By the end of this you will be able to use ChatGPT or Claude to find your way around any policy in minutes, and you will know exactly where to stop trusting it.
A client calls. Am I covered for this?
You open the policy PDF. Ninety-some pages. Somewhere in there is the answer. Maybe on page 3. Maybe in an endorsement stapled to the back.
AI can get you to the right page fast. It cannot be the one who answers the client. Hold onto both halves of that and you'll do fine.
What AI is good at in a policy
Finding, sorting and quoting. A model will read every page, list every form, and pull every exclusion into one place faster than you can scroll.
Judging coverage? That takes the facts of the loss, the whole policy read together, and sometimes state law. That's your job.
It helps to remember how a policy is built. The Kentucky Department of Insurance's guide to understanding your policy walks through the standard parts: the declarations, the insuring agreement, the definitions, the exclusions, the conditions and the endorsements. AI does well when you point it at one part at a time.
Start with the forms schedule
The dec page tells you who, when, how much. The forms and endorsements schedule tells you what the policy actually is.
On a commercial general liability policy you'll see the coverage form, often ISO's CG 00 01, followed by a list of endorsements by form number and edition date. Additional insured endorsements like CG 20 10 and CG 20 37 live there. So do the exclusions that quietly take things away.
Personal lines work the same way. A homeowners policy is built on a form like ISO's HO 00 03, a personal auto policy on a form like PP 00 01, and the state endorsements stacked behind them change the details. Different forms, same habit: find the schedule, then read.
Ask the model to list that schedule first. Everything else hangs off it. ChatGPT and Claude both read attached PDFs; OpenAI's file uploads FAQ lists the current size limits, so check them before you attach a big commercial package.
Attached is a commercial insurance policy for [CLIENT NAME].
1. Quote the declarations: named insured, policy period, limits,
deductibles, premium. Give the page number for each.
2. List every form and endorsement on the schedule with its number,
edition date and title, exactly as printed.
3. Tell me which listed forms are missing from the attached pages.
If something is not in the document, write NOT FOUND. Do not guess.
Step 3 matters more than it looks. Policies arrive with pages missing all the time, and a model will happily summarize the pages it has.
Prompts that make it quote the page
A paraphrase is where errors hide. Covers becomes may cover. Excludes becomes limits. So don't let it paraphrase.
For each question below, answer in this format:
Answer: [yes / no / unclear]
Exact wording: "[quote the policy language]"
Where: [form number, edition date, page]
Changed by: [any endorsement that modifies this wording, or NONE]
Questions:
- [Does the policy include a waiver of subrogation for scheduled parties?]
- [Is there an exclusion for work in a specific state or project type?]
The Changed by line is the one that saves you. Base forms get rewritten by endorsements, and a model reading the base form alone will give you the old answer with total confidence.
Five questions to run on every policy
| Question | Why it matters |
|---|---|
| Is the named insured exactly right, including every entity? | A missing LLC is a missing insured |
| Do the dates, limits and deductibles match what was bound? | The policy can differ from the quote |
| Which endorsements add coverage, and which take it away? | The schedule is where coverage changes |
| What exclusions apply to this client's operations? | Generic exclusions read very differently against real work |
| Are additional insureds, waivers and loss payees on it? | Certificates and contracts depend on them |
The second question is its own discipline, called policy checking. Our guide on AI policy checking shows how to compare the issued policy against the quote and binder.
Where AI gets a policy wrong
Watch for these every time:
- Scanned pages. If the PDF is an image, text recognition can garble numbers. A limit of 1,000,000 that reads as 100,000 changes everything.
- The base form without the endorsement. It reads the CG 00 01 and misses the endorsement on page 61 that rewrites it.
- Two policies in one chat. Last year's policy and this year's renewal blur together. Use a fresh chat for each.
- Confident paraphrase. The summary sounds certain. The wording underneath is not.
- Occurrence versus aggregate. Models mix up per-occurrence limits and aggregates, especially on dense dec pages.
Every one of those is a reason to check the quote against the page yourself.
Never advise from a summary alone
Telling a client what their policy covers is a professional act. Errors there are a familiar source of agency E&O claims, and E&O carriers and the Big "I" write about it often. Utica National's common causes of agents' E&O claims and the Big "I" tips on preventing the most frequent E&O claims are good places to start. This isn't legal advice. Talk to your E&O carrier about your own procedures.
What does that mean for AI? Use the summary to find the language. Read the language. Then answer the client in your own words, with the policy wording behind you and the claim going to the carrier for the coverage decision.
A clean workflow
- Use a business plan with training off. See is ChatGPT safe for insurance client data.
- One policy per chat. Attach every page you have.
- Run the schedule prompt. Chase any missing forms.
- Run the quote-the-page prompt for the question in front of you.
- Open the policy to each cited page and read it yourself.
- Write the answer to the client. Keep the prompt, the output and your notes in the client file.
Questions about a policy often turn into work: a change request, a certificate, an endorsement. That work is what Salience does for agencies. You forward the request, Salience reads it against the policy and prepares the change, and nothing goes out until you approve it.
More prompts for jobs like this live in ChatGPT prompts for insurance agents, and the rest of the task guides are on the AI by task shelf.
Give us a call
Hey, all the tech is great, and innovation is great. There's a lot of responsibility and liability across the entire space, and a lot of people are saying the same words that mean different things. Give us a call. We plug in specifically to your processes and take care of the GDS first to unlock you and bring home the bacon.
Call +1 (916) 866-7867 or book a walkthrough.
Would you rather approve this than type it?
Forward the request or the prior application. Salience prepares the form, flags what's missing, and nothing goes out until you approve it.
Frequently asked questions
Can ChatGPT read a PDF and answer questions?
Yes. Attach the policy and ask. Ask it to quote the exact wording and give the page or form number for every answer, because a paraphrase can drift from what the policy says.
Can AI summarize an insurance policy?
It can map one quickly: who is insured, for what period, which forms and endorsements are attached, the limits and deductibles. Treat that summary as a map for your own reading, not a coverage opinion.
Is it safe to put a client's policy into ChatGPT?
Only on a business plan whose terms say your data is not used for training, with your agency's AI policy behind it. Personal plans have different defaults, so check before a client's policy goes in.
What should I ask AI when reading a policy?
Start with the forms and endorsements schedule, then ask which endorsements change the base form, then ask for every exclusion quoted word for word with its form number. Make it say NOT FOUND when something is not in the document.
Can I send a client the AI summary of their policy?
Not as advice. Read the policy yourself, correct the summary, and make clear that the policy language controls. A summary a client relies on becomes your agency's statement.