Pilot feature
Profile Details, Carefully
Found results can now show who is behind them. The rule we built it around: less detail is fine, wrong detail is not.
When PROBE confirms that an account exists, the result can now show some of what that profile says about itself: a profile picture, display name, short bio, location, public links, when the account was created, and its follower count. It is a pilot. It works on part of the catalog, and we are watching it on live searches before calling it finished.
Where the details come from
Every search already loads a page from each site to decide whether the account exists. Most of those pages also describe the profile in standard, machine-readable ways: the preview tags sites add for link sharing, structured data meant for search engines, or the JSON a site's own app reads. PROBE now reads those same fields from the page it already has, so for most sources the details cost no extra requests.
For nine sources, the page that answers "does this account exist?" is not the profile itself. Steam, Roblox, Tumblr, and a few others are checked through a lightweight endpoint that confirms the account but says nothing about it. For those, PROBE makes one extra request to the public profile page, but only after the account is confirmed found, only over a secure connection, only if that page can fill something still missing, and within a strict time limit. If that request fails, the result keeps its verdict and simply shows fewer details.
Proving a value before showing it
Reading a field is easy. Knowing it describes the person and not the website is the hard part. A "name" tag might hold the site's brand, a "bio" might be the same marketing sentence on every page, and a "picture" might be the site's logo.
So a rule is only accepted for a source when all of this holds:
- Two real accounts agree on where the value lives. The same field has to produce a believable value for two different known accounts.
- A made-up account does not produce it. We check a randomly generated username alongside them. Anything the fake account also returns, such as a default avatar or a generic description, is rejected for that source.
- A person reviewed the output. We read every extracted value from a live run. Fields that were templated or not really about the account were switched off by hand. In this round that included auto-generated "played N games" summaries on Lichess, post-count prefixes on a Mastodon server, and boilerplate on Roblox profiles.
- The value mentions the right account. A follow-up page is only used if it actually contains the username that was searched.
Values are cleaned conservatively: placeholder names like "Private" are dropped, bios that just repeat the username or a number are ignored, and site names tacked onto the end of a display name are removed. If anything looks doubtful, the field is left empty.
Honest coverage
An earlier survey of the automatic sources found some kind of profile value on far more sites than now show details. The gap is deliberate. Some values could not be proven with two accounts, some sources block automated requests, and some only reveal profile details after scripts run in a full browser, which a normal search does not do.
In our latest measurement run, 117 of the 211 results PROBE reported as found carried profile details, and none of the results for missing accounts did. In a separate live check of every source with a rule, 119 sources returned details for a known account and none returned details for a made-up one. The accuracy of the found and not-found decisions was unchanged by the feature, at 98.83% with zero observed false matches. Reading the details added roughly 50 milliseconds to a typical source check.
Those numbers come from our own network. Some sites treat servers more strictly than home connections, so live coverage may be somewhat lower, and that is one of the things this pilot is meant to show.
What is not stored
Profile details are read while the search runs and sent straight to your browser with the result. PROBE does not save them, build profiles from them, or keep them after the search. The details are the site's own public description of the account, and a found result with a picture and a name is still a lead, not proof that two accounts belong to the same person.
What comes next
The free details will keep improving as we add rules and re-check existing ones. Sites that only show profile details after scripts run, or that block ordinary requests, need a real browser and sometimes extra network routing. That costs money on every request, so it will come later as an optional deeper lookup that someone chooses for a specific result. Everything described here stays part of the normal, free search.