Chat agent for your Google Search Console data

Chat agent for your Google Search Console data

An n8n workflow that answers questions about your Search Console data. A webhook takes a message and a session id, and an
OpenAI agent lists your Search Console properties, builds a query from your request, calls Google's API and replies with a
Markdown table. A Postgres table keeps each session's chat history.

What is inside

• workflow.json: the workflow as n8n exports it, 30 nodes, fourteen of them sticky notes. It holds no credential.
  The agent and its tool sit in one file: the tool is the lower half of the canvas, which starts at Tool calling.
• README.md: these notes.

What we changed from the original copy

• Every credential link is gone, so you select your own: OpenAI, Postgres, Basic Auth and the Google OAuth2 one.
• Call Search Console Tool pointed at a workflow id in its author's n8n, which was the id of this same workflow. It now
  reads {{ $workflow.id }}, so the tool calls whichever copy you import.
• The webhook's path was its author's webhook id followed by /chat. It is now search-console-chat.
• The workflow was saved as active. Mindket's copy is inactive until you publish it.
• The big sticky note held the author's screenshots, a link to n8n's guide and some advice on scopes. It now holds a short title.
• Nothing else. The file has no Code node, and its expressions needed no repair.

Before you start

• An n8n instance, on n8n Cloud or your own server, with a public address for the webhook.
• An OpenAI API key.
• A Postgres database for the chat history, such as a Supabase project's. The memory node creates its table.
• A Google Cloud project with the Search Console API enabled, an OAuth client, and a Google account that can see the
  properties you want to ask about.
• A chat front end that can post JSON to a webhook, or a command line tool that can.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OAuth2 API credential for Google with your client's id and secret and the read-only scope
   https://www.googleapis.com/auth/webmasters.readonly, then connect your account. Select it in
   Search Console - Get Custom Insights and ## Search Console - Get List of Properties.
3. Add a Postgres credential and select it in Postgres Chat Memory. The table is insights_chat_histories.
4. Add an OpenAI credential and select it in OpenAI Chat Model.
5. Add a Basic Auth credential with a username and password of your choosing, and select it in Webhook - ChatInput.
6. Save the workflow. Call Search Console Tool calls it by its own id, so it needs the saved copy.
7. Open the webhook's test address and post a message with its Basic Auth login and a body such as
   {"chatInput": "Which sites can you see?", "sessionId": "test-1"}. The reply holds output with the agent's answer and
   intermediateSteps with the raw tool results.
8. Press Publish. The production address ends in /webhook/search-console-chat.

How a run goes

1. Webhook - ChatInput takes a POST with Basic Auth. Set fields copies chatInput and sessionId from the body, or
   from the top level, and adds today's date as date_message.
2. AI Agent reads the message and the date. On the first message of a session its instructions have it fetch the list of
   properties before it replies. It then confirms the property, the dates, the dimensions and the row count with you.
3. It calls Call Search Console Tool with a JSON string. request_type is website_list or custom_insights.
4. The lower half starts at Tool calling. Set fields - Consruct API CALL parses the string into fields, and Switch
   routes on request_type.
5. custom_insights goes to Search Console - Get Custom Insights, a POST to the property's searchAnalytics/query
   endpoint with the start and end date, the dimensions and the row limit. website_list goes to
   ## Search Console - Get List of Properties.
6. A Set node and an Aggregate node fold the rows into one field, response, which the tool returns to the agent.
7. The agent writes a Markdown table, and Respond to Webhook sends the agent's item back as the webhook's reply.

Change the look

• The agent's behaviour: its system message in AI Agent, which sets the greeting, the defaults of 3 months and 100 rows
  and the table format.
• The model: gpt-4o in OpenAI Chat Model, with a limit of 16000 output tokens.
• How much history the model sees: Postgres Chat Memory keeps its default window of 5 exchanges, and the table name is
  set there.
• The query: the body of Search Console - Get Custom Insights. It sends dataState as all.
• The webhook: its path and its authentication in Webhook - ChatInput.

Good to know

• Models: OpenAI's deprecations page, read on 12 November 2026, gives 14 January 2027 as the shutdown of gpt-4o, with
  gpt-6-luna as the replacement. Change the model in OpenAI Chat Model before then. The sticky notes still name
  gpt-4o and gpt-4o-mini.
• Open endpoint: the webhook is protected only by the Basic Auth you add. Without it anyone with the address can use your
  OpenAI key and read your properties.
• The query always starts at row 0, so the agent cannot page through results. Ask for a larger row limit instead. Google's
  API allows 1 to 25,000 rows a request and 1,000 by default, and dataState of all includes fresh data.
• The agent's prompt lists the dimensions page, query, searchAppearance, device and country. The API also takes date and
  hour.
• A session id is the only key to a chat's history, so use ids that are hard to guess.
• Your data: the figures Search Console returns, your questions and the chat history go to OpenAI. The history is also kept
  in your Postgres table.
• The webhook's trigger only answers requests, so a chat window of your own must call it.