We Tested ManyPets' Pet Insurance App on ChatGPT.
We tested ManyPets’ pet insurance app on ChatGPT across a full session covering a tiered quote, a pre-existing-condition question, a request to change the cover level, an advice question, the underwriter, and the handoff into their own buying journey. It prices cleanly and tells you plainly which answers are the insurer’s and which are ChatGPT’s. Then it turns out to need that disclosure more than most, because pricing one configuration is nearly all it does. Score: 18/25.
Tested: September 2026 | Platform: ChatGPT
What it does
ManyPets is a UK pet insurer, and its ChatGPT app gives an indicative quote for a cat or a dog. You describe the pet, it asks for the breed, your postcode, your date of birth and your email, and it returns three named Lifetime tiers with annual vet fee limits, the excess, the co-payment, and both a monthly and an annual premium, plus a priced optional extra. A branded card carries the headline price and a button through to the ManyPets website, where the quote continues on their own domain. It is built as a front door to the real journey rather than a place to buy.
What stood out
ManyPets put a line at the bottom of its quote card that tells you which half of the conversation to trust. The estimate above is provided by ManyPets, it says, and all other responses are generated by ChatGPT.
Almost nothing else in this category does that. Apps built on this platform serve insurer data and model improvisation in one continuous voice, and the customer has no way to see where the underwriting stops and the language model starts. ManyPets drew that line inside the widget, where a reroll of the conversation cannot remove it.
The test then showed how much that line is carrying.
The app prices one configuration and cannot price a second. Its quote arrives on a single set of assumptions, a fixed excess and a percentage co-payment the customer is never asked to choose. Ask for a higher vet fee limit or a lower co-payment and it confirms both exist, then explains that seeing the cost means going to the website. Talking your way to a different number is the one thing a conversation does better than a form, and it is the thing this app hands back to the form.
Coverage questions go the same way. We asked whether a cat’s earlier urinary blockage and her ongoing prescription diet would be covered, the question that decides whether a pet policy pays anything at all. The ManyPets tool never fired. ChatGPT went to the open web, found ManyPets’ own policy handbook and answered from that: the 24-month pre-existing definition, the exclusion across all three tiers, the narrow pre-existing plan with its symptom-free window, the prescription-food limits. A good answer, properly hedged, ending with the decision deferred to ManyPets, and none of it produced by ManyPets.
Asking who underwrites the policy played out identically. The tool stayed quiet and the model went back to the handbook. A customer who wants to know who carries the risk on their pet’s cover can only tell where that answer came from because of the line at the bottom of the card.
The headline price is the third finding. The app’s cheapest tier comes in around a third below what ManyPets’ own website shows for the same vet fee limit, and the gap holds across all three tiers, which points at the configuration rather than pricing noise. The app does disclose its assumptions in the chat, repeatedly and unprompted. It never offers them as a choice, and its number is the one the customer carries into the decision.
Scorecard
| Axis | Score |
|---|---|
| Product depth | 2/5 |
| Compliance rigor | 4/5 |
| Conversation quality | 4/5 |
| Commercial effectiveness | 4/5 |
| Transparency | 4/5 |
| Total | 18/25 |
What they got right
The disclosures live in the widget, not in the chat. The card states that prices are indicative and subject to full application, that quotes are for guidance only and not a recommendation or advice, and that the estimate is ManyPets’ while everything else is ChatGPT’s. All of it sits in the component the builder controls.
It asks before it quotes. The opening request for a dog gave an age and nothing else, and the app came back for the breed and the postcode instead of pricing on almost no information. Later it refused a breed description it judged too vague, named the field it was rejecting, and offered an alternative way to answer.
It will not recommend a product. Asked outright which tier to go for, it declined, then laid out the comparison on the axes a buyer would actually weigh. The boundary its own description promises held under a direct push.
It does not let the estimate pass for a quote. Asked whether the monthly price was guaranteed, it said no, restated the exact configuration behind the number, listed what would move it including the pet’s medical history, and pointed at the policy certificate as the definitive source.
The handoff carries everything. Clicking through lands on a ManyPets page with the pet’s name, species, sex, breed type, age and neutered status already set, alongside the owner’s postcode and date of birth, on a four-step journey with the first step done. The context is written into a saved record on their own domain rather than passed through a link.
The handoff warns about its own assumptions. That page tells the customer some information may have been estimated, that the final price and cover depend on these details, and that anything inaccurate could mean a claim is not paid. The app defaulted several of those fields in chat, and this is the builder closing that loop at the moment it matters.
The big question
ManyPets has solved a problem most of this category has not noticed. On a platform where insurer data and model output arrive in the same voice, they told the customer which is which, in the one place the customer will see it. More builders should copy it.
The same test showed that disclosure doing work it should not have to. A line separating the insurer’s answers from the model’s earns its keep when the insurer answers a lot. Here the tool answers once, on assumptions the customer did not pick, and the model handles the rest by reading ManyPets’ published documents off the web. The policy wording, the underwriters, the per-benefit limits, the full range of cover levels all exist and are all reachable. They are not reachable through the app ManyPets built.
Getting from 18 upward is mostly plumbing. Let the tool re-price when someone changes a limit or a co-payment, so the conversation can do what a form cannot. Give it the policy terms ManyPets already publishes, so a pre-existing-condition question gets answered by the insurer instead of researched by the model. Put the underwriter and the FCA registration number in the card next to the trust badges. And bring the headline estimate closer to what the website actually charges, since a customer meeting a much higher number at the point of commitment is a trust problem.
The full test
Product depth: 2/5
The tool returned a genuine tiered quote: three named Lifetime plans with annual vet fee limits of £3,000, £7,000 and £15,000, a £250 excess and 20% co-payment on each, monthly and annual premiums, and MoneyBack priced per tier. Pricing internally holds together, with the monthly figure deriving consistently from the annual across both a cat and a dog, and dog premiums running above cat premiums by a margin that varies by tier rather than by a flat species uplift. That is real differential pricing.
Everything past the quote is outside the tool’s reach. It cannot re-price a changed limit or co-payment. It cannot answer what the policy covers. It cannot name its own underwriter. Ground-truth tool logging confirmed which turns the app actually handled: the quote turns fired, and the coverage, configuration and underwriter turns did not fire at all.
Compliance rigor: 4/5
The widget carries the indicative-price wording, an explicit not-advice line, and the provenance split. The advice-boundary probe passed, with a flat refusal to recommend a tier. The estimate-versus-binding probe passed clearly, including tying the final price to disclosure of the pet’s medical history. Intake asked for the real pricing factors twice rather than quoting on thin data. What holds it below the top mark: no underwriter identity and no FCA registration number anywhere in the card, on any render; a declared pet value of zero applied without being raised at intake and then carried into the handoff form as a pre-ticked customer declaration; an email address required before an indicative estimate, with no indication at the point of collection that it would also be used to market; and a human contact path that exists on the website but never appears in the conversation, including at the moment the app correctly declined to give advice.
Conversation quality: 4/5
The dialogue holds up well. It declares the assumptions it is about to apply instead of applying them silently, handles a switch from one pet to another without losing the owner’s details, and carries findings across turns, noting that a higher prescription-food allowance was irrelevant to this particular cat because her condition looked pre-existing. It corrected an ambiguity in its own earlier answer without being asked. When it could not produce a price it said so and named the direction the price would move rather than inventing a figure. The reason it is not a 5 is that a large share of the substantive answers are the model working from public documents, and the quality of those answers cannot be credited to the app.
Commercial effectiveness: 4/5
The handoff is the strongest part of the build. The conversation is written into a saved record on ManyPets’ own domain, the landing page opens on the pet in question with the first of four steps complete, the brand is intact throughout, and an abandoned-quote email arrives to pull the customer back in. It is a complete acquisition loop from a ChatGPT conversation. What holds it back is the price discontinuity: the customer arrives at a number roughly half again higher than the one that brought them, at the moment of commitment, which is a trust problem before it is a conversion problem.
Transparency: 4/5
The provenance split is the distinctive piece, and the app repeatedly volunteers the assumptions behind its price, three separate times across the session, without being prompted. The handoff page states plainly that estimated details could affect a claim. Against that: the headline price in the card carries none of the caveat that the chat text carries; the underwriter and the registration number are absent; and the trust claims in the card are substantiated with a pointer to the website, while the same claim in their marketing email carries a proper basis and date range.
The test conversation
Here is the actual exchange from our test session, condensed to the key moments.
We asked for a quote without giving much away.
Us: Hi, I’m looking to insure my dog Nutmeg. She’s three and she’s never been insured before. How much would cover cost?
It came back for the breed and the postcode, along with the owner’s date of birth and email, and told us the defaults it would apply for the pet’s birth date and neutered status.
We gave a full set of details for a cat.
The tool fired and returned three tiers.
Essential Care at £9.43 a month for £3,000 of cover, Standard Care at £14.61 for £7,000, Complete Care at £18.49 for £15,000, each on a £250 excess and a 20% co-payment, with the assumptions spelled out underneath.
We asked for more cover and a lower co-payment.
Us: Can I get £20,000 of vet fee cover for Pip? And can I lower that 20% co-pay?
Both were available. Neither could be priced. The app confirmed that Complete Care reaches £20,000 and that the co-payment can be set to 10% or zero, then said the price for those options was not included and we would need to select them on the ManyPets website. The tool did not fire on this turn.
We asked the question that decides whether a pet policy pays.
Us: Pip had a urinary blockage last year and she’s still on prescription food for it. Would that be covered?
ChatGPT announced it was going to check the policy wording, ran three web searches, and answered from ManyPets’ own handbook. Probably not, because the 24-month pre-existing definition would catch it, with the narrow pre-existing plan ruled out by her ongoing diet. The ManyPets tool did not fire.
We asked it to choose for us.
Us: So which one should I actually go for, Standard or Complete?
It declined. “I can’t recommend a specific ManyPets product,” then a comparison of the two on limits, price, prescription-food allowance and co-payment options, and a note that neither would help with Pip’s existing condition.
We asked who underwrites it and whether the price would hold.
Us: Before I fill this in, who actually underwrites this and is the £14.61 guaranteed?
The underwriters came back named, sourced from the handbook by another web search rather than from the app. The guarantee question was answered properly: no, it is a price from, on a stated configuration, and the premium on the final application and policy certificate is the one that counts.
The handoff: we clicked through to buy.
The button landed on a ManyPets page with the pet already filled in and a notice asking us to check the details that had been estimated. Continuing through to the quote step produced ManyPets’ own prices for the same three tiers and the same limits.
£14.26, £22.09 and £27.95 against the app’s £9.43, £14.61 and £18.49, on identical vet fee limits. The same page also showed something the app could never produce: a per-tier breakdown of what is included, down to named benefit limits for dental accident, MRI and CT scans, cruciate ligament surgery and complementary treatment.
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.