What we do - and don't do - with public data

People-search products have a deserved credibility problem. The category is full of services that imply access to things they don't have, produce confident profiles from thin evidence, and are vague about where any of it came from.

So rather than a marketing page, here is a plain description of what DeepSearch does, what it deliberately doesn't, and what we won't support it being used for.

Where the results come from

DeepSearch can use several source categories, depending on the lookup and product surface:

• public web pages
• public social media profiles
• public records and directories
• news and published articles
• licensed telecom carrier and numbering services that identify a number's carrier, line type, and assignment region
• public breach-notification indexes that name incidents associated with an email address
• on research surfaces that explicitly disclose it, licensed people-data services that return attributed identity-graph associations

Public-web claims should link to a page a reader can inspect. Licensed numbering metadata, people-data associations, and public breach-notification records are labeled by source category or provider. Paying for a dataset does not make its identity associations verified facts, and none of these categories provides private-account access or leaked credential contents.

What we don't touch

This part matters more than the list above, because it's where the category tends to blur.

No leaked credential contents. A public breach-notification index may tell us the name of an incident associated with an email address. We do not expose or search stolen passwords, hashes, private fields, or the contents of a breached record.

No private accounts. We don't log into anything, don't bypass privacy settings, and don't access content behind a follow request. If a profile is private, it stays private.

No call or message contents. We don't have them and can't get them. Neither can anyone selling them to you legally.

No private carrier subscriber records. Licensed numbering data can identify a carrier, line type, and assignment region. It does not identify the account holder, reveal communications, or show the device's location.

No live location. We can't tell you where someone is. Nothing public supports that, and services that claim it are selling a fiction.

If you've seen a competitor advertise any of the above, that's worth treating as information about the competitor.

Every claim is linked to its source

This is the part we care most about, and it's a design constraint rather than a feature.

Every public-web fact in a profile should carry a link to the page it came from. Not because it looks thorough, but because a claim you can't check is worth very little. If our summary says someone is a designer in Austin, you should be able to click through and see the page that says so, and decide for yourself whether that page is credible.

It also keeps us honest. A system that has to show its sources cannot quietly fill gaps with plausible-sounding invention, because the gap becomes visible.

The AI summarises; it doesn't know

There's an AI layer that reads what was found and writes it up, and answers follow-up questions. Its job is narrow: describe the evidence that was collected. It is not a source of facts in its own right.

That means it should say "no public record ties this person to that address" rather than guessing, and "possible match" rather than asserting an identity the evidence doesn't support. A model asked about a person will happily produce a fluent, confident, wrong answer, and the entire engineering effort here is aimed at not shipping that.

Matching is deliberately conservative

Most wrong answers in this category come from merging two different people who share a name. So the matching is built to resist that:

• names alone never establish a match
• a single signal is a hypothesis, not a conclusion
• corroboration across independent sources is what raises confidence
• results are ranked with a confidence score, so you choose the right person rather than being handed one

The consequence is that we sometimes return less than a competitor would. That's the intended trade. A confident wrong profile about a real human being is a worse outcome than an incomplete one. We wrote a longer note on that decision: Why a name is never enough.

What we won't support it being used for

Some of this is law and some is our own line. Both are in our terms.

Not for FCRA-covered decisions. We are not a consumer reporting agency. Results must not be used to decide employment, credit, housing, insurance, or tenancy. Those decisions require a screening provider operating under the Fair Credit Reporting Act, with the subject's consent, their right to see the report, and their right to dispute it. We provide none of that, which is exactly why we say don't use us for it. The longer note is A people-search result is not a consumer report.

Not for stalking or harassment. Aggregating public information to monitor, locate, or pressure someone who doesn't want contact is the harm this category is criticised for, and the criticism is fair. The information being public doesn't make the aggregation harmless.

Not for doxxing. Publishing someone's home address or workplace to expose them is not research.

If you're in the results

If you appear in a result and want the profile removed, contact us with the details and we'll handle it. That's in our privacy policy and the route is on the contact page.

We think that's a basic obligation for anyone in this business. A company that profits from surfacing information about people and makes removal difficult has its incentives pointed the wrong way.

Why publish this

Partly because we think it's the right way to run this, and partly because it's a commitment that's easy to check. Every constraint above is either visible in the product - the source links, the confidence scores, the hedged language - or written into our terms.

If you find us falling short of any of it, we'd like to hear about it. The contact page reaches us.