We Tested Neptune Flood's Insurance App on ChatGPT.

We tested Neptune Flood’s real-time quoting app on ChatGPT across 4 turns covering quoting, re-quoting, deductible scenarios, and NFIP comparison. The widget auto-populates property details from a single address and returns a personalized quote from Neptune’s Triton engine. Score: 22/25.

Tested: March 2026 | Platform: ChatGPT

Neptune Flood has a live app on ChatGPT that generates personalized flood insurance quotes in real time. One address in, full quote out: property details auto-populated, annual cost with tax and fee breakdown, seven optional coverages expandable, and the most comprehensive widget disclaimer we have seen in our audit series. It scored 22 out of 25.

What it does

Neptune Flood is a private flood insurance MGU. Its ChatGPT app connects to Neptune’s Triton underwriting engine to generate real-time, personalized flood insurance quotes. The user provides an address. The tool auto-populates property details from public records, calculates a quote with coverage breakdown and tax/fee split, and renders the result in a branded widget with a “Get My Flood Quote” CTA that hands off to Neptune’s website. The widget supports parameter changes (occupancy type, coverage levels) through re-quoting, and exposes seven optional coverages for customization. Contact information (email and phone) is embedded in the tool data for ChatGPT to surface at conversion moments.

What stood out

Neptune’s app is a study in how much a single-provider tool can accomplish inside a conversational interface.

The widget does the heavy lifting. From one address input, it auto-detects seven property characteristics, computes a personalized quote with full cost transparency, and presents it in a branded panel with clear labeling. Base premium versus taxes and fees, building versus contents breakdown, deductible, flood zone, property details: everything is visible and verifiable. This is not a price lookup. It is a product configurator with seven optional coverages that the user can expand and explore.

The re-quoting works as expected. When we changed the occupancy type, the tool re-fired and rendered a new widget with the updated field. ChatGPT explained why the price did or did not change rather than pretending something had shifted. That kind of explanation (pricing driven by location and construction rather than occupancy for certain property types) was accurate, informative, and non-directive.

The compliance design is thoughtful. The widget disclaimer covers the essential bases: not an agent or broker, not a binder, subject to underwriting, 60-day validity. But the standout line is the one addressing platform risk directly: “Neptune Flood is not responsible for any reliance placed on information interpreted, summarized, or communicated by third-party AI systems.” Neptune is acknowledging a risk that most builders either do not see or choose to ignore: the provider controls the tool output, but the platform controls the narrative around it.

ChatGPT’s behavior in this conversation was unusually balanced. When asked to compare Neptune with the NFIP, it gave pros for both sides without picking a winner. When asked for a recommendation, it self-disclaimed (“Because I’m not providing licensed insurance advice”). When it used its own estimate for NFIP pricing, it labeled it as a rough estimate, not an official quote. This level of restraint is not something the builder can guarantee (ChatGPT’s behavior varies across sessions), but it is worth noting because it contrasts sharply with what we have seen in other insurance app audits.

The conversion path has the right structure. A branded CTA on the widget, contact email and phone surfaced from tool data, and a multi-step handoff flow with several steps pre-filled: Neptune provides multiple paths from conversation to purchase.

Scorecard

What they got right

The widget is a rich single-carrier integration. One address input produces a full quote with seven auto-detected property details, coverage breakdown, tax/fee split, and seven optional coverages. The user can verify every data point. This is not a price display; it is a product configurator inside a conversational interface.

The disclaimer explicitly addresses the platform risk. “Neptune Flood is not responsible for any reliance placed on information interpreted, summarized, or communicated by third-party AI systems.” Neptune is not just disclaiming their own output; they are disclaiming what the AI platform does with it.

ChatGPT’s behavior during this test was notably balanced. Self-disclaiming on a recommendation question, labeling its own estimates, presenting a balanced NFIP comparison, and framing deductible scenarios as options rather than advice. We cannot attribute this to Neptune’s tool design (ChatGPT’s behavior varies), but the conversation produced the kind of measured, non-directive exchange that regulators would want to see from an AI-assisted insurance interaction.

The big question

Neptune scored 22 out of 25. The points it lost tell a clear story.

Compliance dropped one point because ChatGPT’s balanced behavior is not guaranteed. The self-disclaimer, the balanced comparison, the non-directive framing: all of this happened during our test, but none of it is enforced by the tool. Neptune’s widget disclaimers are comprehensive and consistent. ChatGPT’s behavior is session-dependent and outside the builder’s control. The gap between what the widget disclaims and what the conversation produces is a structural risk that Neptune has acknowledged but cannot fully mitigate from the builder side.

Commercial effectiveness dropped one point because the handoff failed during our test. The multi-step flow with pre-filled fields is well-designed (confirmed in prior testing), but a server error broke the digital conversion path. For a distribution channel that depends on a seamless quote-to-bind experience, reliability is not optional. The contact information fallback (email and phone surfaced by ChatGPT) partially compensates, but a broken handoff undermines the strongest part of the funnel.

The broader lesson is about where the value sits. Neptune built one of the most complete AI distribution tools in insurance: real underwriting engine, comprehensive disclaimers, full cost transparency, multiple conversion paths. But the tool operates inside a platform the builder does not control. The disclaimer language (“not responsible for information interpreted by third-party AI systems”) is honest about this boundary. The question for every carrier considering this channel is whether that boundary is acceptable, and what infrastructure would be needed to close the gap between what the tool delivers and what the platform adds around it.

The full test

Product depth: 5/5

Neptune’s Triton engine delivers a real, personalized quote from a single address input. It auto-detects year built (1956), foundation (slab), occupancy (single family), residency (primary home), FEMA flood zone (X), construction (wood), and prior losses (none). Coverage is set at $250,000 building, $100,000 contents, and $5,000 deductible, with seven optional coverages expandable in the widget.

Re-quoting works on parameter changes. Switching from primary home to rental owner triggered a new widget render with the updated field. The price ($585/yr) did not change for this property, and ChatGPT explained the pricing logic rather than leaving the user to wonder.

The price breakdown is complete: $422 base premium plus $163 taxes and fees, totaling $585/yr. This is not an average or an estimate range. It is a personalized quote from a live underwriting engine, computed for a specific property at a specific address.

Compliance rigor: 4/5

The widget disclaimer covers six distinct points: ChatGPT is not an agent, broker, or representative of Neptune Flood; the quote does not constitute a binder or offer; coverage is subject to underwriting review, eligibility, terms, conditions, exclusions, and state regulations; users should consult a Neptune agent or insurance professional; the quote is valid for 60 days; and Neptune is not responsible for reliance on information interpreted by third-party AI systems.

That last line explicitly addresses the invisible boundary problem: when ChatGPT adds context, comparisons, or advice around the widget output, the user may attribute that content to Neptune. Neptune’s disclaimer says, in effect, “what the AI says around our data is not our responsibility.”

ChatGPT’s in-conversation compliance was strong during our test. The self-disclaimer (“Because I’m not providing licensed insurance advice”) on the NFIP comparison question is the only instance across all our audits where ChatGPT voluntarily disclaimed during a recommendation request. The comparison was balanced, with pros for each option and no directive. The deductible explanation was framed as scenarios to consider, not advice to follow.

The one gap: ChatGPT’s behavior is session-dependent. We cannot attribute the balanced behavior to Neptune’s tool design, because ChatGPT’s compliance varies across sessions and accounts. What we can say is that the widget disclaimers are consistent, comprehensive, and builder-controlled.

Conversation quality: 4/5

When the tool fires, the conversation is grounded in real data. All property details, coverage parameters, and pricing come from Neptune’s engine. ChatGPT’s additions on tool turns were accurate and useful: the base versus fee split, monthly equivalent, flood zone implications, and area risk context added value without contradicting the tool output.

On the non-tool turns, ChatGPT performed well. The deductible explanation with three dollar scenarios ($3k, $20k, $150k claims) was practical and easy to follow. The landlord-specific guidance (contents coverage, tenant insurance, lender requirements) showed domain depth. The NFIP comparison was balanced, with ChatGPT clearly marking its own estimate versus tool-sourced data.

The one limitation is structural: the tool does not fire on follow-up questions. The deductible scenarios, the NFIP comparison, and the landlord advice all came from ChatGPT’s general knowledge, not from Neptune’s tool. The content was accurate during our test, but it is not verifiable against Neptune’s data. When the tool fires, every claim is traceable. When it does not, the user is relying on ChatGPT’s training data.

Commercial effectiveness: 4/5

Neptune provides multiple conversion paths. The widget includes a “Get My Flood Quote” CTA. The tool data contains contact information (quote@neptuneflood.com, (727) 217-5343) that ChatGPT surfaced at the purchase question. The 6-step handoff flow pre-fills 3 steps from the conversation (address, product selection, building information), reducing the user’s work to profile details and final coverage confirmation.

The Neptune brand is prominent on every widget render (dark blue header with Neptune logo) and carries through to the handoff. ChatGPT did not introduce competitors or suggest alternative carriers at any point during the quoting flow. When the user asked about the NFIP, ChatGPT treated it as a comparison exercise, not a redirect away from Neptune.

The server error on our test’s handoff is a practical concern. When the handoff fails, the user sees a “Start new quote” button and loses the conversation context. Neptune’s contact info (surfaced by ChatGPT from tool data) serves as a fallback, but the digital path is broken. Based on our prior test, the working handoff is smooth (roughly 3 minutes to bindable quote), but reliability matters for a live distribution channel.

Transparency: 5/5

When the widget fires, every component is visible and labeled. Base premium, taxes and fees, total annual cost, building coverage, contents coverage, deductible, FEMA flood zone, and all seven auto-detected property details. “ESTIMATED ANNUAL COST (INC. TAX AND FEES)” is clear, honest labeling.

ChatGPT’s transparency was also strong. It distinguished between tool-sourced data and its own estimates (“This is only a rough estimate, not an official NFIP quote”). It clearly marked which information came from Neptune’s engine and which came from general knowledge. When the price did not change after the occupancy switch, ChatGPT explained why rather than obscuring the result.

The widget disclaimer addresses the platform transparency problem directly. By stating that Neptune is not responsible for information “interpreted, summarized, or communicated by third-party AI systems,” the disclaimer signals to the user that not everything in the conversation comes from Neptune. This is not a full solution to the source attribution problem, but it is the most explicit acknowledgment of it that any builder has implemented.

The test conversation

Here is the actual exchange from our test session, condensed to the key turns.

Turn 1: We asked for flood insurance.

Us: I need flood insurance for my home in Clearwater, Florida.

No tool fired. ChatGPT asked for the full property address and ZIP code. A straightforward intake step.

Turn 2: We provided the address.

Us: 2228 NE Coachman Rd, Clearwater, FL 33765

The Neptune widget fired immediately. A dark blue, Neptune-branded panel appeared with the header “QUOTE OVERVIEW” and returned a full personalized quote from a single address input.

The widget auto-populated seven property details from public records: year built (1956), foundation (slab), occupancy (single family), residency (primary home), FEMA flood zone (X), construction (wood), and prior losses (none). Coverage was set at $250,000 building, $100,000 contents, and a $5,000 deductible, with 0 of 7 optional coverages selected (expandable in the widget). The price: $585/yr ($422 base + $163 taxes and fees), labeled “ESTIMATED ANNUAL COST (INC. TAX AND FEES).”

Below the widget, a disclaimer block included: “ChatGPT is not an insurance agent, broker, or representative of Neptune Flood.” It continued with non-binding language, underwriting review caveats, 60-day validity, and the line that sets Neptune apart from every other app we have tested: “Neptune Flood is not responsible for any reliance placed on information interpreted, summarized, or communicated by third-party AI systems.”

ChatGPT added useful context around the widget: base premium versus taxes/fees split, monthly equivalent (~$49/month), flood zone X implications (low-to-moderate risk, not mandatory for mortgages), elevation (~67 ft), and area risk data (~$5.7B flood damage in Tampa Bay over 5 years, ~46k claims). It offered to explore lower deductibles, coverage scenarios, and an NFIP comparison.

Turn 3: We changed occupancy and asked about the deductible.

Us: What if I switch this to a rental property instead of primary home? And what does the $5,000 deductible mean in practice?

The tool re-fired. A new widget rendered with the residency field changed to “Rental Owner.” All other property details remained the same. Price: still $585/yr.

ChatGPT explained why the price did not change: for this property type in Zone X, pricing is driven more by location, flood zone, elevation, and construction than by occupancy. It noted where differences would appear (higher-risk zones, multi-unit properties, removing contents coverage). Then it walked through three claim scenarios with the $5,000 deductible: a $3,000 claim (fully absorbed by the deductible), a $20,000 claim ($15,000 paid by Neptune), and a $150,000 claim ($145,000 paid). It offered to simulate different deductible levels ($2,500, $10,000) and provided landlord-specific guidance on dropping contents coverage, tenant insurance interaction, and lender requirements.

The deductible explanation with concrete dollar scenarios was genuinely useful. The landlord follow-ups showed domain depth. All AI-improvised, but accurate and well-framed as options to explore rather than recommendations to follow.

Turn 4: We asked about NFIP and how to buy.

Us: Should I go with Neptune or the NFIP? And how do I actually buy this?

No tool fired. ChatGPT answered from general knowledge combined with data from the previous tool responses.

It presented a balanced comparison: Neptune ~$585 versus NFIP ~$600 (with the NFIP estimate clearly marked as ChatGPT’s own rough estimate, not from a tool). The framing was notably even-handed: Neptune (private, digital, flexible coverages) versus NFIP (government-backed, familiar to lenders). It provided a six-factor comparison checklist: premium, deductible, limits, waiting period, claims handling, and optional coverages.

Then came the compliance highlight of the entire test. ChatGPT stated: “Because I’m not providing licensed insurance advice, the safest neutral read is: neither option is obviously cheaper.” This is the only app we have tested where ChatGPT explicitly self-disclaimed during a recommendation question.

For the purchase path, ChatGPT surfaced data from the tool response that was not visible in the widget: Neptune’s contact email (quote@neptuneflood.com), phone number ((727) 217-5343), and a note that Florida does not require an agent for flood insurance (drawn from the tool’s IsDiligentEffortState field). It described a clear purchase flow and recommended getting an official NFIP quote for a side-by-side comparison.

The handoff: We clicked “Get My Flood Quote.”

The CTA opened Neptune’s website showing a 6-step flow: Address, Product Selection, Building Information, Profile, Coverages, Your Quote. On this test, a server error prevented the flow from loading (“Unable to load your quote. Unable to connect to the server.”).

However, our original test of Neptune’s app (March 2026, before the retest) confirmed that the handoff lands at step 4 of 6, with address, product selection, and building information pre-filled from the ChatGPT conversation. The remaining steps ask for profile details and final coverage confirmation. Estimated time to bindable quote: roughly 3 minutes.

At WaniWani, we help financial services companies launch, optimize, and evaluate their AI distribution apps. If you are thinking about shipping on ChatGPT, Claude, or Gemini, these are exactly the questions we help you navigate.