Your Inspector's Report Was Co-Written by AI. The Licensing Board Doesn't Know That Yet.
Somewhere in America today, a buyer will initial a stack of closing papers and accept the keys to a house she has lived in for exactly zero nights. Her confidence rests on a document she probably skimmed in twenty minutes: the inspection report, the only independent technical assessment in the entire transaction, the one paper whose job is to tell her what is wrong with the thing she is about to spend thirty years paying for, and it now routinely contains sentences no human typed.
Three companies shipped AI into the inspection workflow in the first half of 2026, and none of them asked the licensing boards first, because there was nobody to ask: the boards have said nothing, the trade associations have written nothing, the insurers have published nothing, and the entire accountability framework is an empty chair. Nobody is sitting in it.
Meet the three new co-authors
In February, Palmtech 11 went generally available with an AI Image Defect Detector built in: upload inspection photos and the software scans them for cracks, moisture damage, and other visible issues, then drafts the report comments, descriptions and recommendations included. Porch Group, Palmtech's parent, sells this to solo inspectors starting at $50 a month, and the detail worth pausing on is that the AI does not transcribe the inspector's judgment but performs its own visual analysis of the photos, finding things tired eyes may have skipped and proposing language the inspector then approves. "You're in control" is the marketing line, which in this telling means the delete key.
In June, Spectora announced AI Report Assist, currently in early access: the inspector speaks observations aloud and snaps photos while walking the property, the AI matches the audio to comments previously reviewed and approved in a template library, and when nothing matches, the AI drafts a new narrative on the spot. Early-access inspectors report saving about 25 percent of their time per inspection, finishing reports on site instead of at the kitchen table that night, and with 12,000 inspectors on Spectora's platform, a second tool, an AI Scheduling Agent that answers the phone while the inspector is on a roof and books the job, extends the automation from the report to the business itself.
In March, Denver's Alpine Building Performance released Alpine Intelligence, a free ChatGPT-powered forecaster: an agent uploads an MLS listing sheet and receives era-based risk predictions, system-by-system likely defects, and buyer talking points, all before any licensed inspector visits the property. Its disclaimer explicitly limits the tool to general informational use, not a replacement for professional inspection, though disclaimers are cheap and predictions handed to agents during offer strategy are not.
Adoption, guaranteed by arithmetic
Spectora's 25 percent figure deserves scrutiny, since it comes from early-access users chosen by the vendor, and independent verification does not exist. But run it anyway, because even the conservative version ends the debate about whether inspectors will adopt these tools.
Assume a standard inspection consumes four hours, two and a half on site and ninety minutes on the report and admin, which is conservative for older homes and generous for new condos but close enough to show the shape of the money. A 25 percent saving frees one hour per inspection. An inspector performing 250 inspections a year frees 250 hours, which converts to roughly 62 additional inspection slots. At the national average fee of $343, that is $21,266 a year in additional revenue capacity, per inspector, against software that is currently free in early access and $50 a month at Palmtech's pricing.
Skeptics get their own version: suppose only the ninety-minute report portion shrinks by the claimed quarter, saving twenty-two minutes per inspection, ninety-four hours a year, about twenty-three extra inspections, roughly $7,900, still a raise worth having for changing nothing about how you walk a roof.
Regulators should have noticed this part, because when the private return to adoption is five figures and the compliance cost is zero, no rule addressing the practice at all, adoption does not wait for permission, and it never does. Savings arrive first. Rules arrive later, if at all.
No entry in the rulebook
Home inspection is a licensed profession in roughly 35 states. Its standards of practice, published by ASHI and InterNACHI, define the minimum scope of an inspection in considerable detail, what must be examined, what must be reported, what may be excluded, yet neither standard mentions artificial intelligence, automated comment generation, or machine-drafted narratives. As of this writing, I could find no published AI-specific guidance, advisory opinion, or rule from any state home-inspector licensing board.
Neighboring fields show what managed regulation looks like: when automated valuation models started deciding what homes were worth, six federal agencies spent years writing a rule under Dodd-Frank, effective October 1, 2025. When AI reached construction liability, ISO and Verisk published standardized AI exclusions for commercial general liability policies, the CG 40 47 and CG 40 48 forms, effective January 1, 2026, a development this publication covered in article 739. Inspection errors-and-omissions insurance, the actual backstop when an inspector misses the foundation crack, has no parallel instrument. No carrier we could find has published AI-specific endorsements or exclusions for inspectors. Code is silent, contract is silent, insurance is silent, which in American professional liability means the courts will eventually write the rule one $300,000 settlement at a time.
That figure is not hypothetical: in July, InspectorPro published a case study in which an inspector's own body-camera footage became the evidence against him, hours of raw video handed to the client containing the unflagged water damage he had walked past while joking about the refrigerator two feet away. Initial demand $425,749; settlement $175,000; total incurred $300,000. From this the insurer drew a lesson worth keeping: the final report should be a deliberate work product, not a data dump. Now invert it: when the report contains AI-drafted comments the inspector confirmed with a tap at 9 p.m., is that deliberate work product? No standard requires the report to say which sentences the machine wrote. Neither the buyer reading it nor the inspector's E&O carrier can tell which sentences those are; only the vendor's logs know, and the vendor is not a party to the inspection contract.
Three lawsuits waiting for a plaintiff
Because nobody has answered these, allow me to pose them precisely.
First comes the dismissed flag: Palmtech's detector spots a hairline foundation crack in a crawlspace photo, the inspector deletes the suggested comment as noise, and a year later the buyer finds structural movement. Discovery surfaces the vendor's logs proving the defect was machine-visible before the report was finalized. Does deleting an AI flag convert an ordinary miss into evidence of negligence? Plaintiff's attorneys dream about this fact pattern the way inspectors dream about dry crawlspaces.
Second, the rubber stamp: Spectora drafts a new narrative on the spot, the inspector confirms it without re-reading because speed is the tool's entire point, and the comment understates a moisture intrusion. "The AI wrote it" is not a defense under any existing standard of practice. But no standard requires documenting which comments were machine-drafted, so the defense will never need to be raised. Every sentence in the report carries equal authority, and the buyer has no way to discount the ones that deserved discounting. Equal authority. Unequal authorship.
Third, the pre-inspection forecast: an agent shares Alpine Intelligence talking points with a buyer who waives the inspection contingency on a 1920s bungalow because the forecast looked clean. Its disclaimer said informational use only, but courts have a long, unimpressed history with disclaimers on tools marketed to professionals for transaction decisions, particularly when the professional repeats the output to a client as guidance. Alpine built its disclaimer correctly; whether it holds is a question for a judge, not a marketing page.
Arguing against this article, at full strength
Now the objection, at full strength, because it might be right: these tools may reduce inspection liability rather than increase it.
Start with the baseline, where even careful operators miss things: one multi-inspector firm that audits its own work published a 1.1 percent avoidable-miss rate across nearly 4,000 inspections, and that firm is unusually diligent, while industry summaries put missed defects behind 45 percent of liability claims. An AI that scans every photo for the crack the inspector's tired eyes skipped at dusk plausibly catches more defects than it invents. A "review, edit, or delete" loop keeps a licensed human as the final author, which is precisely what the standards of practice require: professional judgment applied to observed conditions. Nothing in the SOPs demands that every sentence originate in the inspector's frontal cortex. If AI-assisted reports show measurably lower claim rates three years from now, today's anxiety will look like the early panic over infrared cameras, a tool the establishment also greeted with suspicion before adopting universally.
My thesis, stated honestly, is narrower than the headline. It was never the technology that posed the risk. It is the accountability vacuum: machine-proposed findings entering a liability system built entirely around human observation, with no provenance markings, no disclosure duty, and no insurer pricing the difference. Fill the vacuum and the tools are probably a net good. Leave it empty and the first landmark case will fill it instead, messily, at someone's expense.
What this article does not prove
My "no board has issued AI guidance" finding rests on web research, not a fifty-state statutory survey: absence of evidence is not evidence of absence, and a board could hold an unpublished internal position. That $21,266 figure assumes freed hours convert fully into billable inspections at the national average fee, ignoring scheduling friction, seasonality, and demand caps, so real gains will be smaller. Vendor-selected early users produced the 25 percent time-savings claim, and no independent measurement exists. An industry summary, not peer-reviewed research, stands behind the claim-cause percentages. And no E&O carrier responded in our research window about AI-specific policy language, so "none found" describes published materials, not private underwriting manuals.
What to do on Monday
If you are buying: ask your inspector two questions before hiring. "Do you use AI to draft any part of the report?" Then: "Do you document which findings were AI-flagged versus personally observed?" An inspector who cannot answer the second question is telling you something about the first.
If you are inspecting: write a one-page AI-use policy. Which tools, what the human review checklist covers, and a log of AI flags you dismissed and why. It costs nothing, and it is the first document a plaintiff's attorney will request, since "we have always done it this way" is not a policy.
If you are an agent: a forecast is preparation, not diligence. Never let a client waive a contingency on the strength of a prediction, however clean it looks. That disclaimer protects the vendor; it does not protect you.
On the closing table, the report remains the load-bearing document of the transaction. It is worth knowing who wrote it. Right now, in most states, nobody is required to tell you.