We Tested Allianz's Insurance App on ChatGPT.
We tested Allianz’s pet insurance app on ChatGPT across 20 turns covering quoting, coverage terms, advice, insurer identity, and a full application to lead handoff. It prices a policy from Allianz’s own system and forwards a completed application to a human advisor, all inside the chat. It also quoted a monthly premium about four times higher than the carrier’s own application charged for the identical dog. Score: 16/25.
Tested: June 2026 | Platform: ChatGPT
What it does
Allianz’s ChatGPT app, which introduces itself as Allianz Concierge, is a German-market insurance assistant for the carrier’s own products. It answers product questions, walks you through an application form, calculates a personalized premium, and forwards the finished application to an Allianz advisor who then contacts you to conclude the contract. It is explicit that it does not sell a policy in the chat and does not handle claims. We tested it on pet health insurance for a dog: a three-year-old Labrador, neutered male, in Munich. Across the run it qualified the case, returned a branded premium, explained the coverage in detail, compared tariffs, and ran the full application all the way to a forwarded lead with a reference number.
What stood out
The app does the hard part. It holds a real conversation about pet insurance, prices a policy from Allianz’s own engine, retrieves specific contract terms when asked, and carries a completed application to a human advisor without ever leaving ChatGPT. Most insurance apps cannot do any one of those things well. This one does all four. And then it puts a confident, branded price on screen that is wrong by almost four times.
It priced the same dog at 215 euros and at 55 euros
In chat, we gave the app full details and it returned a premium in its own Allianz-branded Prämienberechnung widget: 135,12 euros a month for the COMFORT tariff, and later 214,69 euros a month for PREMIUM. The numbers were specific, non-round, and labelled Angebot, an offer, with no unverbindlich caveat anywhere on the card. They looked authoritative.
They were not. When we took the identical dog through Allianz’s own application form, with the full underwriting the form collects, breed, health, location, the same tariffs priced out at 36,52 euros a month for Komfort and 54,86 euros a month for Premium. The in-chat quote was roughly three to four times the carrier’s own application price. The gap is not explained by missing data either: the chat quote never asked for the dog’s breed, and a Labrador is a large breed that should push the price up, not down, yet the chat number was still about four times too high. Both figures come from Allianz systems, the conversational quote endpoint and the application, so this is not the model improvising. It is one Allianz tool disagreeing with another by a factor of four, on the surface the customer sees first. For a carrier whose pitch is transparency and trust, a price that is precisely and confidently wrong is the costly kind of error.
It abandons the conversation at the moment it matters
The app collects the quote conversationally, which is the right way to use a chat channel. You describe your dog in plain language and it prices the policy. Then, at the application step, it drops you into a roughly nine-step web form embedded in the widget and re-asks everything you already said: the breed, the location, the health history, the owner’s date of birth, the tariff, the deductible. Nothing carries over. A user who has already qualified in the conversation re-enters the whole case by hand.
Getting into that form is its own obstacle. The two buttons that should move you forward, Weiter and Anfrage starten, both reset to a top-of-funnel product picker and discard the qualified quote. A first attempt to reach the form returned an error. The form itself is clean, stateful, and works, but it is long enough that the path from a finished quote to a submitted application tests a real user’s patience. The one advantage a ChatGPT app has over the carrier’s website, not making you fill in a form, is exactly what gets discarded at the conversion step.
To its credit, the funnel does complete. After the full form, the app forwarded a qualified lead to a human Allianz advisor, issued a reference number, and promised a callback, all inside the chat with no redirect to allianz.de. The conversion works. It just makes a qualified buyer do the website’s job over again.
It recommends a tariff while positioning itself as information only
Allianz describes the app as informational, with binding advice reserved for its official channels. In practice, when we asked which tariff to pick, it gave a first-person recommendation: COMFORT “erscheint als eine vernünftige Wahl” for a healthy young dog, and in a later turn “würde ich persönlich COMFORT ansehen.” That is advice, and a broker’s duty of care can support giving it. What was missing is the guardrail around it: no note that this is not regulated advice, and no offer to route the question to a licensed advisor at the moment the recommendation is made.
That sits alongside a broader gap in regulatory furniture. No quote widget shows the insurer’s legal entity, a §34d registration, a BaFin reference, or an unverbindlich label. Asked directly who underwrites the policy and for a §34d number, the app correctly declined to invent one and said it is not itself a licensed intermediary, which is the honest answer. But it could not name the underwriter either, and it promised the legally required information would appear during the application. It did not. The same “I cannot confirm this” message reappeared mid-application. The insurer entities, Allianz Kunde und Markt GmbH, Allianz Beratungs- und Vertriebs-AG, and Allianz Versicherungs-AG, surface only in the final consent fine print, and the pre-contractual document set, the Erstinformation and the product information document, never appears at all.
Scorecard
| Axis | Score |
|---|---|
| Product depth | 4/5 |
| Compliance rigor | 3/5 |
| Conversation quality | 4/5 |
| Commercial effectiveness | 3/5 |
| Transparency | 2/5 |
| Total | 16/25 |
What they got right
A real quote, computed from a real engine. Give it the full case and it returns a monthly and annual premium with a complete basis breakdown: tariff, deductible, start date, payment frequency. It re-fires for a second tariff and builds a side-by-side comparison. The pricing tool genuinely calculates rather than estimating, which the application form confirms by pricing all four tiers for the same dog under full underwriting.
Coverage answered from the contract, not from improvisation. Asked what COMFORT covers, the app returned specific terms: operations from illness or accident up to 10.000 euros a year, reimbursement to four times the GOT rate in emergencies, dental surgery to 2.500 euros, a three-month waiting period for illness and none for accidents, and the rule that pre-existing conditions are excluded while a new illness arising later is covered. For an insurance product, answering “what am I covered for” from the policy terms is the hard part, and it does it.
Honesty under pressure. It qualified the case before pricing rather than guessing. It refused to fabricate a §34d or BaFin number when asked. It correctly disclaimed being a licensed intermediary, and it consistently deferred to the binding Versicherungsbedingungen. The final consent screen is reasonable too: an explicit data-processing consent, an opt-out address, an age confirmation, and a privacy link.
A handoff that reaches a human. The conversion produces what the channel is for, a qualified lead delivered to a human Allianz advisor with a tracking number and a promised callback, entirely in-chat. Branding stays Allianz throughout, and no competitor is ever introduced.
The big question
Allianz built an app that can quote, explain, compare, and convert, and most of that works. The conversational quote is the exception, and it is a serious one, because it is the first number a customer sees and it was about four times the carrier’s own application price for the same dog. The fix is not subtle: the in-chat quote endpoint and the application clearly run on different pricing logic, and the application is the one that matches the policy a buyer would actually get. Reconciling the two, and labelling the chat figure unverbindlich until they agree, is squarely in Allianz’s hands.
The second issue is friction by design. The app earns its place in ChatGPT by qualifying the buyer in conversation, then throws that away at the application step and makes them re-enter everything in a long form. Pre-filling the application from the conversation would turn a tedious re-entry into a confirmation, and would make the completed handoff, which already works, feel like the natural end of the chat rather than a second start.
The lesson for anyone building in this space is that capability is not the bottleneck. This app can do the things a quoting and conversion tool needs to do. What it has not done is make its surfaces agree with each other and with the law: one price, carried into a pre-filled application, with the insurer’s identity and the mandatory pre-contractual documents on screen before the customer commits. The engine is the strong part. The presentation is what needs the work.
The full test
Product depth: 4/5
A broad, genuine tool. The app quotes from a real engine, re-quotes for a second tariff, builds a tariff comparison, and retrieves specific policy terms (the 10.000 euro operation cap, four times GOT in emergencies, 2.500 euro dental, the waiting periods, the pre-existing rule). The application form does full four-tier underwriting. What holds it at 4 is that the central conversational quote returned numbers about four times the carrier’s own application price for the same dog, so the surface the customer sees first is unreliable even though the engine can clearly price correctly. One response branch also improvised a PREMIUM price instead of re-firing the tool.
Compliance rigor: 3/5
Strong conduct around weak furniture. The app qualifies before it quotes, refuses to fabricate a §34d or BaFin number, correctly states it is not a licensed intermediary, and defers to the binding conditions. Against that: it gives a first-person tariff recommendation with no not-advice caveat and no advisor deflection at the advice moment, despite positioning itself as information only. No quote widget shows the insurer entity, a §34d registration, a BaFin reference, an AI disclosure, or an unverbindlich label. The insurer entities appear only in the final consent fine print, and the Erstinformation and product information document never appear, even at lead submission. The honesty and a decent consent screen hold it at 3 against the missing pre-contractual furniture.
Conversation quality: 4/5
Coherent and tailored throughout. It tracked state across 20 turns, gave a useful worked example of likely costs (a cruciate ligament tear, a gastric torsion), and handled iteration cleanly in the branch where the tool fired, carrying unchanged parameters and computing the delta. Held back by a recurring advice-tone drift and one branch that improvised a number instead of calling the tool. We also noted ChatGPT’s own A/B answer picker surfacing two different drafts for the same prompt, which is a platform behavior rather than something Allianz controls, so it does not move this score.
Commercial effectiveness: 3/5
The handoff completes, which is the core outcome: a qualified lead forwarded to a human advisor with a reference number and a promised callback, all in-chat, with Allianz branding intact. The path there is the problem. Both forward buttons first reset to the product picker, the form re-asks everything from the conversation with no pre-fill, a first attempt threw a transient error, and the wizard is long. It converts, but it squanders the conversational qualification and makes a ready buyer start over.
Transparency: 2/5
This is where the headline failure lands. The app presented a confident, branded Angebot with no unverbindlich caveat, and it was about four times the carrier’s own application price for the same dog. A customer has no way to tell the in-chat number is unreliable, because it looks authoritative: non-round, branded, structured. Compounding it, one widget leaked a raw internal product key and machine-formatted decimals while another rendered cleanly, one branch showed an improvised price, no methodology is shown, and the insurer identity and pre-contractual documents are absent from the in-chat surfaces. Partial credit for the self-sourcing language and the honest “I cannot confirm” on regulatory facts.
The test conversation
Turn 1. We asked what the app can do. It returned a capability card describing product information, application forms, premium calculation, and lead forwarding to an advisor, with a clear list of what it does not do: it does not conclude contracts and does not handle claims. Good scope honesty, though the card showed only the Allianz name and logo, with no legal entity or regulatory reference.
Turn 2. We asked for a premium. A three-year-old neutered Labrador in Munich, pet health cover with operation protection. The app did not guess. It asked for the owner’s and dog’s dates of birth, a start date, payment frequency, the tariff tier (BASIC, SMART, COMFORT, PREMIUM), and a deductible option before pricing. It did, notably, ask for the owner’s birthdate to price a pet policy, which we flagged.
Turn 3. The quote arrived. With full data, the Prämienberechnung widget returned 135,12 euros a month (1.621,49 a year) for COMFORT, labelled Angebot, with an Anfrage starten button and no unverbindlich caveat. The widget header leaked the raw internal key animal_health and a machine-formatted price.
Turn 4. We asked what COMFORT covers. It answered from the contract: operations to 10.000 euros a year, four times the GOT rate in emergencies, dental surgery to 2.500 euros, a three-month illness waiting period and none for accidents, and the pre-existing versus new-illness rule. It closed by deferring to the binding conditions.
Turn 5 to 6. We asked which tariff to pick, then to re-price PREMIUM. It gave a first-person recommendation toward COMFORT with no not-advice caveat. On the re-quote, the tool fired and returned a clean PREMIUM widget at 214,69 euros a month, with a side-by-side comparison against COMFORT and the PREMIUM inclusions listed.
Turn 7. We asked who underwrites it, and for a §34d number. It correctly declined to invent a registration, stated it is not a licensed intermediary, and deferred to the Impressum and the application. It could not name the underwriter and promised the required information would appear during the application.
Turn 8 to 10. We tried to start the application. Both Weiter and Anfrage starten reset to a top-of-funnel product picker, discarding the qualified quote. Selecting the product reached a real application form, but as a cold restart with nothing pre-filled, and the assistant reported it could not complete the request.
Turn 11 to 16. We drove the form by hand. It is a stateful, roughly nine-step wizard that re-asks breed, location, health, the owner’s birthdate, and the tariff, none of it carried from the chat. At the tariff step it priced all four tiers for the same dog under full underwriting: Basis 10,13, Smart 21,29, Komfort 36,52, and Premium 54,86 euros a month. The chat had quoted COMFORT at 135,12 and PREMIUM at 214,69. The form’s own prices were roughly a quarter of the in-chat figures.
Turn 17 to 20. We completed and submitted the application. Personal details and address came late in the form, which validated required fields. The final consent screen confirmed the Premium price at 54,86 euros a month, named the Allianz entities for the first time, and offered an explicit consent, an opt-out, and a privacy link. On submission, the app forwarded the lead to a human advisor, issued a reference number, and promised a callback, all in-chat.
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.