Search

Search returns what the web says about a query, as a ranked list. It is the thinnest product in the API: a query goes in, results come out in the same response, and nothing is summarised or interpreted for you. A call costs 5 points per page of 10 results — so 10 results cost 5 points, 30 results cost 15.

When to use it

Use Search when you want the links, not an answer:

discovery before a deeper lookup — find the page, then hand it to AI Extract
a “sources” or “related links” panel in your product
checking what is publicly indexed about a company or person
feeding your own ranking, filtering or RAG pipeline

If you want the sources read and synthesised into a report, that is AI Research — it runs the same searches, then opens the pages and writes the answer.

Example request

curl -X POST "https://api.neuralverge.ai/functions/v1/run-search" \
  -H "Authorization: Bearer YOUR_ACCESS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "gartner magic quadrant data loss prevention 2026",
    "settings": { "max_results": 3, "country": "us", "language": "en" }
  }'

{
  "session_id": "11111111-1111-1111-1111-111111111111",
  "kind": "search",
  "query": "gartner magic quadrant data loss prevention 2026",
  "pages_requested": 1,
  "results": [
    {
      "rank": 1,
      "title": "Best Data Loss Prevention Reviews 2026",
      "url": "https://www.gartner.com/reviews/market/data-loss-prevention",
      "snippet": "GTB Technologies DLP is a software designed to address data loss prevention needs...",
      "source": "organic"
    }
  ],
  "total_points": 5
}

Request reference

Field	Type	Description
`query`	string	The query, in plain text. Required.
`settings.max_results`	number	1 to 100, default 10. Rounded up to whole pages of 10 for billing.
`settings.country`	string	Two-letter country code, default `us`. Biases results and the exit location.
`settings.language`	string	Two-letter language code, default `en`.
`settings.device_type`	string	`desktop` (default) or `mobile`. Mobile results differ on many queries.

Response reference

Field	Description
`results[]`	Ranked results: `rank`, `title`, `url`, `snippet`, and `source` (`organic`, `ad` or `other`).
`pages_requested`	How many pages of 10 were fetched — this is what you are billed for.
`session_id`	The run, readable again through Get Session Status, even though there is nothing to poll.
`total_points`	`pages_requested` × 5.

Notes worth knowing

Ads are labelled, not hidden. Check source before treating a result as organic.
max_results is a ceiling, not a promise. A narrow query can return fewer results than requested; you are billed for the pages fetched.
Ask for what you need. max_results: 11 costs the same as 20, because it spans two pages.
No deduplication across calls. Paging through the same query twice returns overlapping results; dedupe on url if that matters.

Typical pattern

run-search  →  pick the right URL  →  run-extract on that URL

Two calls, 10 points, and you have structured fields from the best page for a query — without writing a crawler.

Next steps

Run Search — full request and response reference
AI Extract — turn one of these URLs into fields
AI Research — searches, reads and synthesises in one run

Introduction

Product

API Basics

Integrations

Search

Search

When to use it

Example request

Request reference

Response reference

Notes worth knowing

Typical pattern

Next steps

​Search

​When to use it

​Example request

​Request reference

​Response reference

​Notes worth knowing

​Typical pattern

​Next steps

Search

When to use it

Example request

Request reference

Response reference

Notes worth knowing

Typical pattern

Next steps