ChatGPT for insurance agents: what it does well, what never to paste, and what it can't tell you at renewal

ChatGPT can explain a coverage term in plain words and tidy the follow-up after a review. It cannot tell you what Mrs. Alvarez said about her roof in the February thread. This guide covers what it costs, where it saves an agent time, what to keep out of it document by document, OpenAI's own settings step by step, prompts for client work, and where the client record actually lives.

Short answer: yes, for the first draft and the plain-words explanation, as long as nothing about a client goes in without a decision and nothing it writes leaves unread. It is one tool on the map our guide to AI for insurance agents draws; this page is about that one tool. ChatGPT is open on your desk, and you want to put it to work on client business without a declarations page landing somewhere it shouldn't.

Two things before the detail. This guide comes from Equerry, whose assistant keeps the client record, not the policy: it does not quote, rate, bind or advise on coverage, and it never phones, texts or messages a client of yours. It appears near the end, where it belongs. None of this is legal, compliance or coverage advice; what may go into any tool is set by your agency, the carriers you represent and your state. Every statement about ChatGPT was read on OpenAI's own help center and pricing page, in a browser, on 12 October 2026.

What ChatGPT is, and what it costs an agent

ChatGPT is OpenAI's general assistant: a website, an iPhone app and an Android app that draft, rewrite and read files you upload. It is not a rater, a comparative quoting tool or a carrier portal, and no plan makes it one. The plan settles usage, memory, and whether what you type trains OpenAI's models by default. The dollar figures are OpenAI's; on currency, its Go article says (read 12 October 2026): "All purchases are in USD - we offer a local currency billing in a limited set of countries."

Two details. OpenAI does "not support annual billing or the option to pay for multiple months in advance for ChatGPT Go, Plus, or Pro subscriptions", so a personal plan runs month to month. And two licensed people already meet the Business minimum, where the training default runs the other way. Neither settles whether a client's information may go in; that comes further down. The same questions in two neighboring trades: ChatGPT for real estate agents and ChatGPT for accountants.

Where ChatGPT saves an agent time

It is quickest at the words around the sale: a first version of something you could write yourself, or material with no client in it.

• A coverage term in plain words, with the policy as the authority. Replacement cost against actual cash value, what it means for a ring to be scheduled, what separates a named-perils form from an open-perils one. Ask for the explanation plus a built-in line that the client's own policy wording governs; you still open their form before answering.
• Your renewal-review questions. Last year's notes on a household, names replaced, become this year's list of things to confirm. It will not know what changed; it will stop you forgetting to ask.
• The follow-up after a call. Three lines of what the client asked and what you owe them become a tidy email containing nothing you didn't put in.
• A carrier bulletin, mapped. Paste a form-change or guideline bulletin in full and ask what changed, from when, and which accounts it touches, with the sentence behind each point. Some carriers mark bulletins confidential; check before one goes into an outside tool.
• Anything with no client in it. The website page on how an independent agency is paid, a plain-words note on what happens after a claim is reported, the agenda for Monday's producer meeting.

Where it falls short

• Ask it for a number and it gives you one. Nothing connects it to a carrier's rating engine, so the premium it offers is made up, and a coverage answer arrives with the same confidence whether or not it has seen the form. Price questions go to the carrier's quote; coverage questions go to the issued policy, and to the carrier where the wording leaves room. Whatever you repeat, the client will remember as yours.
• It does not know your forms or your state. It describes insurance in general; the carrier's endorsements, your state's rules and the declarations page decide the case.
• You sign what it drafts. Once a letter leaves under your name, nobody asks who typed it; the client and the carrier read it as your statement about coverage. The draft stays a draft until you have read every sentence that touches what is or isn't covered and rewritten the ones you would not have written yourself.

What never to paste, document by document

An agency's files are mostly other people's personal information, so the useful question is not "is ChatGPT safe" but "what is in this document". The questions to settle first, your agency's policy, the carriers' rules and the primary-source reading list on the privacy and telemarketing rules, live on the hub page and are not repeated here. This table sorts what is on your desk.

Two habits make the table livable. Keep a redacted copy of the documents you reach for most. And for help reading a policy, paste the standard form text, which carries no client, rather than the issued policy, which carries everything.

Connecting an account is not the same as pasting one document. OpenAI's Memory article counts "content from connected apps, such as Gmail" among the sources memory can draw on, by plan and region, and says disconnecting an app "does not delete past conversations that already used app content." An agency inbox is a year of client files; can ChatGPT read my emails covers what the connection reaches and how to undo it.

ChatGPT's own settings, step by step

Three settings decide what happens to a conversation. The wording and the steps are OpenAI's, read 12 October 2026; your own screen is the authority if it differs.

1. Keep a one-off question out of history and memory. "Open a new chat." "Select Temporary." "Choose Personalized or Unpersonalized before you send your first message." Choose Unpersonalized, which "Does not use memory, custom instructions, or plugins." "You cannot change personalization after the conversation starts," and "OpenAI may keep a copy of a temporary chat for up to 30 days for safety purposes." Saving it later "converts it into a regular chat, including when it started without personalization." Detail in ChatGPT Temporary Chat.

2. Stop new conversations training the model. "When Improve the model for everyone is off, your new conversations won't be used to train OpenAI models." Web, signed in: "Open your account menu." "Select Settings." "Select Data controls." "Select Improve the model for everyone, turn it off, and select Done." iPhone and Android: "Open the sidebar." "Select your profile icon to open Settings." "Select Data controls." "Turn off Improve the model for everyone." The switch "does not delete or hide saved chats," and feedback reverses it for one chat: "the entire conversation associated with that feedback may be used to train OpenAI models." Never give a thumbs-up or thumbs-down on a chat that mentions a client.

3. See what it remembers. "Open Settings." "Select Personalization." "Select Memory." The controls there "may include Memory, Reference saved memories, Reference chat history, Memory summary, Manage, or Saved memories," by account. The summary is "a high-level view" that "does not necessarily include every detail or source that Memory can reference," and "Deleting a chat alone does not necessarily delete a separate saved memory created from that chat." If a client's details have already reached a regular chat, how to delete ChatGPT memory covers the clean-up.

OpenAI keeps the levers apart: "Memory controls how ChatGPT personalizes responses. Model-training controls determine whether eligible content can be used to improve OpenAI models." Neither makes a conversation confidential in any legal sense; they are facts for whoever sets your agency's policy.

ChatGPT prompts for insurance agents' client work

Four prompts, each with one job, placeholders where a client's name or policy number would go, and an output you can verify. Fill the brackets with a placeholder or a general case, never the real account.

A coverage term, with the policy as the authority

"Explain the difference between [replacement cost and actual cash value] to a homeowner in under 150 words, in plain words, with no term left undefined. End with one sentence saying that their own policy's definitions and declarations page govern and that I will confirm the detail against them. Give no dollar figures or percentages."

Then check: every sentence is true of insurance in general, and none of it answers this client's coverage question. That answer comes from their policy.

This year's renewal-review questions

"Here are my notes from last year's review of a household, with names and addresses replaced: [notes]. Write the questions I should ask at this year's review, grouped by what might have changed: the home, vehicles and drivers, work done from home, valuables, travel, and liability exposure. Do not recommend any coverage, limit or product."

Then check: nothing it wrote is a recommendation, and add the question only you know to ask, the one the client mentioned in passing last time.

A reading map for a coverage question

"A client has asked whether [a detached workshop they rent out at weekends] would be covered under a standard homeowners policy. Do not answer the question. List the parts of the policy I should read to answer it myself: the definitions, the coverage sections, the exclusions, and the kinds of endorsement that commonly change the result. End by saying that the issued policy and the carrier decide."

Then check: open those sections in the client's actual form. The answer you give is yours.

The follow-up after a call

"From these call notes, with the client as [Client]: [notes]. Draft a short, warm follow-up email confirming what [Client] asked me to look into, what I will send, and by when. Do not describe what any policy covers, and quote no figure."

Then check: the date you promised is the date in the email, and nothing about coverage crept in.

None of them asks for a premium, a coverage determination, a recommendation about a real client, or an email to send unread.

What ChatGPT can't tell you at renewal: what each client said

The gap is not forgetting. ChatGPT's memory is real; on a personal plan it can draw on past chats, saved memories, uploaded files and connected apps. The gap is control over what reaches it: you typed, pasted or uploaded it, or you connected an account and it reaches whatever that account reaches. Typing keeps you in charge and falls behind on a busy day. Connecting keeps up, and hands over the mailbox rather than the one thread. What you can audit afterwards is "a high-level view", beside this line of OpenAI's: "Memory does not retain every detail from every conversation. ChatGPT decides which available information is relevant to a response, and Memory can change as your context changes."

Here is the renewal problem in one thread. In February Mrs. Alvarez emails about a billing question. You answer, she replies, and in the fourth message she mentions in passing that the roof is being replaced in April and that her son will be driving by summer. The thread ends. In September the carrier's renewal offer arrives, and the two facts that belong at that conversation sit in a billing thread nobody will reopen. Nothing reached ChatGPT, because nobody pasted it, and connecting your mailbox to catch it would hand over every other client's thread as well. Your agency management system holds the policy and the renewal date; it holds the roof only if someone typed it in that afternoon. The habits that close this gap without a new tool are on the hub page, under keeping what each client told you.

Where Equerry fits

Equerry is a personal assistant who helps you get things done. In an agency its job is that thread, not the policy: it does not quote, rate or bind, gives no coverage or suitability advice, never phones, texts, messages or emails a client, and connects to no agency management system, CRM, rater, carrier portal or Outlook. It holds what you send it and answers from that.

Take the February thread. Equerry comes with an email address of its own, at mail.getequerry.app. Had it been on the CC line of Mrs. Alvarez's billing thread, the fourth reply would have arrived like the others: replies that keep the address on the thread keep arriving. It never asks for your inbox, and the original stays in your own mailbox. A thread forwarded after the fact works the same way. What Equerry remembered from it is listed in Settings → Your memory, each line marked as from that email, and you can delete any line.

Three Routines, exactly as they work. A Reminder is a push on a date you pick, once or repeating, carrying the text you wrote; Equerry does not work out renewal dates, so enter the date yourself when the offer lands. A Watcher runs on a schedule you choose, hourly, every six hours or daily, over what you have sent it, the web, or both, and pings you when a run finds something new on its topic; it does not fire the moment an email arrives. With calendar access allowed and Meeting Prep set up, a short brief lands before a client review, built from what you have shared.

The limits, plainly. Equerry holds only what reaches it; a thread nobody forwarded is not there. Everyone on a CC'd thread can see the address, so where an extra name would be awkward, forward the email instead. Making what you send searchable is work done in part by the outside providers named on our AI disclosure page. And Equerry claims nothing under any insurance, privacy or telemarketing rule: sending it a client's details is the same decision as pasting them into ChatGPT, taken under the policy your agency sets, what your carriers require and what your state allows. The difference is scope: the threads, notes and files you chose, instead of a connected account.

Keep ChatGPT for the plain-words explanation and the first draft, with the settings above. Use Equerry for what each client told you between one renewal and the next. You can find it on the App Store for iPhone and iPad.

Sources, each read 12 October 2026 in a browser (OpenAI's help center turns away automated readers). OpenAI Help Center: What is ChatGPT Plus? (Plus price; no annual billing); About ChatGPT Pro tiers (three Pro plans, US prices); What is ChatGPT Go? (Go billing, purchases in USD; no annual billing for Go, Plus or Pro); Data controls in ChatGPT (training setting and steps, feedback, saved chats, managed workspaces, memory and training as separate settings); Temporary chat in ChatGPT (steps, personalization, retention, saving); Memory in ChatGPT (sources, settings path and controls, memory summary, deletion, connected apps). OpenAI, ChatGPT pricing (plan features, per-user pricing, the Business minimum). OpenAI revises these pages often, and the current page outranks this one wherever they differ. General information only; not legal, compliance or coverage advice.

Last updated 2026-10-12.

All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.