecting comparables, forming adjustments, signing off conclusions).
4. The verification step required before any output enters a report.
5. The workfile record you keep.
6. Who to ask when engagement terms are unclear.
7. A review date, because tools and guidance change several times a year.

If you work with others, extend it to anyone who prepares drafts for you. A trainee or assistant using a chatbot without your rules is the most common route to a problem.

Keeping up with a moving target

Guidance on AI in appraisal is young. The advisory opinion was exposed for comment and then adopted within months, and professional bodies are running courses and revising FAQs as they learn. Build a habit of checking three places each quarter: the Appraisal Foundation's pages for standards updates, the Institute's Guide Note and education listings for FAQ changes, and your board and key clients for new requirements. Note each check in a log with the date. When you read a summary, including this one, treat it as a pointer to the primary text, not a substitute.

For a wider picture of AI in appraisal work, AI prompts for appraisers collects task-level examples, and the AI guide links the rest of the series. If you would like help turning the checks above into a policy and workfile template for your own practice, that is what my AI implementation consultation covers, through the contact page.

Common questions from appraisers about AI guidance

Has the Appraisal Institute banned AI for appraisers?

I found no ban. The Institute's AI material I could confirm is educational, such as its AI on AI webinar series, which treats AI as a decision-support tool whose outputs must be independently reviewed and verified. The enforceable standards are USPAP, and Advisory Opinion 41 explains how they apply to AI and other technology.

Does using AI change who is responsible for the report?

No. Summaries of Advisory Opinion 41 describe technology output as something the appraiser must review and own, not as a conclusion that transfers responsibility. You sign the report, so you answer for every figure and sentence in it, however it was drafted.

Can I paste a subject property's details into a public chatbot?

Only after checking confidentiality duties, your engagement terms, and the tool's data settings. Many appraisal assignments carry confidentiality obligations to the client. Use anonymised data or a business plan with training switched off, and ask the client when in doubt.

Do I need to disclose AI use in the report?

Not routinely, in one society's view: in July 2026 the International Society of Appraisers said routine AI-specific disclosures are not necessary, but that disclosure may be appropriate when AI materially contributes to assignment results or leaving it out would mislead. Client terms can require more, so read them, and keep a workfile record of what the tool did either way.

Further reads

• UAD 3.6 Checklist for Solo Appraisers: Templates, Photos and AI Tools — Templates, photos and tools for the new data standard.
• AI for Commercial Appraisers: Lease Abstracts and Income Prep — Verifying AI lease abstracts before they feed an income approach.
• AI for Canadian Appraisers: CUSPAP, Privacy and Practical Uses — A privacy-focused view for appraisers under a different standards body.
• Can AI Help Draft a Reconsideration of Value Response? — Using AI on a reconsideration request without losing control.
• AI for Real Estate Appraisals: Report Narratives and the USPAP Checks — Draft narratives and adjustment notes with AI while keeping USPAP responsibility, confidentiality and the workfile where they belong: with you.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Appraisal Institute education listings (the AI on AI series and a chapter seminar referring to FAQ #110 and the 2 January 2026 exposure draft) and its March 2026 conference post; summaries of USPAP Advisory Opinion 41 from the International Society of Appraisers (April and July 2026), BVWire (April 2026) and an appraisal firm (June 2026), all checked November 2026. Read the primary documents on the Appraisal Foundation's site before relying on any summary.