Social Profile Finder

You rarely have the full picture of a subject, just one clue. A name, a company, a website domain, an email address, an @handle: any one of those is enough to start from here. What comes back is the public profiles that belong to that subject on X (Twitter), Instagram, LinkedIn, Facebook, TikTok, YouTube and GitHub.

Every row carries the profile URL, the handle, the display name on the account, and a confidence score with the evidence behind it.

No account, no cookies, no login, no browser.

• One input box takes all of it. "Patagonia", "patagonia.com", "press@patagonia.com", "@patagonia" or a profile URL.
• Seven networks per lookup, each checked on that network's own public surface, so a match is confirmed rather than guessed.
• Every row scores the match from 0 to 0.99 and lists the evidence that produced the score, so you can audit it instead of trusting it.
• Set a confidence floor and weak matches are never written, never charged and never land in your CRM.
• Empty input returns one labelled sample row, free.

Price

$0.90 per 1,000 profiles. Plus $0.001 per run as the platform start fee, billed per gigabyte of run memory, so exactly that on the default 1 GB.

No volume tiers, no minimum spend, no subscription.

Profiles	Total
100	$0.091
1,000	$0.901
10,000	$9.001
100,000	$90.001

What gets charged

• One profile-found event per profile row written to the dataset. Nothing else is metered per row.
• Free: the sample row an empty run returns, and every diagnostic row. They all carry "charged": false.
• Lookups that find nothing produce an uncharged diagnostic row explaining why.
• Profiles scoring below your confidence floor are dropped before they're charged.
• Duplicate profiles found twice in the same run are collapsed before they're charged.
• A run that finds nothing costs the start fee and nothing else.
• Rows never leave the dataset without a charge and are never charged without a row. The billed event is a named one, so nothing is quietly attached to apify-default-dataset-item.

Input

{
  "queries": [
    "patagonia.com",
    "Marques Brownlee",
    "press@patagonia.com",
    "@MrBeast",
    "https://github.com/torvalds"
  ],
  "networks": ["x", "instagram", "linkedin", "facebook", "tiktok", "youtube", "github"],
  "minConfidence": 0.65,
  "maxItems": 50
}

Field	What it does
queries	What to look up, one entry per line. A person or company name, a domain, an email address, an @handle, or a link to a profile you already have. Up to 50 per run. You can also pass objects like {"name":"Patagonia","domain":"patagonia.com"} when you have more than one clue about the same subject. Combining them raises the confidence of the result.
networks	Which networks to check. Leave empty for all seven. Dropping the ones you don't need makes each lookup faster and cheaper, because a network that isn't checked is never charged for.
minConfidence	The confidence floor, 0 to 0.99. Default 0.65, which is deliberately strict. Lower it to about 0.5 if you'd rather review weak matches yourself. Raise it to 0.8 for self-declared and badge-verified profiles only.
maxProfilesPerNetwork	How many profiles to keep per network per lookup. Default 1, the highest-scoring one. Raise it to 2 or 3 for brands that legitimately run several accounts on one network.
maxHandleVariants	How many handle spellings to try when all you have is a name: "janedoe", "jane.doe", "jane_doe" and so on. Default 3, maximum 5. More variants means more coverage and more requests.
checkWebsite	Whether to read the subject's own website for the profiles it links to. Default on, and it's the single strongest signal available, so leave it on unless you only ever pass bare handles.
maxItems	Total profiles across every lookup in the run. Default 50, hard ceiling 2,000. Keep it low while testing, since you pay per profile.
proxyUrls	Leave empty. Only for callers who want traffic to leave through proxy servers they already pay for, as http://user:pass@host:port.

Run it with empty input and you get one labelled sample row, free.

Output

One row per profile. This is a real row from a real run:

{
  "ok": true,
  "charged": true,
  "recordType": "profile",
  "query": "patagonia.com",
  "subject": "patagonia",
  "subjectDomain": "patagonia.com",
  "network": "instagram",
  "networkLabel": "Instagram",
  "profileUrl": "https://www.instagram.com/patagonia/",
  "handle": "patagonia",
  "displayName": "Patagonia",
  "matchConfidence": 0.95,
  "confidenceLabel": "high",
  "matchEvidence": "verified-on-network | linked-from-own-website | structured-data-sameas | display-name-matches | platform-verified-badge | established-audience",
  "source": "website-link",
  "followers": 5498775,
  "platformVerified": true,
  "bio": "We're in business to save our home planet.",
  "profileWebsite": "http://sprout.link/patagonia/",
  "verifiedLive": true,
  "scrapedAt": "2026-08-16T15:45:01.719Z"
}

Field notes

• network is one of x, instagram, linkedin, facebook, tiktok, youtube, github. Lower case and stable, so it's safe to switch on in code.
• profileUrl is the canonical, openable profile address. For a YouTube channel found by link this is the /channel/UC... form, which doesn't break when a creator changes their handle.
• handle is the account name as that network spells it, taken from the network's own answer rather than your input, so capitalisation and any redirect are already resolved.
• displayName is the name shown on the account. Null when the network won't hand it over, which is honest reporting rather than a scraping failure. The confidence score already accounts for it.
• matchConfidence runs 0 to 0.99. Never 1.0, because this is public evidence rather than proof of identity.
• confidenceLabel is high at 0.8 and above, medium from 0.6, low below that. A convenience for spreadsheet filters.
• matchEvidence lists the named reasons behind the score, pipe separated. This is the field to read when a match surprises you.
• source is website-link if the subject's own site published it, handle-probe if a candidate handle was confirmed on the network, cross-link if another confirmed profile pointed at it.
• followers is the audience size where the network publishes it, otherwise null. TikTok, YouTube and LinkedIn member profiles frequently return null here.
• platformVerified is true when the network itself marks the account as verified. Null means the network didn't say, which isn't the same as false.
• verifiedLive is true when the profile was confirmed against the network during this run. False only appears for a link the subject published on their own site that the network wouldn't confirm at that moment.
• subjectDomain is the website the lookup was anchored to. You either passed it directly or it came out of an email address.

Real rows carry "charged": true. Sample rows carry "_sample": true, diagnostic rows carry "_diagnostic": true and an errorCode. Neither is billed.

How it works

Each entry is classified first: an email, a domain, a handle, a profile link or a plain name. Guessing that wrong wastes the whole lookup.

If a website is known, it gets read for the profiles it links to, including the machine-readable sameAs block many sites publish. A profile the subject published themselves is the strongest evidence there is.

Candidate handles are then built from what you gave: the domain label, the email local part, the name run together, and a few common spellings of it. Every candidate is checked against that network's own public surface, the endpoint the site itself uses to render a profile, so an account is confirmed to exist and its display name is read straight from the network. Nothing is reported on a guess.

The confirmed profiles are scored against each other: display name against the name you asked for, the profile's own outbound link against your domain, the network's verification badge, the audience size, and whether one confirmed profile points at another. Anything below your floor is dropped before it's written.

Requests leave through a rotating pool of addresses, so a per-address rate limit is answered by moving rather than waiting.

How the confidence score is built

The score is additive and every term is named in matchEvidence, because a number nobody can audit is a number nobody should trust. A profile starts at zero and collects:

Evidence	Weight	What it means
verified-on-network	+0.30	the network's own endpoint returned this profile during the run
linked-from-own-website	+0.35	the subject's website links to it
structured-data-sameas	+0.05	and it was in the site's machine-readable profile block, not just a footer icon
explicit-handle	+0.20	you supplied this exact handle or profile link
handle-matches-domain	+0.20	the handle is the domain label, e.g. patagonia for patagonia.com
handle-from-email	+0.10	the handle is the local part of the email you gave
handle-matches-name	+0.10	the handle is the name with the spaces taken out
display-name-matches	+0.20	the name on the account covers the name you asked for
display-name-similar	+0.10	it partly covers it
display-name-differs	-0.15	both names were readable and they don't match
links-back-to-domain	+0.15	the profile's own website field points back at your domain
platform-verified-badge	+0.10	the network marks the account as verified
established-audience	+0.05	10,000 followers or more, so not a freshly squatted handle
cross-linked-from-github	+0.10	another confirmed profile of the same subject names this handle
not-verified-on-network	—	the subject's site published this link but the network wouldn't confirm it during the run

The result is clamped to 0.99.

Why the default floor is 0.65. A bare name can only ever earn 0.30 + 0.10 + 0.20 = 0.60, which sits below the default. So a handle that merely exists and merely echoes the name back, the classic squatter or fan account, isn't reported at all. It takes a verification badge, a real audience, a link back to your domain, or the subject's own website vouching for it, before a name-derived guess is shown to you. A wrong profile in a lead list is worse than a blank cell, because the blank cell is honest. Set minConfidence to 0.5 if you'd rather review the weak matches yourself.

The name comparison is asymmetric on purpose. It measures how much of the name you asked for is accounted for by the name on the account. "NASA" is fully accounted for by an account called "NASA - National Aeronautics and Space Administration", so that's a match. "Marques Brownlee" is only half accounted for by an account called "Marques", so it isn't. During testing that exact rule is what stopped an 87-follower impostor from being reported.

What people use it for

• Enriching a lead list. Paste a column of company domains and get their X, Instagram, LinkedIn, Facebook, TikTok and YouTube accounts back as rows, ready to join on the domain.
• Vetting an influencer or partner. One handle in, every network they're on out, with follower counts and verification badges next to each.
• Recruiting research. An email address, or a name plus a company domain, resolves to the public professional profiles.
• Brand protection. Search your own brand name and see every account carrying it, then look at the ones your website doesn't link to.
• Filling gaps in a CRM that has a name and an email but no social handles, on a schedule, with a floor high enough that nothing dubious gets written back.
• Due diligence and journalism, where a documented evidence trail per match matters more than volume.

Reading the output

• Real rows carry "charged": true and "recordType": "profile". One billed event each.
• The sample row carries "_sample": true and "charged": false. There's exactly one, only when the input had nothing to look up.
• Diagnostic rows carry "_diagnostic": true, "charged": false and an errorCode: NO_RESULTS when a lookup produced nothing above your floor, NETWORK when a network couldn't be reached, TIME_BUDGET when time ran out, RATE_LIMITED when a network throttled the run. Each carries the query it belongs to and a plain-English explanation.

Filter on charged == true and you have exactly the rows you paid for.

Limits

• Seven networks are covered: X (Twitter), Instagram, LinkedIn, Facebook, TikTok, YouTube and GitHub. Nothing else is checked, so a subject who is only on a network outside that list comes back empty.
• A plain name with no domain, email or handle is the hardest input there is. Unless the account carries a verification badge, a real audience or a link back to a website you supplied, a name-derived match won't clear the default floor. Give it a domain or an email whenever you have one.
• Instagram meters its public profile endpoint hard per address. When it refuses, the run falls back to a surface that confirms the account exists but returns no display name. If that's refused too you get an uncharged diagnostic row rather than a guess. Instagram is the least complete of the seven.
• LinkedIn answers member profiles with a bot-check status that means "not now" and "doesn't exist" at the same time. That's never treated as a match and never as a clean absence. It produces a diagnostic row, so LinkedIn member coverage is best-effort while LinkedIn company pages are reliable.
• Facebook coverage is Pages and public profiles that permit embedding. A personal profile with tight privacy settings won't be confirmed.
• YouTube channels found by a /channel/UC... link are confirmed and named. A channel with no published videos may come back with a null display name.
• The score is evidence, not proof. A high row means several independent public signals agree, not that identity has been established. Treat 0.99 as "very likely" and check anything that matters.
• Sites behind a bot check may refuse to be read. The lookup still runs every handle check, and an uncharged diagnostic row records that the website couldn't be read.
• Follower counts and verification badges are a snapshot at read time.
• Private, suspended, deleted or renamed accounts are reported as absent, because the network reports them as absent.
• Only public information is read. No login, no cookie, no session belonging to anyone.
• Hard ceilings: 50 lookups and 2,000 profiles per run.

Questions

What exactly counts as a billable profile?

One confirmed profile row that cleared your confidence floor. A lookup that returns nothing isn't billed, a profile below your floor isn't billed, and diagnostic rows are never billed. To spend less, raise minConfidence or trim the networks list.

Why did a lookup come back with fewer networks than I expected?

Either the subject isn't on that network under any handle derivable from what you gave, or the match scored below your floor, or the network wouldn't answer. The diagnostic rows name which of the three it was.

Can I trust a high confidence row without checking it?

For a company with a website, yes in practice. A high row there usually means the company's own site links to the profile and the network confirmed both the account and its name. For an individual found from a name alone, read matchEvidence first. The field exists so you never have to take the number on faith.

Do I need an API key or a login for any of the networks?

No. Everything read here is what a logged-out visitor sees.

How do I look up a thousand companies?

Pass up to 50 entries per run and raise maxItems, then schedule or queue the rest. Nothing is held between runs, so splitting a list changes nothing except how long each run takes.

Will the run fail if a network blocks it?

No. A blocked or throttled network produces an uncharged diagnostic row and the run still finishes as succeeded, with whatever the other networks returned. A failed run would still bill the start fee, which would mean paying to be told something went wrong.

I have a name and a company domain for the same person. How do I pass both?

As one object: {"name":"Jane Doe","domain":"acme.com"}. Both clues are used for the same subject, and the extra evidence raises the confidence of every match instead of producing two separate lookups.

Can I use this to find someone's private accounts?

No, and it isn't built to. It reports accounts that are publicly visible and publicly attributable, with the evidence for each.