You’re under contract. Inspection was Tuesday. The report just hit your inbox. 80 pages of red text and hundreds of photos. Every imperfection in the house, individually flagged.

You did what most buyers do in 2026. You opened ChatGPT, dragged in the PDF, and typed something like “tell me what’s wrong with this house.”

Your favorite AI LLM(Grok, Claude (our preference), ChatGPT, etc. just rewrote the script to the Money Pit. If you haven’t seen it, add it the list for 80s movie night.  The AI isn’t lying. but the question you asked shaped the answer, and the answer in front of you is making a perfectly normal house look catastrophic.   

Here’s how to use AI on an inspection report the right way. And where to stop trusting it.

Why every inspection report looks terrifying

A home inspection report is designed to flag everything. A thorough inspector walks a house and notes every imperfection. A hairline crack in the driveway. A chipped grout line in the master bath. A slow drain in the guest bathroom. A missing GFCI outlet in the garage. The standard is “if I see it, I write it up.” A 60-page report on a 1925 Sellwood bungalow is normal. A clean report with five items on it would be a red flag, because it would mean the inspector didn’t actually look.  

So when you open the report and see 200 items, that’s the inspector doing their job. Not the house failing.

Four kinds of items in every inspection report

Once you’ve calmed down about the volume, the harder sort starts. Every item the inspector flagged belongs to one of four buckets. Which bucket it lands in tells you what you can actually ask for.

1. Things you already knew about and priced into your offer.

The listing said the roof has a five-year certification. You offered less because of it. Now the inspection confirms five years left. You don’t get to ask for a new roof. The price you offered already accounted for it. And if you signed “as-is, no repairs” to win the deal, this whole category disappears.

2. Things every reasonable buyer would ask for.

Safety items. Anything required by the lender for financing. Anything required for occupancy under local code. The non-negotiable floor. These aren’t asks specific to you. They’re what any buyer in any market would put on the addendum.

3. Things no one knew about.

The actual surprises. A failed sewer line that hadn’t been scoped. An undiscovered oil tank in the yard. A furnace about to fail. The hidden defects the inspection exists to surface. These are the strongest asks because nobody priced them in. Not you. Not the seller.

4. Things you personally want.

Wish-list items. Nice-to-haves. New carpet because you don’t love the color. Replacing a working dishwasher because you’d prefer a different brand. These are your list, not the universal list. They have the lowest claim and the highest cost to your overall negotiating position. Sometimes worth including. Sometimes not.

The bucket determines what’s possible. So does the context of how you got the house. A buyer who beat ten offers with an “as-is” clause has very different room than a buyer who was the only offer at full asking. Market context decides what you can ask for, what you’ll get, and what you should let go.

This is the part AI can’t help with. AI doesn’t know what you priced in. AI doesn’t know how many offers were on the table. AI doesn’t know what’s typical in this market this month. This sort happens on the call with your agent, with your contract and your situation on the table.

Why “tell me what’s wrong with this house” is the wrong prompt

When you give ChatGPT an unconstrained prompt like “tell me what’s wrong with this house,” the model does what it’s trained to do. It summarizes the document comprehensively. Every flagged item comes back. The output reads like a list of disasters because every imperfection gets equal weight, whether it’s a $40 GFCI outlet or a $14,000 sewer line.

That output is technically accurate and practically useless. It’s the equivalent of asking a doctor “what’s wrong with my body” during a physical and getting back “your left earlobe is slightly larger than your right” alongside “your blood pressure is concerning.”

You don’t need the AI to summarize. You need it to sort.

Four prompts that work

Here are the prompts we send our buyer clients when their report lands. Paste them one at a time, with the inspection PDF attached, into the AI of your choice.

Prompt 1: Triage.

Read this home inspection report and group the findings into three buckets: (1) safety or health concerns, (2) major systems likely to need repair or replacement in the next 1–3 years, and (3) routine maintenance and cosmetic items. Don’t editorialize, just sort. Use the inspector’s own language for each item.

This forces the AI to do the sorting work for you, in the inspector’s words. You’ll usually get back five to ten items in bucket 1, ten to fifteen in bucket 2, dozens in bucket 3. The report feels less overwhelming already.

Prompt 2: Age check.

This home was built in [YEAR]. For each item flagged in the report, tell me whether it’s typical and expected for a home of this age in the Pacific Northwest, or whether it’s unusual and worth real concern. Keep your answer to one sentence per item.

This is the one that changes minds. A lot of things that look scary in isolation are completely normal for a Portland home of a certain vintage. Knob-and-tube remnants. Galvanized supply lines. A heaved front walk. Asking the AI to calibrate against the home’s age separates “alarming” from “expected for this neighborhood, this era.”

Prompt 3: Cost handling.

For any items likely to cost more than $1,000 to address, flag them as “get a real bid.” Do not estimate dollar amounts. Costs vary too much by region and contractor, and I don’t want a wrong number anchoring my thinking. My real estate agent has vetted local contractors and will pull actual numbers for the items that matter.

This is the most important prompt in the set. More in a moment.

Prompt 4: Questions for your agent.

Based on this inspection report and what I’ve told you about the home, what are the top 5 questions I should be asking my real estate agent before deciding how to respond? Be specific to this report, not generic.

This converts your anxiety into a productive call with your agent. Instead of arriving with “I’m freaking out about everything,” you arrive with five focused questions tied to actual findings.

Where AI ends

Here’s the line.

AI is good at sorting and summarizing. It’s good at telling you whether a finding is normal for a 1925 house or unusual.

AI is bad at two things. Really bad.

It hallucinates costs. Ask ChatGPT what a sewer line replacement costs and it will confidently quote you a number. That number is a national average pulled from training data of unknown vintage. It doesn’t reflect what contractors in Northeast Portland or Battle Ground or McMinnville actually charge in 2026. We’ve seen buyers walk into negotiations anchored to a $4,000 AI estimate for a job that bids out at $14,000. They get laughed out of the room.  We’ve got the data from local vendors to ballpark the cost before getting real time estimates. 

It doesn’t know what to ask for. Every inspection negotiation has the same three options for every Major item. Ask for the repair before close. Ask for a credit at close in lieu of repair. Ask for a specialist evaluation. Which one to use depends on the price point. The seller’s situation. The days-on-market. What was already disclosed. What’s typical for that neighborhood. A dozen other things AI doesn’t have access to.

AI gives generic answers. Generic answers leave money on the table. Or kill deals that should have closed.

The pattern we see when buyers walk

We work mostly with buyers in the Portland metro and Mid-Willamette Valley. We’ve worked through inspection reports on homes from 1898 to last year’s new construction. The pattern is consistent.

Buyers who panic and walk away from a deal almost always panic because of bad information. An inspector finding they didn’t understand. An AI cost estimate that was way off. A generic online article that scared them about something normal for the neighborhood.

Buyers who close on a great house and feel good about it had two things on their side. A clear framework for sorting the report. And a calibrated read on what’s reasonable to ask for in this market.

We can’t promise you a clean report. No report is clean. But when your inspection comes in, we’ll have done this 2,500+ times before, we’ll have actual local cost data and a vetted vendor list instead of national averages, and we’ll know which findings in your specific neighborhood are typical and which ones aren’t.

What to do when your report lands

If you’re a Lovejoy buyer, your agent has already walked you through this. The day-of-inspection handout you got at the property is the same framework you just read. The four prompts above are the same ones we email you the day the report drops.

If you’re working with another agent and you’re reading this because you’re staring at your own report right now, try the prompts. They work whether you’re our client or not. 

Either way, don’t walk from a good house because the AI gave you a scary list. Sort the list first.

The vetted vendor list and local cost data lives in the Lovejoy app. Free to install. The four prompts above get you sorted. The app gets you actual names and numbers for the items that matter. Most of our buyer clients use both. The AI to organize. The app to bid out.

Get the Lovejoy App

Full RMLS and WVMLS search, saved favorites, instant alerts, and our vetted homeowner resources. Free to install. Your data stays between us.