Seven situations agents hit every week — grouped by what you're trying to do,
with what ordinary tools return and what Search returns instead. Each has its
own page:

## Reach sources other tools can't

- [The blocked portal](/search/examples/blocked-portal.md) — question to
  cited answer, captured end to end: the host's native fetch refuses the
  portal outright; Search reads the same page in a full browser and the
  agent answers with citations, provenance included.
- [The quiet failure](/search/examples/quiet-failure.md) — ordinary tools serve the challenge page as "content"; Search reports an explicit failure — shown with a real captured error, both routes named.

## Control what reaches your context

- [Schema-first discovery](/search/examples/schema-discovery.md) — `schema: true` returns an unknown source's shape before the payload: select-ready paths for JSON, a heading outline for markdown; the cached follow-up extracts exactly what's needed.
- [The oversized source](/search/examples/oversized-source.md) — the compare endpoint answers with 1.26 MB; `select` trims it server-side to 28 KB, the fields alone, shape intact.

## Automate research runs

- [Queued reads](/search/examples/queued-reads.md) — a captured run: four URLs queued, acknowledged instantly, all resolved in 29 seconds — three read, one genuine failure reported verbatim and retried.
- [The nightly release digest](/search/examples/release-digest.md) — a
  production run from question to posted digest: the generated summary with
  per-line PR citations, its two selected reads (28 KB of a 1.26 MB
  response), and the measured cost.

## Understand the basics

- [Agent-friendly content: search and direct fetch](/search/examples/fast-path.md) —
  question to cited answer on an open site: ranked structured results, the
  page as clean markdown capped with `max_bytes` — next to the uncitable
  paraphrase ordinary tools return.