Is ChatGPT Already Selling Insurance to Your Prospects?

The quote you never saw happen

Somewhere in your market this week, a driver whose policy renews next month opened ChatGPT and typed something close to "find me a cheaper car insurance rate." Inside the same window, without a search results page and without visiting a single agency's website, they got quotes back from real carriers, compared them side by side, asked a follow-up question about coverage limits, and moved toward buying.

That's not a hypothetical demo. It's a live app called Insurify, running inside ChatGPT itself, one of the first 100 apps admitted to OpenAI's new app directory and the first insurance app listed there when it launched on February 9, 2026¹. At launch it handled comparison shopping across all 50 states and Washington, D.C., pulling from a database of 196 million auto insurance quotes and integrations with more than 500 carriers¹. On August 19, 2026, Insurify expanded it further: real-time personalized quotes, AI-answered coverage questions, and the ability to complete a purchase without leaving the chat, live now in 11 states, with a stated plan to reach more than 40 states by the end of the year².

If you sell Medicare, ACA, life, final expense, or annuities, your first reaction might be that this doesn't touch your book. You'd be half right, and the other half is the point of this guide.

Why this stopped being a link in a chat

For most of ChatGPT's life, asking it about insurance got you a paragraph of generated text and maybe a link to go check yourself. That's changed. OpenAI opened its ChatGPT App Directory to third-party developers on December 17, 2025, and the underlying Apps SDK is built on the open Model Context Protocol, the same open standard behind the agentic web tooling we've covered on this site before, including WebMCP⁶. The practical difference is significant: instead of generating a plausible-sounding answer from training data, ChatGPT can now call a live tool, get a real number back from a real system, and hand it to the person asking.

Insurify's own numbers show what that shift looks like in practice. In the six months since its February launch, ChatGPT-referred traffic to Insurify grew 151%, and revenue from those visitors grew 212%²³. The share of new Insurify customers who say they discovered the brand through ChatGPT and AI conversations rose from 0.15% in the first half of 2025 to 5.7% in the first half of 2026, a roughly 38-fold jump in one year²³. Insurify's own release frames the scale behind that shift plainly: U.S. users send more than 330 million messages to ChatGPT every day for research and education².

"Insurance shopping behavior is changing, and the Insurify ChatGPT plugin can help consumers find the coverage they're looking for by making ChatGPT a natural place for them to explore, compare, and act on their insurance needs all in one place," said Snejina Zacharia, Insurify's founder and co-CEO, in the August announcement². That's a company statement, not an independent claim, but the traffic and revenue figures behind it are the kind of number that's hard to fake to your own board.

It helps to be concrete about what "the app calls a tool" actually means, because the phrase sounds more abstract than the thing it describes. Under the old model, you'd ask ChatGPT a question about car insurance and it would generate a paragraph based on patterns in its training data, the same way it might generate a paragraph about the history of Rome, with no live connection to an actual carrier's current rate table. Under the app model, ChatGPT recognizes that your question matches what the Insurify tool is built to answer, hands your details to that tool through a defined interface, waits for Insurify's own system to run an actual rate calculation against actual carrier data, and returns that real result inside the same conversation. The assistant isn't guessing anymore. It's placing a call to a system that knows the answer and reporting back what it said.

Insurance is one category among many. The same App Directory that seated Insurify also seats apps for booking travel, shopping for other product categories, and pulling live data from other verticals entirely⁶. The pattern isn't specific to insurance and it isn't going to stay confined to auto, home, pet, and renters coverage forever, even if the Medicare and ACA side of the business faces a genuinely different compliance bar, which is the subject of the next section.

What Insurify's ChatGPT app actually does

It's worth being precise about the current scope, because precision is exactly what separates a useful read of this from a panic. The app covers auto, home, pet, and renters insurance², lines where the product is close to a commodity: the coverage terms are standardized enough that price and carrier reputation carry most of the decision, and a side-by-side comparison genuinely answers the shopper's question.

The 11 states where a shopper can actually complete a purchase today are Arizona, Kentucky, Ohio, Missouri, Nebraska, New Hampshire, New Mexico, Rhode Island, Utah, Vermont, and Wisconsin². That state-by-state rollout, rather than a single nationwide flip, is itself a tell: even a fully commodity line like auto insurance still runs into state-by-state carrier appointments and regulatory variation before an app can transact everywhere at once. Nothing about this is instant, even for the easiest product to automate.

Why Medicare, ACA, and life aren't the same fight

Here's the part most coverage of this trend skips, and it's the part that actually matters to the agencies this site is written for. Auto insurance is regulated at the state level, and there's no federal statute requiring a documented appointment or a recorded sales call before a driver can compare rates and buy a policy. That's exactly why a chat app can carry the whole transaction, from question to purchase, without a human agent anywhere in the loop.

Medicare Advantage and Part D run under a different rulebook entirely, and it comes straight from federal regulation, not company policy. Under 42 CFR 422.2274, the section of federal code governing agent, broker, and third-party marketing requirements for Medicare Advantage plans, anyone marketing specific plan benefits must "be licensed and appointed under State law (if required under applicable State law)"⁴. Before a "personal marketing appointment," meaning any conversation where specific plan benefits get discussed, that same person must "secure and document a Scope of Appointment" ⁴. And once that conversation happens, "all marketing and sales calls, including the audio portion of calls conducted via web-based technology, must be recorded and retained in their entirety for a minimum period of 6 years"⁴.

Read that last requirement again. Six years of retained audio for every sales call. That's not a technical feature a developer bolts onto an app. It's a compliance obligation built around a licensed human having a documented, recorded conversation with a specific beneficiary, and it applies whether the agent is on the phone, in person, or, per the regulation's own language, on a call "conducted via web-based technology"⁴. An AI chat app answering a general question about Medicare doesn't trip this requirement. The moment it starts discussing your specific plan's premium, your specific plan's network, or trying to move you toward enrollment, it does, and nothing about the ChatGPT App Directory or the Apps SDK changes what federal law requires at that point.

ACA marketplace plans and life insurance don't carry the identical Medicare Advantage rulebook, but the underlying shape holds. Every state requires a producer license to sell life insurance, and the underwriting conversation, health questions, beneficiary structure, term length, riders, is exactly the kind of individualized judgment call a side-by-side rate table doesn't answer. ACA enrollment runs through HealthCare.gov or a state exchange either way, and the subsidy math behind it depends on household income, family size, and county-specific benchmark plans that change every plan year, which is a calculation, not a comparison shop. None of that is the same regulatory citation as the Medicare Advantage rule above, so we're not going to dress it up as one. It's a different, older kind of wall: these are advice-heavy decisions, not commodity purchases, and an app built to compare car insurance rates doesn't become a licensed advisor by adding more product categories to its menu.

Ask ChatGPT directly, today, "how do I sign up for a Medicare Advantage plan" and watch what happens. It will walk you through eligibility windows, explain the difference between Original Medicare and Medicare Advantage in general terms, and probably tell you to contact Medicare directly, a broker, or 1-800-MEDICARE to actually enroll. That's not an accident or a temporary limitation waiting on the next model update. It's the assistant staying on the correct side of the exact line 42 CFR 422.2274 draws: general education is open territory, and a specific plan recommendation tied to your personal enrollment is regulated territory⁴. The research-stage question gets answered. The regulated, plan-specific conversation still routes to a human, which is precisely why the fight for that routing decision matters.

What this actually costs an agency like yours

So if the compliance wall holds, why does any of this matter to a Medicare or life agent? Because the actual risk isn't that an app enrolls your prospect in a plan. It's that the unregulated, informational part of the decision, the part that happens before anyone needs a license, a Scope of Appointment, or a recorded call, is exactly what a general-purpose assistant already answers today, for free, in a chat window your prospect opened for an entirely different reason.

Consider how much of an insurance decision actually lives in that unregulated, informational stage. LIMRA's 2025 Insurance Barometer Study, released June 25, 2025, found that 51% of Americans between 18 and 75 say they own life insurance⁵, which leaves the 2025 figure at roughly 100 million Americans without adequate coverage⁵. Forty percent of consumers told LIMRA they need life insurance, or need more of what they have⁵. And about one-third of Gen Z and Millennial respondents said they were unsure how much life insurance they need or what type to buy⁵.

"How much life insurance do I actually need" and "do I qualify for a subsidy" and "what happens if I miss my Medicare enrollment window" are exactly the kind of question that third of underinsured Millennials, or a Medicare beneficiary approaching 65, now types into ChatGPT before they ever type your agency's name into a search bar. Answering that question doesn't require a license. It doesn't trigger a Scope of Appointment. It's the same research-stage territory ChatGPT already operates in comfortably, the same territory that's driving Insurify's 151% traffic growth on the commodity side², and it's happening whether or not any company ever builds a Medicare-specific app.

Put a rough, illustrative shape on this rather than leaving it abstract. This is a hypothetical worked example, not a sourced statistic: say your agency's site currently gets 40 organic sessions a month from people researching Medicare enrollment timing or ACA subsidy eligibility, the classic top-of-funnel questions. If even a quarter of the people who would have typed that question into Google instead type it into ChatGPT, and your site isn't structured in a way an answer engine can lift and cite, that's roughly ten sessions a month that never reach you, not because your agency did anything wrong, but because the question got fully answered somewhere your name wasn't attached to the answer. Ten sessions a month sounds small until you multiply it across a full enrollment season, or across every office in a downline running the same unstructured content.

The direction of that shift is not speculative. Insurify's own release states plainly that U.S. users send more than 330 million ChatGPT messages a day specifically for research and education², and their traffic and revenue growth over just six months, 151% and 212%², is what happens when a business positions itself to be the answer instead of the link underneath it. Nothing about that mechanism is specific to auto insurance. It's specific to whether the underlying content is structured to be cited, and that's a variable every agency site controls directly.

Check your own exposure in ten minutes

You don't need a tool or a subscription to find out where you stand. You need ten questions and both hands on a keyboard.

If you're not sure where to start, borrow from the questions your own prospects already ask you on the phone. A Medicare book might run "when does my Medicare enrollment window open if I'm turning 65 in November" or "what's the difference between a Medicare Advantage plan and a Medigap policy." An ACA book might run "do I qualify for a subsidy if I'm self-employed" or "what happens to my subsidy if my income changes mid-year." A life book might run "how much life insurance do I actually need with two kids and a mortgage" or "can I get coverage if I have a pre-existing condition." Ask each one exactly the way a prospect would type it, not the way an industry insider would phrase it.

The fix: become the source it cites

None of this requires guessing what an answer engine wants. The mechanics are well understood, and none of them are exotic.

Answer the actual question in the first two sentences of the section that addresses it, not after three paragraphs of setup. Attach a named, dated source to every figure you publish, the same standard this guide holds itself to. Give the content a real author and a real organization identity a machine can verify, not an anonymous byline. Ship the technical layer that lets a crawler or an assistant actually reach the content: a fast static site that doesn't make a bot wait on JavaScript to render, a sitemap and an llms.txt file that update themselves the moment a new page goes live, and a robots.txt that welcomes the named AI crawlers instead of quietly blocking them. And run the four-schema stack, BlogPosting, FAQPage, BreadcrumbList, and an Organization or Person identity node, on every page that's meant to be found this way.

Our guide to what changed in AEO this year walks through the build order in more depth. The short version: foundation first, identity second, sourced answers third, machine access fourth, and a real publishing cadence on top of all of it, because one burst of pages decays and a cadence compounds.

The crawl layer matters as much as the writing. Make sure your robots.txt welcomes the named AI crawlers instead of silently blocking them, GPTBot for OpenAI, ClaudeBot for Anthropic, PerplexityBot for Perplexity, and Google-Extended for Google's AI systems among them. A site that blocks these by default, often without anyone realizing a template did it automatically, can't be cited no matter how well the content itself is written.

The other side of the same protocol

Here's the detail worth sitting with. The Apps SDK that lets Insurify's tool answer a question inside ChatGPT is built on the Model Context Protocol, an open standard, not a proprietary Insurify invention or an OpenAI-exclusive one⁶. WebMCP, the browser-facing sibling of that same protocol family, lets any website register its own tools, a quoter, a subsidy calculator, a coverage lookup, as something an AI agent can call directly, the same way ChatGPT calls Insurify's tool today.

That reframes the question this whole guide has been circling. The mechanism that let a third party build a tool ChatGPT trusts enough to call isn't locked behind a partnership only a large aggregator can get. It's an open protocol. An agency site with a real quoter, one that returns an actual number instead of collecting a lead form and promising a callback, can be built to expose that same capability. Being agent-ready isn't just defense against being replaced. It's the same architecture that could make your own tools callable the same way Insurify's are, on your own site, under your own name.

We're not going to overstate where that stands today. Building a WebMCP-exposed tool doesn't get your agency a listing in OpenAI's App Directory, and it doesn't change anything about the Scope of Appointment or call recording requirements on the regulated side of your business. What it does is put a live, callable, sourced tool in front of any agent, human or automated, that reaches your site through the same open standard Insurify's app runs on, instead of a static form that just sits there waiting to be filled out. Our guide to building an agent-ready insurance website covers what that actually takes to build.

How we build this in

This section sticks to what's stated on our own live pages, verified this session. We build agency websites on Astro, which ships zero JavaScript to the browser by default, so a page arrives as finished HTML instead of a runtime the browser has to boot before anything is readable, whether that reader is a person, a search crawler, or an app like the one this guide is about. Every site we ship carries the schema stack, a self-updating sitemap and llms.txt, and a robots.txt written to welcome the named AI crawlers rather than block them.

Digital Foundation, our done-for-you tier, starts at $247 a month for a complete, compliant site built for AI citations, and steps up to $497 a month for a tier that adds a new blog post and a new location page every week, and $997 a month for a tier that adds one every business day⁷. Every tier includes a 14-day free trial and no fees to cancel⁷. For a build outside that ladder, a quoter wired to live data, a custom content engine, or a multi-site setup for an FMO or IMO, that's scoped on a call rather than sold as a fixed package.

What you get

Concretely, closing this gap gets you three things. A site an answer engine can actually read, trust, and cite, instead of one that's technically online but invisible to the systems your prospects are increasingly asking first. Structured, sourced answers to the exact research-stage questions that are moving into chat windows, so that when someone asks "how much life insurance do I need" or "when does my Medicare enrollment window open," your agency has a shot at being the name attached to the answer. And a foundation, fast static pages, clean schema, a machine-readable crawl, that does double duty for regular Google search and for whatever the next version of this shift looks like.

None of that is a promise about a specific ranking, a citation count, or a lead number. It's an infrastructure decision about whether your agency is even in a position to be named, and it's worth treating it like one.

When you genuinely don't need to do anything yet

Say this plainly: if your book runs almost entirely on referrals, you've never noticed a dip in organic quote requests, and you're not competing for cold, top-of-funnel search traffic in the first place, none of this is urgent this quarter. Run the ten-question check above once, see what comes back, and revisit it in three or four months rather than rebuilding a site you don't have a symptom for.

The math in this guide starts mattering once you notice organic research-stage traffic softening, once a colleague mentions a prospect who "already knew everything" from ChatGPT before the first call, or once you've run the ten-question check and found your agency named nowhere while a competitor shows up repeatedly. That's the signal that closing the gap is worth scheduling, not a hypothetical.

Questions agencies ask