New York Insurance › Data

What New York ask for in auto insurance: the demand dataset

Every request a person or an AI assistant makes to New York Insurance becomes one record, the moment it is made: what they drive, where (to the ZIP3), what coverage they want, what the state's published rates showed them, which door they came through and which AI asked, and what happened next: whether they asked to be contacted, proved their phone, how many agencies received them, whether they withdrew. No name, phone, email, date of birth or five-digit ZIP, ever. It is the shopping-intent signal carriers, marketplaces and rate consultants pay enterprise prices for, observed first-hand and in real time, including the part nobody else has: what AI agents ask for on behalf of their humans.

What is in it, right now

19 records since 2026-09-02, 19 in the last 30 days · 0 asked to be contacted · 0 proved a phone · 6 withdrew · 84% asked by an AI agent.

By state: TX 11 · NV 5 · CA 2 · NY 1
By door: mcp 18 · form 1
By AI vendor: anthropic 15 · unknown 3 · openai 1
By basis: unknown 19

The free sample: the last 19 records

JSON: https://newyorkautoquotes.com/data/sample.json · the schema and rollups: https://newyorkautoquotes.com/data.json

What it costs

$0.02 per record, minimum $5 per pull; the rollups, the schema and the sample are free. Pay with a prepaid buyer key (register, then buy credits by card on your status page) or at the door with x402, MPP or AP2 (open now: stripe-checkout).

curl -s "https://newyorkautoquotes.com/v1/dataset?state=NY&limit=1000" -H "X-Buyer-Key: k_…"
# MCP: market_data { "state": "NY", "limit": 1000, "buyer_key": "k_…" }   or payment / credential / mandate

Who buys it

• Carriers and MGAs pricing a market they cannot see from their own book
• Insurtechs and rate consultants tracking quote-to-bind gaps by ZIP3
• Lead desks and ad-tech sizing a state before they spend
• Investors and analysts watching shopping intent and the AI-agent share of it

The schema

id

one-way hash of the request; joins outcomes, never a person

t

the hour the request was made (UTC)

state

two-letter state

entity

the entity asked

domain

the domain asked

door

mcp | form | consent-link | voice | agency-page

requester

agent | ai-crawler | sdk | browser | probe | impostor

vendor

openai | anthropic | google | perplexity | …

confidence

cryptographic | ip-verified | declared | user-agent

zip3

first three digits of the garaging ZIP

age_band

driver age in five-year bands (16-20 … 80+)

vehicle_year

model year

make_model

make and model, lowercased

coverage

state minimum | standard | full

mileage_band

annual miles: <5k | 5-10k | 10-15k | 15-20k | 20k+

violations

in three years: 0 | 1 | 2+

prior_continuous

currently insured

years_licensed_band

<3 | 3-9 | 10+

licensed_here

whether the entity showed prices

quoted

insurers priced

price_low

lowest monthly price shown (USD)

price_median

median monthly price shown

price_high

highest monthly price shown

basis

filed sample rate | state-published sample rate | indicative

consented

the person asked to be contacted

scope

contact_consumer | sell_identity

verified

phone proven by one-time code

delivered

agencies that received the request

withdrawn

the person withdrew

Never in a record: full_name, phone_number, email_address, date_of_birth, garaging_zip, pool_id, street_address, lead_id, household_drivers, ip, ua.

Is there personal information in it?

No. Age is a five-year band, the ZIP is three digits, the vehicle is a make and model, and the timestamp is the hour. The record id is a one-way hash. People who asked to be contacted are sold as leads only under their own consent, on a different rail, never through this dataset.

How fresh is it?

A record exists within a second of the request. Outcomes (consent, verification, delivery, withdrawal) fold in as they happen. The free sample is the last twenty records.

What does it cost?

Per record, with a minimum per pull; the catalog states both. Pay with a prepaid buyer key, or at the door with x402, MPP or AP2. The rollups, the schema and the sample are free.

What is unique about it?

It is observed at the moment of shopping, not modelled afterwards, and it records which AI assistant asked. Nobody publishes the AI-agent share of insurance demand. It also carries what the state's published rates showed at that moment, so price and intent sit in the same row.