You can feed live website data into an AI agent without writing code. Browse AI collects the data from the website and keeps it current, and Make or n8n passes each new result to your AI model or agent, then sends the answer wherever you need it.
The flow looks like this:
Website → Browse AI robot → Make or n8n → AI step (ChatGPT, Claude, Gemini or your own agent) → Slack, email, a spreadsheet or your app
Use this when you want an agent to work from current information it can't reliably fetch on its own: competitor prices, product availability, listings, job postings, policy or pricing pages, or any list you check by hand today.
📖 Just want to ask ChatGPT or Claude about a dataset once? You don't need automation for that. See how to use Browse AI data with ChatGPT, Claude and other AI assistants.
Before you start
A Browse AI robot that extracts the data you want. Point Browse AI at the page, AI proposes the columns, and you review and approve them.
Optional: a monitor on the robot, so it checks the page on a schedule and your agent gets new data automatically.
A Make or n8n account.
An account with the AI provider you want to use (OpenAI, Anthropic or Google), or your own agent's endpoint.
API access on your Browse AI workspace. Webhooks and the Make integration both run on the Browse AI API.
Which option to choose
Make | n8n | |
How it connects | Browse AI app in Make (sign in, pick your robot) | A Browse AI webhook sends data to an n8n Webhook node |
Setup | Easiest, fully no-code | No code, one copy-and-paste step |
Best for | Teams who want a quick, managed setup | Teams who want to self-host, or use n8n's AI Agent node |
Option 1: Make
Step 1: Start the scenario with Browse AI
In Make, create a new scenario.
Add the Browse AI app and choose the Watch Tasks trigger.
Connect your Browse AI account when asked.
Pick your robot.
Choose when the scenario should run:
Task finished successfully, to process every new set of data.
Change detected in captured data, to react only when a monitor finds something new on the page.
For the full connection steps, with screenshots, see how to integrate your robot with Make.com.
Step 2: Shape the data
If your robot extracts a list (for example, a table of products), add an Iterator to handle each row, or an Array Aggregator to send the whole list to the AI step at once. Sending the whole list is usually better for summaries and comparisons.
Step 3: Add the AI step
Add the module for your AI provider, for example OpenAI (ChatGPT), Anthropic (Claude) or Google Gemini, and map the Browse AI data into the prompt. A prompt that works well for monitoring:
You are a monitoring assistant. Here is the latest data from a website we track: [map the Browse AI data here]
1. Summarize what is new or changed in 2 to 3 sentences.
2. Say whether anyone needs to act, and why.
3. Answer in plain language for a busy team.
Sending data to your own agent instead? Use Make's HTTP module to POST the data to your agent's endpoint.
Step 4: Send the answer somewhere useful
Add one more module for the destination: a Slack or Microsoft Teams message, an email, a new row in Google Sheets or Airtable, or a record in your CRM. Then turn the scenario on.
Option 2: n8n
Step 1: Create the webhook in n8n
Create a new workflow in n8n.
Add a Webhook node, set the method to POST and the response mode to respond immediately.
Copy the webhook URL n8n gives you.
Step 2: Point your Browse AI robot at it
Open your robot in Browse AI and go to Integrations.
Choose Webhooks, paste the n8n URL and pick when to call it:
Task is finished successfully, to send every new set of data.
Monitoring check is finished successfully with changes detected, to send data only when something changed.
Save, then send a test with the last task so n8n receives a sample.
Step 3: Add the AI step
Add an OpenAI, Anthropic or Google Gemini node, or n8n's AI Agent node if you want the model to use tools and memory. In the prompt, map the Browse AI data from the webhook: task.capturedLists holds lists (for example, a table of products) and task.capturedTexts holds single fields.
Step 4: Route the result
Add a destination node: Slack, email, Google Sheets, a database, or an HTTP Request node to your own app or agent.
For more n8n patterns, including scheduled pulls from the Browse AI API, see how to process Browse AI data with n8n.
Let your agent request fresh data on demand
The steps above push data to your agent whenever the robot runs. If your agent needs data at the moment it's asked a question, it can start a run itself:
In Make: use the Browse AI Run a Task action, then wait for the result in the same scenario.
In n8n or your own code: call the Browse AI API to run a task, then read the result when it finishes.
See the Browse AI API documentation for the endpoints.
Tips
Tip | Why it helps |
Use change detection for monitoring | Your AI step only runs when the page actually changed, so you aren't paying for or reading answers about the same data twice. |
Send only the fields you need | Smaller prompts give clearer answers and cost less with your AI provider. |
Ask for a decision, not just a summary | "Does anyone need to act?" turns an alert into something your team can use. |
Test with a past task | Send the last task to Make or n8n while you build, so you don't have to wait for a new run. |
Next steps
Process the data with Claude: see how to automate workflows with Browse AI and Claude.
Connect any LLM through the API: see how to connect Browse AI to any LLM.
Coming soon: a Browse AI MCP server, so agents like Claude can call Browse AI directly.