Anyone can type a number into a website. So the obvious question about a site built on what people say they pay is: how do you know it's true?
Here's how we handle it, in plain words.
Two kinds of rates
There are two ways to tell us what you pay.
- Share it. Type in your premium and a couple of details. It takes about 20 seconds.
- Verify it. Upload your policy declarations page, the summary page at the front of your policy or renewal that lists your coverages and your price.
Both count. They don't count the same.
4×
A verified rate counts in full. A shared rate counts a quarter as much. Shared rates help most where we don't have many verified ones yet, and they can never outweigh the verified records and public data in an area, no matter how many come in.1 That protects the numbers from honest mistakes and from anyone trying to tip them.
Those weights are a starting point. Once there's enough data, we'll compare shared rates with verified ones in the same places, measure how far apart they run, and adjust. Every change will be posted with our methodology.
What happens when you upload
1. A computer reads the page. We use Claude, an AI model made by Anthropic, through Anthropic's API. Anthropic may keep what it's sent for a limited time to monitor for abuse, and we're requesting zero data retention. Claude reads your declarations page and pulls out the fields we need: the company, the premium, the policy dates, the coverages, the deductibles. Each field comes back with the page it was found on and how confident the reading is.
2. You check it. We show you what we read. If anything looks off, or the reading wasn't confident, you confirm it or fix it.
3. Code checks the math. The AI reads; it never does the math. Plain code checks that:
- the line-item premiums add up to the total;
- the coverages make sense together;
- the policy period is a normal length and ended no more than 18 months ago;
- the ZIP code matches the one you gave us;
- the policy form matches the coverages on it.
4. It's verified, or it goes to a person. If every check passes and you changed the premium by 2% or less, the rate is verified. If something doesn't line up, a person reviews it.
We also watch for tricks. The same document can't be counted twice. Neither can the same policy for the same term. And a rate far outside what's normal for the county goes to a person for a look.
What we keep, and what we never take
We keep the numbers: premium, coverages, deductibles, the policy form, the ZIP code and similar facts about the policy.
We never store:
- names of people;
- driver's license numbers;
- dates of birth (we turn them into an age range, then drop them);
- full vehicle ID numbers;
- mortgage loan numbers;
- phone numbers.
Your policy number is kept only in a scrambled form that can't be turned back into the number. It's there so the same policy isn't counted twice.
Why we delete your document
24 hours
Once we've read the page, we don't need it. So we delete it within 24 hours. You can also click "Delete it now" the moment you're done.
We do this because a declarations page has more on it than we need: your name, your address, sometimes your lender. The safest file is one that doesn't exist.
There are three exceptions, and none happens without you being told first:
- You store it. Tick "Store my document" and we keep it in your vault so it's there at renewal time. It's off unless you tick it, and one click removes it.
- You let us keep it to improve how we read documents. A separate, optional choice, offered after we've read it. We keep the file up to 2 years to test and improve our reading, only our team can open it, and you can delete it any time.
- It's waiting for review. If a person needs to look at it, we keep it until they do, for no more than 7 days. If no one gets to it in time, it's deleted and your rate counts as shared instead.
After deletion, what's left is what we read from it (the policy's figures and details, never the personal things above), a fingerprint of the file so it can't be sent twice, and the time it was deleted, which you can see in your account.
How the numbers you see are built
When we show a typical rate for an area, it's computed the same way everywhere.
- We start from the best public figure for the area, like a state regulator's average or the Census survey. It counts as if it were 5 records.
- Verified and shared rates are blended in by their weight.
- When an area has 10 or more verified rates, we show the verified figure on its own, and say so: "Based on 23 verified policies."
- Below that, we show a blended estimate and label it as one, or just the public figure with its source and year.
We never publish a crowd figure for an area without enough data behind it, and never anything about a single home, household or policy.
Why it's worth it
A verified rate is the only current, real price data in most of the country. In 30 states, the state publishes nothing useful about what homeowners pay. Every declarations page makes the numbers sharper for everyone nearby.
And you get something back: a like-for-like comparison with similar homes near you, not just a rough rank.
We read it, keep the numbers and delete the document, unless you ask us to keep it. That's the deal, and it's the same for everyone.
Sources
-
What My Neighbors Pay methodology, November 14, 2026. Starting weights: verified 1.0, reviewed 1.0, shared 0.25. Shared rates can't add more total weight to an area than its verified records plus the public prior. The prior counts as 5 records. Full details: How we verify. ↩