How to build LinkedIn hiring intelligence from public jobs

Ask a question the data can answer

Useful questions include: which sampled companies currently advertise AI roles in the US, which role families dominate a company's disclosed vacancies, or which markets appear in a company's current public listings? Avoid questions that require private headcount, applicant identity, hiring completion, budget, or purchase data.

These workflows use public job-posting pages and retain source-linked evidence. They do not require a LinkedIn login or collect private profiles, employee directories, or personal email addresses.

Open LinkedIn Company Hiring Signals when you need company-level aggregation. Use broad Jobs Search when individual vacancies are the product.

Choose consistent inputs

Start with two role families, one geography, and a recent posting window. Keep the same inputs across companies you intend to compare:

{
  "queries": ["machine learning engineer", "AI product manager"],
  "locations": ["United States"],
  "datePosted": "r604800",
  "minimumCompanyJobs": 2,
  "maxResultsPerQuery": 50
}

A seven-day window is responsive but sparse. A longer window offers more rows but can include roles no longer central to current activity. Record the exact query terms, location, date window, caps, and run timestamp with every analysis.

Inspect evidence before the score

For each company row, verify total retained jobs, role-family counts, location counts, sample titles, source job URLs, coverage notes, and the time the data was collected. A useful score should be reconstructable from those fields. If the evidence URLs do not support the label, do not keep the label.

Compare companies without hiding coverage

1. Use the same role dictionary, geography, date window, and row caps.
2. Remove vendor self-listings when researching demand for an external tool.
3. Require every requested segment to have data before publishing a benchmark.
4. Show raw counts and shares together; a large share from two jobs is not the same as a large share from 200.
5. Retain sample source URLs for every category.
6. State that public search visibility, ranking, and pagination bound the sample.

Write conclusions at the evidence level

Prefer “the sampled listings show” or “the bounded search retained” over “the company is expanding.” Label inferred categories and preserve unknowns. Technology mentions can indicate role requirements, migration work, internal support, or incidental text; they are not purchase records. A job disappearing from search is not proof that it was filled or canceled.

Operationalize the analysis

• Save reviewed inputs as public or private Actor Tasks.
• Schedule only after a bounded run succeeds.
• Write run ID, input hash, and retrieval time into the downstream table.
• Use webhooks to trigger ingestion after success, then validate dataset completeness.
• Separate owner QA, unattributed page views, external runs, and paid usage.
• Re-run comparisons on a consistent cadence rather than cherry-picking dates.