The Bias Nobody Is Talking About: How AI Is Quietly Setting the Standard Appraisers Will Be Judged Against
by Mark Raney, SRA
Most of the debate over UAD 3.6 has focused on Certification 25: who gets our data, what they can do with it, and what we give up legally by signing it. Those are fair concerns. But there's a quieter consequence of the new form that almost nobody is discussing, and it may end up mattering more than any lawsuit. Artificial intelligence is starting to define what a "normal" appraisal looks like, and appraisers will be measured against it.
From reports to datasets
Under UAD 3.6, the appraisal stops being a narrative document and becomes structured data. Condition and quality ratings, component-level updates, adjustments, and tagged photos all arrive in a format that can be sorted, compared, and analyzed automatically. Certification 25 gives lenders, the GSEs, and downstream recipients the right to store, copy, analyze, and use all of it for internal or external purposes.
At the same time, AI tools are moving into the appraiser's own workflow. Computer-vision products already read listing and appraisal photos at massive scale and turn them into condition and quality scores for automated valuation models. It's only a matter of time before more appraisers adopt similar tools to suggest ratings and adjustments in their own reports, if they haven't already.
Put those two trends together, and you get a feedback loop.
How the loop works
1. AI tools suggest condition ratings, quality ratings, and adjustments based on patterns in past data.
2. Appraisers who rely on those tools submit reports that cluster around the AI's suggestions.
3. Those reports flow into the GSE databases and become the "market norm" that review software checks new reports against.
4. An appraiser who inspected the property, exercised independent judgment, and reached a different and well-supported conclusion now looks like the outlier.
5. That report gets flagged, questioned, or pulled into a repurchase review. The lesson every appraiser learns is that the safest rating is the one that matches the machine.
Nobody designed this as a bias. It's an unintended side effect. But it's a bias all the same, and it compounds over time. The more appraisers conform, the stronger the AI median gets, and the riskier it becomes to disagree with it.
We've seen this before
Think about what Zillow's Zestimate did to residential pricing. It was never an appraisal, but it anchored the expectations of buyers, sellers, and agents. List prices and offers drifted toward it, and appraisers came in "low" or "high" based on how they compared with a number no one had to defend. An AI-driven condition and quality median risks doing the same thing inside the appraisal process itself, except this time it shapes the reviews that decide whether our work is accepted.
Why this is a USPAP problem
USPAP requires credible assignment results based on the appraiser's own independent analysis. An appraiser who changes a condition rating to avoid a review flag, and not because the property supports it, is no longer exercising independent judgment. They're managing risk against an algorithm. When the system rewards conformity and punishes well-supported dissent, it pushes appraisers toward exactly the behavior USPAP is meant to prevent.
It also raises a fairness question. AI models learn from historical data, and if that data carries old patterns of undervaluing certain neighborhoods or market conditions, the median inherits them. An independent appraiser can catch and correct for that. An appraiser pressured to match the median can't.
What should happen
This isn't a call to reject technology or refuse UAD 3.6. That's not realistic. It's a call for the profession and its organizations to start asking questions before the loop sets in:
The GSEs should disclose whether and how AI-derived data is used to build the benchmarks that appraisal reviews rely on.
Review standards should treat a well-supported departure from the norm as professional judgment, not as a red flag on its own.
The Appraisal Foundation should issue guidance on how appraisers can use AI tools and still meet USPAP's independence and credibility requirements.
Professional organizations, including the Appraisal Institute, should put this on the agenda now. Even if they can't change Cert 25, they can make sure the bias it enables is named and scrutinized.
The value of an appraiser has always been that a trained professional walked through the property and made an independent judgment. If the system quietly trains us to agree with the machine, we lose the one thing that makes an appraisal worth ordering. That conversation needs to start now, before the median becomes the standard.
