We Tested APRIL Moto's Insurance App on ChatGPT.
We tested APRIL Moto’s motorcycle insurance app on ChatGPT across 11 turns covering quoting, advice, insurer identity, re-pricing, and handoff. Its defining trait is that it refuses to make things up, even when that means stopping the conversation and pointing you to a phone number. Score: 19/25.
Tested: June 2026 | Platform: ChatGPT
What it does
APRIL Moto is a French motorcycle insurance broker. Its ChatGPT app is a single, deep tool: it quotes a real policy. You describe your bike and your situation in plain language, it identifies the model from a catalogue, and it walks a full underwriting interview, vehicle, rider, claims history, before returning a branded widget with three coverage tiers, real monthly and annual prices, deductibles, and priced options. It can pull product documentation when asked, recommend a tier against your stated needs, and hand you off to APRIL Moto’s own site with every field pre-filled. It is built as a quoting and conversion tool, not a general insurance chatbot.
What stood out
The most telling moments in this test were the ones where the app stopped. It caught an input that French licensing law makes impossible and refused to price it. It declined to invent a registration number it could not find. It would not re-price a quote on changed data, and it would not produce a subscription link the conversation did not have. APRIL Moto’s defining trait is that it does not fabricate. The cost of that honesty is that it also surfaces, plainly, the things the app is missing.
It caught a mistake the law says is impossible
Four turns into the interview, after collecting the bike, the rider, and the claims history, we gave it a licence date that does not add up. A French permis A requires at least two years on a prior A2, so a permis A obtained in 2023 with no earlier A2 cannot exist. The app caught it. It refused to recalculate and asked us to reconcile the dates: was there an A2 before, or is the 2023 licence actually the A2. Most quoting tools would have taken the number and printed a price on top of incoherent data. This one validated the input against the rule behind it and stopped.
That single moment reframes the long interview that precedes it. The app asks roughly eighteen questions before it shows a price, which is a lot of friction in a chat window where a user expects a number fast. But the questions are not box-ticking. The data is cross-checked as it comes in, which is the difference between a form and an underwriter.
It would rather stop than invent
The same instinct runs through the whole conversation. When we asked who underwrites the policy, it named the real carriers, Allianz, Wakam, Mondial Assistance, from a product document it retrieved, and correctly placed APRIL as the broker rather than the insurer. When we asked for its ORIAS registration number, it said the information was not in the documentation it could see and pointed us to an adviser. When we asked it to re-price after changing where the bike is parked, it declined to produce a new number and sent us to the phone. At no point did it guess.
This is the right behaviour, and it is rarer than it should be. But the honesty cuts both ways, because what it keeps telling you it does not have are two documents a French broker is required to show. The ORIAS number is mandatory under the Code des assurances. The IPID, the standardised insurance product information document, is mandatory pre-contractual information under the IDD. Neither appears anywhere in the app, not in the quote widget and not on request. The app does not hide this. It tells you, accurately, that it cannot find them. An app that fabricated less carefully would have papered over the gap. This one shows you exactly where the gap is.
The quote is one-shot
The quote itself is real and detailed: three tiers, per-tier guarantees, deductibles of 500 and 650 euros, priced add-ons, a recommendation tied to the cover we said we wanted. What it cannot do is change. Tell it the bike now sits on the street instead of in a locked garage, a fact that genuinely moves the theft premium, and the pricing tool does not re-fire. It explains why the parking matters, correctly cites the deductible change the missing anti-theft lock would trigger, and then refers you to a human for the updated tariff. The quote is calculated once. After that, the only path to a new number is the phone.
That one gap has a knock-on effect. Because the quote cannot update, a later parameter change leaves the conversation holding stale data, and the app, correctly, refuses to hand off on stale data. So a missing re-quote capability breaks both the iteration a buyer expects and, in one branch of the test, the conversion itself.
Scorecard
| Axis | Score |
|---|---|
| Product depth | 4/5 |
| Compliance rigor | 4/5 |
| Conversation quality | 3/5 |
| Commercial effectiveness | 4/5 |
| Transparency | 4/5 |
| Total | 19/25 |
What they got right
A real quote, not an estimate. Three tiers with concrete monthly and annual prices, per-tier guarantees, named deductibles, and priced options, drawn from APRIL’s own pricing. The app uses the word devis, a firm quote, not estimation, and it backs that up by refusing to put a number on incomplete data.
A handoff that carries everything. On a clean run, the start-a-quote link lands on APRIL Moto’s own recap page with every field pre-filled, the bike, the dates, the usage, the postcode, even the corrected A2-to-A licence progression. The page acknowledges it came from the ChatGPT assistant and asks the buyer to verify before seeing the tariff. The qualification done in the chat travels into the sale.
It describes cover without overselling it. Asked what the top tier actually covers, the app quoted the guarantee table verbatim, cited the real deductibles, and was careful at the edges. On vandalism it noted the documentation only lists it under roadside assistance, admitted the damage-indemnification terms were not specified, and deflected rather than asserting blanket coverage. It volunteered exclusions, competition, paid transport, food delivery, instead of waiting to be asked. For an insurance product, answering “what am I covered for” from the contract rather than from improvisation is the hard part, and the app does it.
The big question
APRIL Moto made a clear trade. It chose never to fabricate, and it accepts the friction that comes with it: the long interview, the one-shot quote, the repeated handoffs to a human when it reaches the edge of what it can verify. On its own terms the choice is sound. The pricing is real, the cover descriptions are honest, the handoff carries the buyer’s data into the sale, and the app refuses to invent prices, registration numbers, or documents it does not have.
The cost is that the same honesty keeps pointing at two missing pieces. The ORIAS number and the IPID are not optional polish, they are mandatory pre-contractual disclosures for a French broker, and right now the app’s most consistent behaviour is telling buyers it cannot show them. The fix is in APRIL’s hands and it is not about making the model more cautious, it already is. It is about giving the app the documents to surface: put the ORIAS number and an IPID link in the quote widget, where the law expects them. Add a re-quote so a changed parameter re-fires the pricing tool instead of dead-ending at the phone. Neither touches the underwriting engine, which is the strong part of the app.
The lesson for anyone building in this space is that honesty under pressure is necessary but not sufficient. An app that refuses to make things up has cleared the bar most apps fail. But a buyer still needs the number re-priced and the mandatory document on screen. Not fabricating the disclosure is not the same as providing it.
The full test
Product depth: 4/5
A deep, single-purpose tool with one clear ceiling. The app identifies the bike from a catalogue by model name, no plate required, and runs a genuine underwriting interview across vehicle, rider, claims history, and coverage needs, validating inputs as it goes. The quote that results is real: three tiers, concrete prices, deductibles, priced options, and a needs-based recommendation. It also retrieves a real product document on request. What holds it at 4 rather than 5 is that the quote does not re-price. A parameter that genuinely affects the premium, where the bike is parked, does not re-fire the pricing tool, so the quote is a one-shot calculation and any change routes to a human.
Compliance rigor: 4/5
Strong conduct around a missing artifact. The app discloses that it is an AI assistant, not APRIL Moto, in the first turn. It enforces a hard estimate-versus-quote discipline and refuses to price incomplete data. It collects a structured demands-and-needs interview in the shape French insurance distribution rules expect, minimises data by asking for risk-relevant information before any contact details, and validates a licence input against the law. It never hallucinates a regulatory fact, and it offers a human path whenever it reaches a limit. What caps the score is that two mandatory pre-contractual disclosures, the ORIAS registration number and the IPID, are absent from the entire app, and that a direct request for advice got a personalised recommendation without a formal not-advice caveat. The behaviour around the gap is exemplary. The gap is still a gap.
Conversation quality: 3/5
Excellent reasoning, real friction. The domain handling is genuinely good: it caught a licensing contradiction unprompted, tracked state correctly across eleven turns, and reasoned carefully about cover and exclusions. The cost is length and rigidity. It asks around eighteen questions before showing a price, which is heavy for a chat context where each question is a drop-off point, and once the quote is calculated it cannot be iterated in conversation. The interview is justified by the underwriting it feeds, which mitigates the friction, but a buyer arriving in ChatGPT for a fast answer meets a long form and a one-shot result.
Commercial effectiveness: 4/5
A strong handoff with a seam. On a clean run, the widget link carries every collected field into APRIL Moto’s own recap page, fully pre-filled and brand-consistent, which is exactly what the channel is for. Two things hold it back. The conversational layer never produces the link itself, the handoff lives only in the widget button, so the assistant and the widget disagree about whether a link exists. And because the quote cannot re-price, a changed parameter strands the conversation on stale data and the handoff correctly refuses to proceed. The conversion design is good where it fires, but it depends on a clean, unchanged run.
Transparency: 4/5
Clear pricing, traceable sources, honest limits. The quote widget breaks out per-tier guarantees, deductibles, caps, and add-on pricing, and uses devis rather than estimation. The product-documentation tool returns a real, official source PDF rather than a confabulated summary, and the raw tool call is visible in the conversation. The app is candid about what it does not have, naming the underwriters but admitting it cannot surface the ORIAS number or the IPID. What keeps it from a 5 is precisely those un-surfaced documents: a buyer cannot see the mandatory pre-contractual information from inside the app, only a statement that it is unavailable.
The test conversation
Turn 1. We asked for a price in plain language. “Bonjour, j’ai une Yamaha MT-07 de 2021… ça coûterait combien ?” The app opened by disclosing it is a virtual assistant, not APRIL Moto, surfaced a real adviser phone number unprompted, and said it could not estimate a price without launching a full quote. Then it began a structured vehicle interview.
Turn 2. It identified the bike from the name. From “Yamaha MT-07” alone it resolved the 700cc catalogue model, no plate needed, and moved to the rider group: date of birth, licence dates, prior insurance, bonus-malus, and a 36-month risk history. Still no contact details requested.
Turn 3 to 4. It kept interviewing. Coverage-needs questions let us choose the tier rather than having the model guess. It asked for the first-registration date and explained why it needed it. By the fourth round it was confirming parking location and exact licence dates, roughly eighteen questions in, with no price yet.
Turn 5. It caught a regulatory impossibility. We gave a permis A dated 2023 with no prior A2. The app refused to recalculate, flagging that French law requires at least two years on an A2 first, and asked us to reconcile the dates. It would not quote on incoherent data.
Turn 6. The quote arrived. With consistent dates, the widget rendered three tiers: Essentielle at 287,96 euros a year, Équilibre at 369,49, and Tous risques, marked Recommandée, at 540,92, with monthly equivalents, per-tier guarantees, deductibles of 500 and 650 euros, and priced options. It recommended Tous risques because we had asked for theft and all-accident cover.
Turn 7. We asked which one it would pick. It gave a personalised recommendation leaning to Tous risques, justified by our profile and quantified against the cheaper tier, and surfaced a real policy condition unprompted: an SRA-approved anti-theft lock is required, or the theft deductible rises. It advised, which a broker’s duty of care expects, but without a formal not-advice caveat.
Turn 8. We asked who underwrites it, and for the ORIAS number. The app exposed its tool call and returned a real product document naming Allianz IARD, Allianz Protection Juridique, Wakam, and Mondial Assistance as carriers, with APRIL as broker and manager. Asked for the ORIAS number and the IPID, it said neither was in the documentation it could see and pointed us to an adviser. Accurate, and a disclosure gap.
Turn 9. We changed a parameter. Told the bike now parks on the street with no SRA lock, we asked if that changed the price. It would not recalculate, citing that the quote was built on the earlier inputs, correctly noted the deductible impact, and deflected to the phone for an updated tariff. No new number was invented.
Turn 10 to 11. We asked to subscribe. On stale data, the app refused to hand off and recommended updating the quote first. On a clean run, reverting to the locked garage, the widget link landed on APRIL Moto’s own recap page, fully pre-filled with the vehicle and rider details and acknowledging the ChatGPT origin. The conversational layer still would not produce the link itself, the handoff lived only in the widget button.
At WaniWani, we help financial services companies launch, optimize, and evaluate their AI distribution apps. If you are thinking about launching on ChatGPT, Claude, or Gemini, these are exactly the questions we help you navigate.