Insurify's AI Agent Gateway, Broken Down

Last updated: 4 November 2026

Insurify has launched its AI Agent Gateway. AI agents such as Muse, Instinct or Dots can now get real-time car insurance quotes from Insurify, and every quote keeps its coverage information while the shopper’s personal data stays protected.

The gateway gives Insurify back control over AI agents, and independent research on how agents use websites supports that choice. It does not fix the problem for the market, because every insurer, MGA and comparison site will now write its own rules, and agents seldom read the rules that sites already publish.

What changed

In September 2026 Insurify blocked Meta’s Muse agent from its website, which stopped personal assistants from filling in its quote forms with no control. The gateway opens the site to AI agents again, on Insurify’s terms:

AI agents must identify themselves and can’t pretend to be human.
They must get quotes through routes built for agents, an MCP server and WebMCP tools, and never through the forms built for people.
The shopper must consent to the data being shared before any quote runs. The agent can’t consent for them.
Every quote is shown with its full coverage details, never a partial view.

This gives Insurify back control over the growing number of AI agents now looking for insurance quotes. It is good progress, and the consent step in particular is something every site that quotes insurance should copy.

What Insurify’s AI Agent Gateway includes

Building block	What it is for
Direct connection for AI assistants (MCP server)	Lets an assistant like ChatGPT or Claude request a quote from Insurify directly, without opening the website
Public description of that connection (MCP server card)	Tells assistants what the connection does and how to use it, so they can find it on their own
Quote tools on the website (WebMCP)	Lets an assistant that browses the web for someone get a quote from Insurify's car insurance pages
Quote page for assistants	A simple form for assistants that cannot use either of the above
List of assistant skills	A menu that tells assistants what they can do on Insurify: get a rough estimate, or get a real quote
Instructions file for AI (llms.txt)	Tells assistants which route to use first and what rules to follow
Rules for bots (robots.txt)	Lets AI tools read Insurify's content and keeps them away from the quote forms
ChatGPT app	Lets people compare car insurance inside a ChatGPT conversation
AI Agent Gateway	The name Insurify gave to the routes above when it announced them together
Poppy file	Insurify’s entry for the Personal Agent Protocol (PAP, said “Poppy”), a proposed open standard from Sierra and Meta that tells personal AI agents how to find a company and what they may do

How Insurify’s consent step works

Before an assistant gets a real quote, the shopper has to approve it. The assistant sends the person a consent link, the person has one hour to agree, and the approval then lasts 24 hours (48 hours when the assistant works through the website). The assistant cannot agree on the person’s behalf.

Clicking Agree is a legal step. Insurify’s terms say it is the shopper’s electronic signature, and it allows Insurify to share their details with carriers and to pull credit and driving-record reports.

What happened when we tested it

We ran Insurify’s quote flow through its direct connection with a made-up test profile: a driver born in 1990, a 2020 Toyota Camry LE, full coverage and a fake Brooklyn address. After a person clicked Agree, the application went through and the results came back with zero carrier quotes and one sponsored GEICO ad.

That result is expected. Carriers will not rate a driver who does not exist, so our test proves that the connection works from start to finish, not what a real shopper would pay. A real personalised price needs a real driver.

What independent research on agents adds

Several studies from 2025 and 2026 have measured how AI agents behave on real websites. None of them looked at insurance quotes, so the figures describe agents and AI crawlers in general. Four of the findings bear directly on Insurify’s gateway.

Blocking AI traffic costs visibility. A study of large news publishers found that those that blocked AI crawlers in robots.txt lost about 7% of their weekly visits in the six weeks after the block, compared with publishers that had not blocked. In Web Bench, a benchmark of browser agents on more than 500 real websites, one of the most common failures was an agent leaving the site it started on for a different one. News is a different market from insurance, and the evidence still points one way. A shopper whose agent is turned away keeps the agent and finishes the task on another site. Insurify moved from a block to a gateway in five weeks, and the research supports that move.

Agents almost never read llms.txt. Ahrefs looked at 137,000 domains in May 2026. 97% of the llms.txt files on those sites received no requests at all, and assistants that browse for a person, such as ChatGPT-User and Claude-User, made 2.5% of the requests that did arrive. Insurify’s llms.txt carries most of its rules for agents: identify yourself, get consent first, don’t fill in the quote forms, don’t quote in bulk. An agent that never opens the file never sees those rules. The rules that hold are the ones built into the tools, like the consent step, which no agent can skip.

Forms are where agents fail most. Web Bench found that the best agents complete more than 70% of tasks that only read a page, while the best result on tasks that write, such as filling in forms or logging in, was 46.6%. ClawBench, which tests agents on everyday tasks on live websites, long forms included, found that the strongest model it tested completed 33.3% of them. A car insurance quote is a long form full of personal details. Sending agents to tools built for them, away from the quote forms, is a choice these benchmarks back directly.

A name an agent gives itself is easy to fake. In August 2025 Cloudflare reported that when Perplexity’s declared crawler was blocked, requests arrived with a generic Chrome identity from addresses outside Perplexity’s published ranges. Perplexity disputed the report. Insurify asks agents to name themselves, but its MCP server needs no key, so a name is all it gets. Signed requests already exist: Cloudflare’s Web Bot Auth lets an agent prove who it is with a cryptographic signature. Checking signatures of that kind is the step Insurify’s gateway is missing.

Our view after reading this research: Insurify was right to open a door, and agents now need a common way to find those doors and prove who they are, because a rulebook in a text file mostly goes unread.

Why one rulebook per insurer does not work

Insurify wrote its own standard. Every other insurer, MGA and comparison site will now write theirs, each with its own way to identify agents, collect consent, name its tools and format a quote. An AI agent shopping for car insurance will have to learn a different rulebook for each site it visits, and first it has to find that rulebook.

You can already see it. Insurify’s MCP server asks for a quote in its own words and its own format: three tools, a consent token and a polling token. The Zebra, another comparison site with a working connection, starts a session, asks the agent for its own identifier and returns only quotes from carriers you can buy online through The Zebra. An agent that wants to compare the two has to learn each one separately and then guess how their answers line up.

Insurers and comparison sites each pay to design and police their own gateway, and agent builders write a new integration for every site they support. Insurify’s case against scraping was that a quote without its context is just a number. The same risk comes back for shoppers when every site describes its quotes differently.

A private rulebook also has limits that one company cannot fix alone. Insurify asks agents to say who they are, but its MCP server needs no sign-in or API key, so nothing checks the answer. Nobody checks the shopper either: our made-up driver went through after one click on Agree. Rules written in a text file only work for agents that read them and choose to follow them. The consent step works because the tool enforces it, and identity needs the same treatment. Verified identity for agents and shoppers has to work across the whole market, and no single insurer can set that up for everyone.

One standard for the whole market

We believe there should be one common standard for how AI agents quote insurance, agreed by insurers and owned by regulators, so that agents can quote safely and under control whatever the insurer or the buyer.

Insurify’s four rules make a good first draft. A shared version would add what one company cannot provide on its own:

One fixed place where every insurer publishes its rules and tools, so an agent finds them without guessing. The Personal Agent Protocol, which Insurify joined in October, is one candidate.
Verified identity for agents, for example signed requests, so a scraper can’t call itself Claude and get the same treatment.
One consent format that a shopper gives once and that every insurer accepts.
A shared quote format, so coverage, exclusions and the carrier’s name travel with the price wherever it is shown.
Supervision by regulators, who already set the rules for how insurance is sold to people.

AI agents are a new distribution channel. Regulators already own the rules for insurance distribution, so the rules for agents belong with them too.

What this means for carriers

Carriers listed on Insurify are already quoted to AI agents under Insurify’s rules. Carriers that open their own routes to agents should take the same care with consent and coverage details, and they should push for a shared standard before each of them builds a private one. Our explainer on what AI distribution is covers where to start.

Until that standard exists, some steps cost little and help today:

Don’t block agents outright. Give them a route you control.
Don’t rely on llms.txt to carry your rules. Put them in your tool descriptions too, and enforce the ones that matter inside the tool.
Point agents to your routes from the pages they already open, such as the home page, /docs and /pricing.
Put the key facts in plain text near the top of each product page, so an agent that reads only part of a page still finds them.
Offer a route for agents that fetch pages as well as one for agents that browse.

FAQ

What is Insurify’s AI Agent Gateway?

Insurify’s AI Agent Gateway, announced on 28 October 2026, lets AI agents such as Muse, Instinct or Dots get real-time car insurance quotes through an MCP server and WebMCP tools. Agents must identify themselves and the shopper has to consent first. Every quote keeps its full coverage details and the carrier’s name.

Can ChatGPT or Claude get a car insurance quote from Insurify?

Yes. Insurify offers a direct connection (an MCP server) that assistants like ChatGPT and Claude can use to request real, personalised car insurance quotes. The shopper approves each quote session through a consent link before any quote runs. A made-up profile goes through the flow but gets no carrier prices, so a real price needs a real driver.

How does Insurify’s consent step work?

The assistant sends the shopper a consent link, and the shopper has one hour to agree. The approval then lasts 24 hours, or 48 hours when the assistant works through Insurify’s website. The assistant cannot give consent for the shopper.

Why did Insurify block Meta’s Muse assistant?

Insurify blocked Muse from scraping its website on 23 September 2026. It said scraped quotes lose coverage details and disclosures, and that mass quoting sets off paid record checks. It kept offering access through its approved routes and later named Muse in its gateway announcement.

What happens when an insurer blocks AI agents?

Research on large news publishers found that blocking AI crawlers was followed by about 7% fewer weekly visits. An agent that is turned away can also finish the task on another site. Insurify chose to give agents a controlled route instead of a block.

Do AI agents read llms.txt?

Rarely. In an Ahrefs study of 137,000 domains, 97% of llms.txt files received no requests at all in May 2026. Rules for agents belong in the tools agents call as well, where they can be enforced.

Can an AI assistant buy insurance on Insurify?

No. Assistants can get quotes, and a person completes the purchase with the carrier or a licensed agent. Insurify’s rules also tell assistants not to click on quote links for the shopper.

Which insurance does Insurify support for AI assistants?

Car insurance only, as of 28 October 2026. For home, renters and life insurance, assistants are given a link to the right page on Insurify’s website.

Is Insurify’s ChatGPT app live?

Yes. It launched on 6 February 2026 and added real-time quotes in 11 US states in August 2026. Insurify plans to reach more than 40 states by the end of 2026.

Why does the insurance market need one standard for AI agents?

Without one, every insurer, MGA and comparison site writes its own rules for AI agents, and agents have to find and follow a different rulebook on each site. One standard, agreed by insurers and owned by regulators, would let agents quote insurance safely and under control whatever the insurer or the buyer.