---
title: "How to use web search and site reading in the AI Assistant"
description: "Example requests: current information from the internet, reading URLs, collecting data from a public client website."
---

# How to use web search and site reading in the AI Assistant?

If an administrator enabled **web search** and/or **URL fetching**, the AI Assistant can supplement answers with **public internet** data. This does not replace ConnectiveOne documentation — platform questions still rely on built-in docs.

## When you need this

- You need **up-to-date information** (news, rates, partner API updates).
- You have a **link to a public page** and want a summary or fact table.
- You want to **collect reference data from an open website** (product catalog, FAQ, public contact pages).

## Good to know

- **ConnectiveOne documentation** takes priority for platform questions.
- The assistant **cites sources** (URLs) when using web search.
- **One message** allows a limited number of searches and page fetches; large jobs continue across **several messages** (“continue the next batch”).
- Internet content is **reference only** — verify critical decisions manually.
- **Fetch URL** returns **page-based text**, not raw HTML and not a guaranteed verbatim copy: the assistant produces **model-generated summary** via the selected provider API (Gemini / Claude / GPT). Large tables, exact numbers, or legal wording may be shortened — verify critical data at the source URL.

## Before you start

- [x] Required features are enabled — see [Enable internet search and URL reading](./enable-ai-assistant-web-research.md).
- [x] Chat **model** has a configured API key (Gemini, GPT, or Claude).

## Example requests

### Current information (web search)

- “What are the latest Stripe API changes for charges?”
- “Find the official Telegram Bot API page about webhooks”
- “Briefly summarize what’s written today about [topic], with links”

**Result:** answer with source URLs. In debug mode — step **Web search → "query" (N sources)**.

### Single page by URL (Fetch URL)

- “Read https://docs.example.com/api/auth and explain OAuth steps”
- “Here is our public pricing page — extract the plan table: https://…”

**Result:** page summary (model-generated text, not HTML) or a structured fact table. If the URL is blocked (login, internal address), the assistant explains why.

### Data from your public site (multiple pages)

State **scope** clearly:

1. “I need the product catalog from `/shop/` on site.example.com — find the sitemap first and show a plan for how many pages”
2. After the plan: “Continue the next 5 URLs from the queue”
3. “Save the result as a table: name, price, URL”

**Result:** the assistant builds a URL queue (sitemap, robots.txt, or navigation), reads pages **in batches**, and reports progress. To load into **Custom Data** or a knowledge base — use **Agent** mode with edit permission and state the target explicitly.

## Work modes

| Mode | Web search / URL | Platform changes |
| --- | --- | --- |
| **Ask** | Yes (if enabled) | No |
| **Agent** | Yes | Yes, per role |
| **Plan** | Yes during discovery | After plan approval |

In **Plan** mode for large data collection, the assistant shows an **execution plan** first (scope, steps, where to store results).

## How to verify it works

- The answer includes **source links** or quotes a specific URL.
- With **Agent Debug Mode**, you see **Web search** / **Fetch URL** steps.
- If the feature is off, the assistant says web access is unavailable and suggests contacting an administrator.

## After you finish

- You get an answer backed by public sources or the content of a page at the URL you provided.
- For large sites, the assistant shows progress and may ask you to continue in the next chat message.

## Related materials

- [Enable internet search and URL reading](./enable-ai-assistant-web-research.md)
- [Use the AI Assistant](./use-ai-assistant.md)
- [What is the AI Assistant](../explanation/what-is-ai-assistant.md)