Person Batch Verify - Crustdata Docs

Batch verify takes a list of people you already have and answers one question for each: is the work history on their profile real? Each result carries a verdict, a confidence score, the reasoning behind it, and the web sources it relied on.

POST https://api.crustdata.com/batch/person/verify

Verification is an add-on, enabled per account. If your plan does not include it, contact your Crustdata account team. Verification jobs are not billed today.

How a profile is checked

For each person the service runs three checks, then asks a language model to weigh the whole picture:

Dates: a full-time job that overlaps full-time study by more than six months is flagged.
Social graph: the person’s own recent posts are fetched. The people who reacted or commented are looked up in the Crustdata people database and matched to the claimed employer. Coworker engagement can lift a verdict to probably_genuine, never higher.
Web search: the model searches the web for independent evidence of the current role. Profile mirror sites that only restate the claim carry no weight.

The current role decides the verdict. Old roles that cannot be verified add a past_roles_unverified flag instead of dragging the verdict down. A job usually finishes in a minute or two.

Submit a job

POST /batch/person/verify takes professional_network_profile_urls, a list of person profile URLs, with up to 10 values per job for now. Sending none returns 400. You can also pass webhook_url, a URL on your side. When the job finishes, Crustdata sends a POST to it with the download link, so you do not have to poll. See Skip polling with webhooks.

curl --request POST \
  --url https://api.crustdata.com/batch/person/verify \
  --header 'authorization: Bearer YOUR_API_KEY' \
  --header 'content-type: application/json' \
  --header 'x-api-version: 2025-11-01' \
  --data '{
    "professional_network_profile_urls": ["https://www.linkedin.com/in/rishabhhq/"]
  }'

{
    "batch_id": "b82e4e5e-8812-4f47-bccd-0439315695b0",
    "status": "pending",
    "entity": "person",
    "action": "verify",
    "identifier_count": 1,
    "entities_requested": 1,
    "status_url": "/batch/b82e4e5e-8812-4f47-bccd-0439315695b0"
}

The action is verify wherever you poll or list jobs.

Poll for the results file

Poll GET /batch/{batch_id} until status is completed. The response then carries download_url, a link to a gzipped JSON lines file with one record per person. The link expires after 5 days.

curl --request GET \
  --url https://api.crustdata.com/batch/3f806ed3-ec4c-4f10-a853-a0f05c723f7d \
  --header 'authorization: Bearer YOUR_API_KEY'

{
    "batch_id": "3f806ed3-ec4c-4f10-a853-a0f05c723f7d",
    "status": "completed",
    "entity": "person",
    "action": "verify",
    "identifier_count": 3,
    "result_count": 3,
    "entities_requested": 3,
    "entities_fulfilled": 3,
    "created_at": "2026-09-10T05:41:24.883865+00:00",
    "completed_at": "2026-09-10T05:43:15.198554+00:00",
    "download_url": "https://crustdata-batch-api-data.s3.amazonaws.com/3f806ed3-ec4c-4f10-a853-a0f05c723f7d/output/results.jsonl.gz?...",
    "download_urls": [
        "https://crustdata-batch-api-data.s3.amazonaws.com/3f806ed3-ec4c-4f10-a853-a0f05c723f7d/output/part-000.jsonl.gz?..."
    ]
}

Download and unpack it:

curl -s "$DOWNLOAD_URL" -o results.jsonl.gz
gunzip results.jsonl.gz

Read the results

When the job completes, each line in the downloaded file is one person wrapped in the standard batch envelope. original_identifier echoes the URL you sent, and internal_id is the numeric Crustdata person ID.

One record from the results file

{
    "original_identifier": "https://www.linkedin.com/in/rishabhhq/",
    "internal_id": 1052069,
    "data": {
        "verdict": "probably_genuine",
        "confidence": 0.65,
        "reasoning": "Current Crustdata Founding Engineer role is carried to 'supported' by social-graph engagement meeting the 3+ verified-employees/2+ posts threshold, corroborated by a Product Hunt maker listing that supplies identity anchors alongside known Crustdata staff. The three past roles have no independent corroborating or contradicting evidence; these are short internships and a mid-level IC role at scale, which is expected to be low-visibility and does not lower the overall assessment.",
        "flags": [],
        "coverage": {
            "social_ran": true,
            "posts_fetched": 2,
            "posts_checked": 2,
            "engagers_seen": 58,
            "engagers_resolved": 56,
            "coworker_hits": 7,
            "timeline_flags": 0,
            "flags": []
        },
        "citations": [
            {
                "url": "https://www.producthunt.com/products/crustdata-2/makers",
                "title": "Crustdata Makers and Employees (2025) | Product Hunt",
                "cited_text": "Rishabh Raj Building Crustdata · People Dataset ..."
            }
        ]
    }
}

reasoning and cited_text are shortened here. A profile that cannot be resolved, or whose check fails, is left out of the results file rather than written as an error, so result_count can be lower than identifier_count.

Fields in `data`

Field	Type	Meaning
`verdict`	string	One of the seven values below.
`confidence`	number	`0` to `1`. Bands are fixed per verdict, so the same verdict lands in the same range across runs.
`reasoning`	string	Plain-text explanation of the verdict.
`flags`	array	Date conflicts and other caveats found on the profile, for example `past_roles_unverified`.
`coverage`	object	What the checks managed to look at, as counts (see below).
`citations`	array	Each source the model relied on: `url`, `title`, and the `cited_text` it quoted.

Verdicts

`verdict`	Meaning
`clearly_genuine`	Independent sources confirm the current role.
`probably_genuine`	Evidence supports the claim, for example coworker engagement, but does not confirm it outright.
`cannot_verify`	No usable evidence either way. Expected for people with little web footprint.
`probably_fabricated`	No evidence was found where some was expected, given the seniority or visibility of the claimed role.
`clearly_fabricated`	Evidence contradicts the claim.
`reclassified_employment`	The relationship is real but not what the profile says, for example an advisor listed as an employee.
`non_person`	The profile is not a person, for example a company or a group page.

Coverage counts

Field	Meaning
`social_ran`	Whether the social graph check ran at all.
`posts_fetched`	Recent posts by the person that were retrieved.
`posts_checked`	Posts whose reactions and comments were examined.
`engagers_seen`	Distinct people who reacted or commented.
`engagers_resolved`	Engagers that resolved to a record in the people database.
`coworker_hits`	Resolved engagers who work at the claimed employer.
`timeline_flags`	Date conflicts found by the dates check.
`flags`	Caveats from the checks themselves, for example `posts_unavailable` when no posts came back.

posts_unavailable means no posts were retrieved. It cannot tell a retrieval failure from a person who never posts, so read a cannot_verify with this flag as “unchecked” rather than “suspicious”.

Rate limits

Submissions are limited to 5 per minute, and verify jobs count toward the live pool of active jobs. See Rate limits.

Errors

Sending a body without professional_network_profile_urls returns 400 with error.type of invalid_request.

What to do next

Need the full profile too? See Person Batch Enrich to pull the record you are verifying.
Only have emails? Person Batch Identify resolves them to profiles first.

​How a profile is checked

​Submit a job

​Poll for the results file

​Read the results

​Fields in data

​Verdicts

​Coverage counts

​Rate limits

​Errors

​What to do next

How a profile is checked

Submit a job

Poll for the results file

Read the results

Fields in `data`

Verdicts

Coverage counts

Rate limits

Errors

What to do next