Here’s the squeeze a lot of adjusters are feeling in 2026. The carriers you work with are pouring AI into claims — the NAIC’s 16-state survey found 84% of health insurers already running AI on claims, with personal-lines giants like State Farm and Allstate right behind on homeowner and auto. AI lanes that used to take two to four weeks now settle simple claims in under ten minutes. And one Florida bill that would have required a human to review AI-flagged denials? It died in committee last June.
So the machines are getting faster on one side of the desk. Meanwhile you’re still hand-typing claim summaries, status updates, and three kinds of letters. That’s the imbalance. And the fix isn’t a $40,000 enterprise platform. It’s the same free ChatGPT everyone already has — used carefully, for the typing, never for the decision.
One former adjuster put it well on X: the dream is AI that “gives them the real job back” by eating the grunt work. That’s exactly the line we’re drawing here.
The hard line, before anything else
Read this twice, because it’s the whole game.
AI drafts and summarizes. You decide. Every coverage call, every liability determination, every dollar of a settlement — that’s yours, full stop. ChatGPT is a fast typist with good grammar. It is not an adjuster, it doesn’t know the policy, and it will state a wrong thing with total confidence. The moment you let it make the call, you become the human “rubber-stamping what the AI decided” — which is precisely the failure regulators and policyholders are furious about.
And the second rule, just as hard: claimant PII never goes into a public chatbot. No names, no policy numbers, no dates of birth, no addresses, no claim numbers. Strip them first. Replace them with placeholders — “the claimant,” “[POLICY #],” “the insured property.” You can paste the facts of a loss; you can’t paste the person.
Turn a messy claim file into a clean summary
This is the task adjusters repeat most, and the one AI is genuinely great at. You’ve got scattered notes — a recorded-statement transcript, photos described in shorthand, three emails, a phone log. You need a clean, structured summary.
Strip the PII, paste the rest, and use a prompt like this:
You are helping me organize an insurance claim file. I have removed all
personal information. Turn the notes below into a structured claim summary
with these sections:
1. Loss summary (what happened, in 2-3 sentences)
2. Timeline of key events (dated)
3. Reported damages (itemized)
4. Open questions / missing information
5. Suggested next steps
Use only the facts I provide. Do not invent details, dates, or dollar
amounts. If something is unclear or missing, list it under "Open questions"
rather than guessing.
NOTES:
[paste your de-identified notes here]
That last paragraph is doing real work. “Use only the facts I provide… do not invent” is what keeps ChatGPT from hallucinating a date or a damage figure into your file. Always read the output against your raw notes before you trust a single line.
Two more daily wins
The status update that writes itself. Status comms are the highest-frequency thing on your desk. Give ChatGPT three bullet points — what’s done, what’s pending, what you need — and ask for “a clear, professional claim status update to a policyholder, plain language, no jargon, four sentences max.” You’ll rewrite one line and send it. Multiply that by every open file.
The 30-page estimate digest. Paste a long repair estimate (de-identified) and ask it to “list the total, the major line items, any duplicate or unusually high entries, and flag five items I should verify.” Frame it exactly that way — as a first-pass reader that flags things for you to check, not an approver. You still open the estimate. You just open it knowing where to look.
For the actual letters — acceptance, denial, “we need more info” — ChatGPT can draft clean, professional language in seconds. But you write the coverage determination in your own words first, then ask it to polish your reasoning. Never the reverse. The letter reflects your decision; the AI just makes it read well.
What this means for you
If you’re a staff adjuster. Your win is volume. The summaries, status updates, and acknowledgment letters that eat your afternoon are exactly what AI clears fastest. Keep the determinations yours; hand it the typing.
If you’re an independent adjuster. You’re juggling more files across more carriers. A consistent prompt for summaries and estimate digests is a genuine throughput multiplier — and it keeps your documentation tidy when a carrier audits the file.
If you’re a public adjuster. You’re on the policyholder’s side, and the carrier’s AI is now scoring claims fast. Use AI to build a tighter, better-documented claim narrative that survives that scrutiny — clear, complete, well-organized. The hard rule for you is the same one you’d want from the other side: never invent a damage or a fact to pad a claim. A stronger story, honestly told, is the whole edge.
If you’ve never pasted a prompt before. Start with one task — the claim summary — and do it ten times until it’s muscle memory. That’s the whole on-ramp.
What ChatGPT can’t do for an adjuster
Five honest limits.
- It can’t make the coverage call. Not “shouldn’t” — can’t. It doesn’t know the policy, the endorsements, or the law. Treat any coverage opinion it offers as noise.
- It will hallucinate. It can invent a date, a figure, or a clause that sounds right. Every output gets checked against your source. Every time.
- It’s not private by default. Free ChatGPT may use your chats to train. That’s the whole reason PII never goes in — strip it before you paste.
- It doesn’t replace your judgment on bad-faith risk. A sloppy AI-drafted denial is still your denial. The accountability doesn’t transfer.
- It won’t keep you compliant on its own. Carrier rules and state regs (and they’re shifting) are on you to know. AI doesn’t track them.
The bottom line
The carriers aren’t slowing down on AI, and waiting it out isn’t a plan. But the move isn’t to fear the technology or to hand it your judgment — it’s to take the same tool for yourself and use it to delete the typing, so your hours go to the work only a licensed human can do. Draft with AI. Decide on your own. Keep the PII out.
If you want a structured, step-by-step path — the prompts, the guardrails, the de-identify-first habit — our AI for insurance client communication course walks through it for claims work, with the coverage decision kept firmly in your hands. First two lessons are free, no signup.
Sources
- Carriers Using AI for Claims but Adoption Is Fragmented — Claims Journal
- Insurance Companies Already Deploying AI to Deny Claims Faster — Futurism
- NAIC Health Insurance AI/ML Survey — National Association of Insurance Commissioners
- Florida SB 794 (2025) Mandatory Human Reviews of Insurance Claim Denials — bill history (Failed) — LegiScan
- How AI Is Affecting Homeowners Insurance Claims — AAA Public Adjusters
- AI for Claims Adjusters: 10 Key Examples — magicplan
