> Parel documentation. Index of every page: https://parel.si/docs/llms.txt. This page in HTML: https://parel.si/docs/tools/search_web

# search_web

> Google organic results for a query as structured rows, for a chosen country and language. ₹0.14 per call; an empty result list is ₹0.

## What it does

Google's organic ranking for a query in a market: up to ten rows with position, title, link and snippet, in `data.output.results`. Use it when the task needs the actual ranking for a country and language; for a quick fact the assistant's own search is free and better. It returns organic rows only: no AI Overview unless `ai_overview` is true, no news ([`search_news`](https://parel.si/docs/tools/search_news.md)), no page text ([`read_page`](https://parel.si/docs/tools/read_page.md)).

India: pass `country: "in"` and `language: "en"` or `"hi"`; the defaults are the United States and English.

## Inputs

Generated from the contract captured from staging on 2026-11-05 ([`GET /v1/tools/search_web`](https://parel.si/docs/api/tool.md)).

| Field | Type | Required | Description |
|---|---|---|---|
| `q` | string | one of | identity |
| `limit` | integer | no | filter; default 10; results to return where the provider bills by depth |
| `country` | string | no | filter; default 'us'; ISO country code; mapped to each provider's location code / gl / region |
| `language` | string | no | filter; default 'en' |
| `ai_overview` | boolean | no | identity |

One of these sets must be complete (anything else is `invalid_input`, ₹0): `q`.

## What drives the cost

Nothing you can turn: the price is flat per call whatever `limit` says. Keep `limit` at the default 10: a deeper page costs more at the source and can be refused by the per-call cost cap as [`estimate_too_low`](https://parel.si/docs/errors.md#estimate_too_low), which is ₹0.

## Price today

| | |
|---|---|
| Price today | **₹0.14** per call (`cost_paise: 14`) at ₹93.93 per US dollar on 17 November 2026 |
| Hold before the call | ₹0.38 (`hold_paise: 38`); when the call settles the real price is charged and the unused part is released |
| ₹0 when | it fails: ₹0 |
| Timeout | 30 s, then `upstream_timeout` (₹0, retryable) |
| Over MCP | a headline tool: call `search_web` directly |
| Over REST | `POST https://api.parel.si/v1/run` with `"tool": "search_web"` |
| Confirmation | not needed (the estimate is under the ₹200 threshold) |

## Request

```bash
curl https://api.parel.si/v1/run \
  -H "Authorization: Bearer $PAREL_KEY" \
  -H "Content-Type: application/json" \
  -d '{"tool":"search_web","input":{"q":"best crm for small business","limit":10,"country":"in","language":"en"}}'
```

Over MCP, `search_web` is a headline tool: Claude calls it directly with the same fields as arguments.

## What comes back

`data.output.results[]`, one row per organic result: `position`, `title`, `link`, `snippet`. `data.output` is the whole answer. An empty `results` is `class: "ok_free"`, ₹0. No recorded example of a success is shown here: the staging environment's connection to the source was being re-keyed when these pages were captured, so the shape above is from the tool's contract.

## Notes

- Organic rows only: no news, maps, images, ads or page text.
- Keep limit at 10. A deeper page can be refused by the cost cap (estimate_too_low); that refusal is free.
- An empty result list is free for you (empty_list rule) even when the provider charged for the miss.
- India: country "in", language "en" or "hi". The upstream default is the United States.

## Siblings

- [`search_news`](https://parel.si/docs/tools/search_news.md) for recent articles instead of web pages.
- [`read_page`](https://parel.si/docs/tools/read_page.md) to read one of the results in full.
- [`check_ai_visibility`](https://parel.si/docs/tools/check_ai_visibility.md) for what ChatGPT, not Google, answers.
