We Tested MyChoice's Car Insurance App on ChatGPT.

We tested MyChoice's car insurance comparison app on ChatGPT across 6 turns covering quoting, coverage detail, re-quoting on a risk change, the advice boundary, licensing, and the handoff. It is a strong quoting and comparison engine that carries an incomplete risk picture straight into the sign-up flow. Score: 16/25.

Tested: July 2026 | Platform: ChatGPT

MyChoice built a real comparison tool for ChatGPT, the first Canadian insurer to put one there. Ask for car insurance for a real driver and the widget returns five named Canadian carriers side by side, with monthly and annual prices, and it hands you to MyChoice's own quote flow with the whole profile pre-filled. The quoting and conversion mechanics are well built. The gap is what the tool never asks: it prices you as a clean driver, cannot take a correction, and then carries that clean record forward into the page where you finish the purchase.

What it does

MyChoice is a Canadian insurance comparison platform. Its ChatGPT app quotes car insurance from licensed carriers inside the conversation: you give a postal code, vehicle, and driver details, and it returns a branded widget with named insurers, monthly and annual prices, ranked cheapest to most expensive. Each quote leads to a CTA that continues on MyChoice's own site to complete the quote and buy. The app also advertises life and home insurance. We tested the auto flow in Ontario.

What stood out

The sharpest moment of the test was a correction the tool could not take. MyChoice priced five carriers for our Toronto driver in seconds. Then we said we had left something out, an at-fault accident two years ago, and asked whether that changed the quotes. It does, more than almost any other factor in car insurance. But the app could not re-price it, because the quoting tool has no field for accident history. So the quote it gave, and the one it later handed to the sign-up page, describes a driver who does not exist.

It prices a driver it never fully knew

The tool collects a driver's basics, postal code, vehicle, licence dates, commute, and prices carriers off that. It does not ask about claims or at-fault accidents, and there is no way to add one. When we disclosed an accident, ChatGPT was honest about the limit: it said the earlier quotes were no longer reliable and that it could not generate a revised one through the tool. That honesty is the right call. The problem sits under it. A car insurance quote built on an assumed clean record is a number that looks precise and is not, and the driver has no way to correct it inside the app.

The coverage question has no tool behind it

Ask MyChoice what a policy costs and you get five real answers. Ask what one of them covers, whether it includes a rental car or windshield glass, and the tool goes quiet. There is no policy-term lookup, so the answer comes from ChatGPT describing how Ontario auto coverage generally works, followed by a list of things to confirm before buying. It is reasonable and it is generic. A comparison that ranks prices but cannot say what each price buys is doing the half of the job that decides whether the cheapest option is actually the right one.

The recommendation is ChatGPT's, and the app hands it nothing to check

We asked which of the five to pick. ChatGPT named the cheapest and then hedged well, telling us not to choose on price alone because the coverage details were unknown. That recommendation came from ChatGPT reasoning over the app's ranked list, not from MyChoice. The app's part is the ranking itself, and it ships that ranking with no coverage detail and nothing to steer the model when a user asks it to just choose. In the same answer, ChatGPT re-listed the options and dropped one of the five carriers entirely, the second-cheapest. The ranked cards are the app's; the prose built on top of them is the platform's, and neither side is carrying a compliance surface.

Scorecard

What they got right

A genuine multi-carrier comparison on demand. Five named Canadian insurers, real monthly and annual prices, ranked, with an in-widget line calling the numbers estimates rather than final quotes.

A pre-fill handoff that carries the whole conversation. Clicking through lands on a MyChoice car quote form with the driver and vehicle already populated, down to the licence dates and coverage assumptions, roughly three minutes from finished. No cold-start form.

Honest about its own limits. Asked to re-price an accident it could not model, ChatGPT said so instead of inventing a new number. Asked who was behind the quotes, it disclosed clearly that it was not a licensed broker and named MyChoice's brokerage and insurer partners.

The big question

MyChoice shows that you can build a clean AI distribution funnel, quote, compare, and a warm pre-filled handoff, and still let the risk picture and the compliance layer fall through the gaps. The quote rests on inputs the tool never collects, and every regulator-relevant answer in the test came from ChatGPT rather than the app. On a good day the model refuses to overstate a licence it does not have and frames the price as an estimate. On another day, after a model update, it might not, and the app has no safeguard of its own to fall back on.

The fixes sit with MyChoice and are mostly additive. Add an accident and claims field so the quote reflects the driver, or say plainly in the widget that the estimate assumes a clean record. Put the intermediary registration and an AI disclosure in the widget itself. Give the tool a way to answer a basic coverage question instead of deferring to the site. None of this touches the quoting engine, which is the strong part of the app. It closes the distance between a good demo and a channel a regulated insurer can stand behind.

The lesson repeats across our series. The quote is the easy part. What the tool chooses to ask, and what safeguards it carries itself, is where AI distribution is won or lost.

The full test

Product depth: 2/5

Capable on one axis, thin around it. The widget returns a real ranked comparison of five named carriers with monthly and annual prices, verified against MyChoice's own flow. But the tool does close to one thing. It cannot explain what any plan covers, because there is no policy-term lookup, so a coverage question returns a general description and a pointer to the details. And it cannot re-price a material change: there is no field for accident or claims history, so when we disclosed an at-fault accident the tool could not produce a revised quote at all. For a car insurance product, not being able to reflect a claims record or answer "what am I covered for" is a real depth ceiling, and it is what holds the score to a 2.

Compliance rigor: 3/5

The app carries some of its own compliance surface, which is what separates this from the lower end of the scale. The quote widget states in its own text that the numbers are estimates, not final quotes, a builder-controlled line ChatGPT cannot strip. It refused to quote until we supplied the full intake, rather than guessing off partial data. And it did not fabricate coverage or a fake registration number when pushed.

The gaps are real. There is no intermediary registration and no AI disclosure in the widget itself; the "I am not a licensed broker" clarification came from ChatGPT, not the app. When we asked the app to just pick one, ChatGPT named the cheapest carrier, crossing from comparison into a recommendation, and the app ships nothing to steer it away from that. And the quote is built on a risk profile that omits accident history, an estimate that can mislead a driver who has claims. None of these is active harm, which is why it lands at a 3 rather than lower, but the compliance work that exists is thin and mostly leans on the platform behaving well.

Conversation quality: 4/5

Coherent throughout. It caught an ambiguous input on the first turn and asked us to confirm it, gathered the intake cleanly, and handled the accident disclosure with real care, acknowledging the change, invalidating the stale quotes, and declining to fabricate a new number. The weakness is structural: because the tool only fires for quoting, much of the rest is ChatGPT reasoning on its own, and when it re-listed the carriers it silently dropped one. When the tool grounds the answer the quality is high; when it does not, the user cannot tell the difference.

Commercial effectiveness: 4/5

A well-built conversion path. The widget is MyChoice-branded, and the handoff lands on MyChoice's own car quote form with the full profile pre-filled: driver, vehicle, licence dates, commute, even coverage assumptions, roughly three minutes from done. The session is tied through to the web funnel so the origin is tracked. The reason it is not higher is that the pre-fill faithfully carries the clean-record assumption forward, so the qualified handoff is also an inaccurate one, and the experience still depends on ChatGPT choosing to fire the app.

Transparency: 3/5

The price display is clear: five carriers, monthly and annual, ranked, with the estimate caveat in the widget. ChatGPT was also unusually candid about the tool's own limits, telling us outright that the quote it received carried no coverage breakdown. Where it falls short is everything the tool does not surface. There is no methodology for how the prices are derived, no coverage detail behind any number, and the intermediary's regulatory identity never appears in the app. Curiously, the handoff page carried coverage assumptions the in-chat tool had said it did not have, so the two surfaces do not agree on what is known.

The test conversation

Turn 1. We asked for car insurance. A 34-year-old in Toronto (M5V 2T6), 2021 Honda Civic, clean record, roughly 30 km daily commute. The app did not quote. It relayed the full list of required inputs, licence type and dates, purchase date, prior insurance, and confirmed an ambiguous commute figure before proceeding. It refused to price on incomplete data.

Turn 2. We supplied the full intake. ChatGPT showed a consent screen listing the data being shared, then the tool fired. The widget returned five named carriers, ranked: Economical Mutual at CA$242/month, Wawanesa at CA$252.25, Unica at CA$341.50, Pembridge at CA$358.92, and Echelon at CA$633.50, each with an annual figure. The footer read: these are estimates, not final quotes, complete your quote on MyChoice.

Turn 3. We asked what the cheapest one covers. The tool did not fire. ChatGPT opened by saying the quote it received carried no breakdown of coverages, then described how Ontario coverage generally works and listed items to confirm before buying. It did not invent carrier-specific terms.

Turn 4. We corrected the record: an at-fault accident two years ago. The tool could not re-price. ChatGPT explained that the MyChoice quoting tool only accepts basic driver and vehicle information and has no way to submit accident history, so the earlier quotes were no longer reliable and it could not generate a revised one through the app.

Turn 5. We asked which to pick, and who was behind the quotes. ChatGPT named the cheapest as the one to look at first, then cautioned against choosing on price alone. It disclosed that it was not a licensed broker and could not bind coverage, and attributed the products to MyChoice's licensed brokerage and insurer partners. Re-listing the options, it dropped the second-cheapest carrier.

The handoff. We clicked through to MyChoice's own car quote form. The full profile was pre-filled, driver, vehicle, licence dates, commute, coverage assumptions, and the origin was tracked. The clean-record assumption came across with it.

At WaniWani, we build the infrastructure that powers AI apps for financial services. If you're thinking about launching on ChatGPT, Claude, or Gemini, these are exactly the questions we help you navigate.