What does an insurance app in ChatGPT do?
Insurance distribution leads: this guide settles whether your products need an insurance app in ChatGPT. Score one product in six questions and you get a build, partner or wait answer you can defend to compliance and the CFO.
A shopper asks "what would car insurance cost me in Ohio", ChatGPT calls the insurer's app, which asks a few questions and returns a price. Then the shopper clicks out to buy. Of nine launch releases we read, the eight that say where the sale finishes all send the buyer to a site the seller controls.
The demand evidence is thin. The only published figures we found are Insurify's own, from its 19 August 2026 release and customer survey. Leapbuzz has not audited them.
Our position: the first movers have proved an assistant can work as a quoting surface, not that every carrier needs one. For the wider channel, see our ChatGPT marketing work.
Which insurers already quote inside ChatGPT?
At least nine insurers, brokers and comparison sites quote inside ChatGPT as of 26 October 2026, by our count from their own releases; others may exist. Four are carriers pricing their own products. The other five are intermediaries: four comparison sites and marketplaces, and a small-business broker.
The early designs split on how much data to take inside the chat. Insurify, a comparison site that grew up as an insurtech, called its 9 February launch the "industry's first ChatGPT app". Simply Business went minimal: its 23 April release returns an indicative price (a figure to anchor on, not a quote you can buy) and says no personally identifiable information is collected in the chat.
Aviva sits between them. Its 24 March release announced an app, due to launch in the next few weeks, that quotes one direct product, Signature Home Insurance, with limits and add-ons chosen on Aviva's website; trade reports say April (unconfirmed by Aviva). Its 1 June release extended the app to initial life quotes, pre-filled into Aviva's application.
Three intermediaries joined them:
- Experian (12 February) plans bindable coverage inside ChatGPT.
- BizCover, an Australian small-business comparison site: purchases happen "not within ChatGPT".
- VisitorsCoverage (company-reported, read as a wire copy) compares travel plans.
Liberty Mutual prices from its own rating engine (the system that turns answers into a price) and sends buyers to LibertyMutual.com to bind. Plymouth Rock's 30 July release calls itself "the first U.S. insurer to bring home insurance quoting capabilities directly into ChatGPT", a claim we report as the company's own.
| Company | Announced | What the chat returns | Data taken in the chat | Where the sale happens |
|---|---|---|---|---|
| Insurify (US car) | 9 Feb 2026 | Real-time quotes in 11 states; estimates elsewhere | Driver, vehicle and coverage details | Insurer site or Insurify's agency |
| Experian (US car marketplace) | 12 Feb 2026 | Estimated rates from 37+ carriers | ZIP code first | Experian website |
| Aviva (UK home, then life) | 24 Mar 2026 (announced) | Initial quote for one direct product | Name, address, contact details, cover type | Aviva website |
| BizCover (Australia small business) | 31 Mar 2026 | Public liability and professional indemnity quotes | Not itemised | BizCover platform |
| Simply Business (US small business) | 23 Apr 2026 | Indicative price | Trade, revenue, ZIP code; no PII | simplybusiness.com |
| VisitorsCoverage (travel) | 30 Apr 2026 | Quotes and plan comparison from several carriers | Trip and cover needs | VisitorsCoverage platform |
| Liberty Mutual (US auto) | 28 May 2026 | Quote from its own rating engine, 7 states | "A short series of questions" (not itemised) | LibertyMutual.com |
| Plymouth Rock (US home) | 30 Jul 2026 | Real-time home quote, its operating states | Not itemised | Not stated in the release |
| Manulife CoverMe (Canada travel) | 28 Sep 2026 | Personalised travel quote | Trip questions | CoverMe.com |
State lists, as published
Insurify real-time quotes (19 August 2026): Arizona, Kentucky, Ohio, Missouri, Nebraska, New Hampshire, New Mexico, Rhode Island, Utah, Vermont, Wisconsin.
Liberty Mutual (28 May 2026): Arizona, Kentucky, Ohio, Missouri, New Mexico, Utah, Wisconsin.
Plymouth Rock (30 July 2026): "every state where Plymouth Rock operates, including" Massachusetts, New Jersey, Pennsylvania, Connecticut, New York and New Hampshire.
The pattern worth copying: every carrier-built app opened with auto, home or travel, where rating inputs are few.
What do OpenAI's rules let an insurer do inside the chat?
OpenAI's rules point to the handoff. Its developer documentation now calls these apps "plugins" (the page is titled "Plugin guidelines"). The guidelines for developers, read on 26 October 2026, say plugins "may conduct commerce only for physical goods" and should use external checkout on the developer's own domain. Insurance is not named; our reading is that a policy is a service, so the purchase is a click-out.
Three more rules shape an insurance app:
- Restricted data. No card data covered by PCI DSS (the card-payment security standard), protected health information, government identifiers or login credentials. For life and health lines, that removes most of the form.
- Minimum data. Collect what the tool needs and nothing "just in case". A 40-question web form will not port.
- General audiences. Plugins must suit users aged 13 to 17, so compliance will ask how yours answers a minor who wants a car insurance price.
The same document bans "unregulated, deceptive, or abusive financial services." A licensed insurer is not the target, but expect review questions about licensing. The guidelines carry no revision date; re-read before submission.
Do this first. Before anyone writes code, mark every field of your quote form that OpenAI's restricted-data list keeps off the chat. If more than half the rating inputs are marked, you are looking at an indicative-price app (the Simply Business pattern) or a partner listing, not a real-time quote.
Should a carrier build, partner, or wait?
Most carriers should partner or wait, because the answer depends on plumbing more than ambition.
Build means your own app and listing, and it needs a rating engine you can call through an API in seconds. Partner means your prices appear inside an aggregator's or broker's app that already has the traffic. Being one of the few prices an assistant shows beats owning a page nobody reaches; see our shortlist economy analysis. To make your business discoverable and quotable by AI agents, see Origin Pi, our parent company.
Partnering has a limit, and Insurify drew it on 23 September 2026. It blocked Meta's Muse agent from its marketplace, saying Muse showed prices without limits, deductibles, eligibility conditions or state disclosures.
"A quote without its context is not a fair comparison. It is a number."
Giorgos Zacharia, co-founder and co-CEO, Insurify, in its 23 September 2026 release
Whoever owns the surface decides what context travels with your price, so write that into the partner agreement before the feed goes live.
Wait is legitimate: build an API-ready quote service and a measurement plan now, and revisit in two quarters.
| Factor | Build your own app | Partner with an aggregator or broker app | Wait and prepare |
|---|---|---|---|
| Best fit | Direct carrier, simple high-volume product, API rating engine | Carrier already on comparison panels; complex products | Legacy rating, or a market where OpenAI surfaces are still thin |
| Compliance surface | Your tool outputs plus ongoing tests of the model's wording | Your product data must be accurate in the partner's app | Minimal |
| Main risk | A listing few people use | Price-only comparison squeezes margin | A competitor becomes the default answer |
The row teams skip is compliance surface. Your tool supplies the facts, but the model writes the words, differently each time, so testing replaces one-off sign-off.
Which of build, partner or wait fits one of your product lines? Send the product and your markets through the contact page, and Leapbuzz returns a scored recommendation.
A six-question scorecard: build, partner or wait
Score each question 0, 1 or 2, and be harsh. The goal is to swap "our competitor launched one" for a number compliance has signed. Leapbuzz built it for insurance marketing teams.
- 1. Can your rating engine return a price through an API in under five seconds? 2 = yes, in production. 1 = only behind the web form. 0 = no.
Score
- 2. Can the product be priced on inputs OpenAI allows in the chat? 2 = all rating inputs are allowed. 1 = an indicative price works on a subset. 0 = pricing needs health data, government IDs or payment details.
Score
- 3. Is the product sold direct? 2 = direct-led. 1 = mixed. 0 = brokers only, who a build would undercut.
Score
- 4. Can the web journey accept a handoff? 2 = answers pre-fill and the customer continues. 1 = a deep link to the right product page. 0 = the customer starts again on a long form.
Score
- 5. Can compliance test what the model says? 2 = a script of risky questions, run before launch and after every change. 1 = ad hoc checks. 0 = nobody.
Score
- 6. Can you measure it? 2 = AI-referred sessions tracked to bind, plus a "how did you find us" question. 1 = referral traffic only. 0 = nothing.
Score
9 to 12: build one product. 5 to 8: partner, and fix the lowest-scoring item before revisiting. 0 to 4: wait, and work on questions 1, 4 and 6, which pay off even if you never ship. A 0 on question 2 caps any build at an indicative price. We have not tested these bands against outcomes; use the score to force a conversation, not as a prediction.
Pick a score for each question, then press the button. Without JavaScript, add the six scores by hand and use the bands above.
Formula: total = sum of the six scores, out of 12. The bands are our heuristic, not a benchmark.
A hypothetical direct home insurer scores 2 on the API, 2 on allowed inputs and 2 on selling direct. It deep-links to the plan page (1), has only ad hoc compliance checks (1) and tracks referrals but not bind (1). Total: 9, so build, but fix handoff and measurement first.
Sell the same product only through brokers and question 3 drops to 0. Total: 7, so partner. In Singapore, Malaysia and Australia many books are broker-led, so a 0 on question 3 is common and most readers will land on partner.
What should insurers in Singapore, Malaysia and Australia watch?
The first Australian ChatGPT insurance app we found is BizCover's (31 March 2026), a comparison site, not an insurer. We found no Singapore or Malaysian insurer or broker, and no Australian carrier, with one as of 26 October 2026.
Canada gave the nearest signal. Announcing CoverMe on 28 September, Manulife's chief artificial intelligence officer Jodie Wallis said: "The foundations behind it are enterprise-wide, so what we learn here compounds across our markets." Regional competitors should assume the build will travel.
A price plus a model-written explanation of cover sits closer to advice than a web form does. Whether in-chat quoting counts as advice or arranging in your market is a question for your regulator (MAS in Singapore, Bank Negara Malaysia, ASIC in Australia) and your counsel. Ask it before the product team asks OpenAI. We state no rule; this is not legal advice.
Three other signals would change the decision:
- OpenAI opening in-chat checkout to services. Experian says it plans bindable coverage in ChatGPT. That would remove the handoff and strengthen the build case; see our guide to the Agentic Commerce Protocol and checkout handoff.
- Aggregators listing first in your market. In the US and Australia the comparison sites moved before the carriers. For Australian SME lines partner is already the working default; in Singapore and Malaysia, watch for the same.
- What ChatGPT already says about your category. Many shoppers may never connect an app, so check this before funding a build; an AI visibility review starts there.
How to measure an insurance app in ChatGPT before you trust the numbers
Measure it with your own analytics and tracking, not Insurify's headline. The 212% revenue growth from ChatGPT-referred visitors, per its 19 August 2026 release, is growth from a small base, reported by the company that benefits. We found no carrier that has published its own survey share.
- Tag the handoff link and follow ChatGPT-referred traffic to quote and to bind in your CRM. The insurance CRM and martech stack guide covers the plumbing.
- Add an AI assistant option to "how did you hear about us". Insurify's 0.15% to 5.7% figure came from this kind of self-reported survey question. Imperfect, but more honest than last-click.
- Run a cannibalisation check. Compare bind rate and new-customer share for AI-referred shoppers against direct and comparison channels over the same weeks.
Give it two quarters. If volume is still flat with a working handoff, retire the app and keep the API.
This week, run the restricted-data audit from "Do this first" on your highest-volume personal line and score question 2, then put the engineering time into question 1.