Practical research examples
Copy focused questions, select output types and test a representative sample.
Use these patterns with typed enrichment: enrichment_start in MCP or enrichment.start through the HTTP API. Each item supplies known facts in input and one question. Submit 1–100 items in one batch, including many phone and email requests. To keep results in a sheet, use sheets_research for new rows or sheets_enrich for existing rows. These are task examples, not pre-verified results.
For hiring questions, false means a negative answer, not an absent result. If the evidence does not establish the answer, it should remain unknown. An allowed-value list restricts format; it does not justify choosing a country when none fits.
Build a useful pilot
Select a few representative inputs, including an ambiguous name and a case where you expect no result. Review entity identity, sources, answer type and actual cost. A well-known company alone is not a representative quality benchmark.
Once the pilot meets your needs, send larger batches. Keep the same question and output definition when comparing runs. Changes to the question, available inputs or source freshness can change the answer.
Use the catalog for fixed operations
If you need an existing platform operation such as a job search or web search, first look for it in the Catalog. A direct HTTP call avoids an extra research-model decision when the tool and parameters already express the task. Direct catalog operations are not public MCP tools. A connected MCP agent uses research when a question requires interpreting or combining evidence.