Weekly Google Analytics compared by AI and saved to Baserow

Weekly Google Analytics compared by AI and saved to Baserow

An n8n workflow that reads your Google Analytics every week and asks an AI model what changed. It pulls three reports: page engagement, Google search results and views by country. For each report it takes the last 7 days and the 7 days before. Three AI answers, each a markdown table and five suggestions, go into one Baserow row.

What is inside

• workflow.json: the workflow as n8n exports it, 18 nodes beside 4 sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

Mindket repaired the copy so it runs today:

• The node for the prior week had a fixed end date, 23 October 2024, and a start date of 14 days ago. The Google Analytics node in n8n stops with an error when the start is after the end. That made every run from early November 2024 on fail. Mindket's copy ends on the date 7 days ago, as the two other prior-week nodes do.
• Both country nodes asked for a metric with no name. The code that reads the answer expects totalUsers in that place, so Mindket's copy names it.
• The first AI request filled "2 weeks ago" with the country data of the prior week, not its page data. It now reads the page data from Parse GA data.
• The three requests named meta-llama/llama-3.1-70b-instruct:free. OpenRouter's model list, read on 12 November 2026, no longer has that free variant. Mindket's copy names meta-llama/llama-3.1-70b-instruct, which the list shows at 0.72 US dollars for each million tokens, in and out.
• The Google Analytics property id and the Baserow database, table and field ids were the previous owner's. They are now placeholders such as <YOUR_GA4_PROPERTY_ID>, <YOUR_BASEROW_DATABASE_ID> and <YOUR_BASEROW_FIELD_PAGE_VIEWS>.
• Four sticky notes were reworded, and a link to the previous owner's own blog is gone.
• No credential is inside.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A Google Analytics 4 property and its property id. The Google search report also needs a Search Console property linked to it, since Google's API says those metrics require an active link.
• A Baserow table with five fields: Name, Country Views, Page Views, Search Report and Blog, ideally long text for the three reports.
• An OpenRouter API key.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add a Google Analytics OAuth2 credential. Select it in the six Google Analytics nodes, and replace <YOUR_GA4_PROPERTY_ID> in each of them with your property id.
3. Add a Header Auth credential for OpenRouter: the name Authorization, and the value Bearer followed by a space and your API key. Select it in Send page data to A.I., Send page Search data to A.I. and Send country view data to A.I..
4. Add a Baserow credential. In Save A.I. output to Baserow, pick your database and table in place of the placeholders, then pick a field for each of the five values and put the name of your website in the Blog value.
5. Run the workflow once with the manual trigger and check the new row. Then publish it.

The five values written to Baserow are:

| Field | Value |
|---|---|
| Name | The date and time of the run |
| Blog | The text Name of your blog, which you replace |
| Page Views | The answer to the first AI request |
| Search Report | The answer to the second AI request |
| Country Views | The answer to the third AI request |

How a run goes

1. Schedule Trigger starts a run every week, on Sunday at midnight in the time zone of your n8n. When clicking ‘Test workflow’ starts the same run by hand.
2. Get Page Engagement Stats for this week reads the last 7 days of page views, active users, views for each user and event count for each page title, and Parse data from Google Analytics turns the rows into one URL-encoded text.
3. Get Page Engagement Stats for prior week reads the 7 days before, from 14 days ago to 7 days ago, and Parse GA data turns them into text.
4. The two search steps do the same with landing pages, and with the Search Console figures: average position, click through rate, clicks and impressions, beside the active users, engaged sessions, engagement rate and event count.
5. The two country steps do it by country, with active users, new users, total users, sessions, engaged sessions, engagement rate and event count.
6. Send page data to A.I., Send page Search data to A.I. and Send country view data to A.I. each post the two texts to OpenRouter with the same prompt. It asks for a markdown comparison table and five suggestions to improve SEO.
7. Save A.I. output to Baserow creates one row.

Change the look

• The prompt in each of the three requests: a different kind of advice, or a different table.
• The model: the model value in the body of each request.
• The metrics and the dimensions: the options of the six Google Analytics nodes. If you change the order of the metrics, change the matching Parse code too, since it reads them by position.
• The day and time: the options of Schedule Trigger.

Good to know

• The two periods share one day. The last 7 days run from 7 days ago to today, and the week before ends 7 days ago.
• Google Analytics 4 only. The report needs the Search Console link for the four search metrics.
• Your page paths, titles and countries go to OpenRouter and to the model it routes to. A site with many pages sends a long prompt, and the model's context window is 131,072 tokens.
• OpenRouter bills each request. Free models come and go on its list, so check the list before you choose one. google/gemma-4-31b-it:free was listed on 12 November 2026.
• The Baserow node signs in with a Baserow username and password by default. You can switch it to a database token in the node.
• The JSON holds no credential.