News posts summarized by AI each week and saved to NocoDB

News posts summarized by AI each week and saved to NocoDB

An n8n workflow for a news page that has no RSS feed. Once a week it fetches the page, collects the link and the date of every post, and keeps the posts from the last 7 days. It fetches each of those posts, and an OpenAI model writes a short summary and three keywords for it. Each post lands as a row in a NocoDB table.

The file is a pattern, not a ready-made feed. You give it the address of your own news page and four CSS selectors, so it knows where the links, dates, titles and texts sit.

What is inside

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

What we changed from the original copy

Mindket repaired and cleaned the copy:

• The original copy was written for one company's news page. That page answers 404 today (read on 12 November 2026), so its address and its four CSS selectors are now the placeholders <YOUR_NEWS_PAGE_URL>, <YOUR_LINK_CSS_SELECTOR>, <YOUR_DATE_CSS_SELECTOR>, <YOUR_TITLE_CSS_SELECTOR> and <YOUR_TEXT_CSS_SELECTOR>. The source name written to NocoDB is now <YOUR_NEWS_SOURCE_NAME>.
• The date filter went back 70 days, although the node's name says 7 days and the schedule runs weekly. Mindket's copy goes back 7 days.
• Both OpenAI nodes named gpt-4-1106-preview, which OpenAI shut down on 23 October 2026. OpenAI's deprecations page, read on 12 November 2026, names gpt-5.6-sol as the replacement, so Mindket's copy uses it. The nodes are n8n's older OpenAI node, which n8n hides from its node list but still runs. They send no temperature or token limit, so nothing else needed to change.
• The NocoDB base and table ids were the previous owner's. They are now <YOUR_NOCODB_BASE_ID> and <YOUR_NOCODB_TABLE_ID>, and the node's API Version is set to v0.200.0 Onwards, the newest of the three it offers.
• Six sticky notes were reworded, and the links and contact details in them are gone.
• No credential is inside.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• A news page that lists its posts with a link and a date for each, and that you are allowed to read with a program. Read its terms first.
• A NocoDB table with six columns: News_Source, Title, Date, Link, Summary and Keywords, and a NocoDB API token.
• An OpenAI API key.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. In Retrieve the web page for further processsing, replace <YOUR_NEWS_PAGE_URL> with the address of your news page.
3. Open the news page in a browser, right click a post link, choose Inspect and copy a CSS selector that matches every post link. Put it in Extract the HTML with the right css class. Do the same for the dates in Extract date, with one date for each post and in the same order as the links.
4. Open one post, and copy a selector for its title and one for its text. Put them in Extract individual posts.
5. If the links on your page are relative, such as /news/post-1, put the site's address in front in the address of HTTP Request1.
6. Add a NocoDB credential with your API token. In NocoDB news database, pick your base and table in place of the two placeholders, and replace <YOUR_NEWS_SOURCE_NAME> with the name of the site.
7. Add an OpenAI credential and select it in Summary and Keywords.
8. Run the workflow once by hand and check the rows in NocoDB. Then publish it.

How a run goes

1. Schedule Trigger each week starts a run every week on Wednesday at 4:32, in the time zone of your n8n.
2. Retrieve the web page for further processsing fetches your page as text.
3. Extract the HTML with the right css class picks the post links, and Create single link items turns the list into one item for each link, in a field named Link.
4. Extract date picks the dates, and Create single date items makes one item for each date, in a field named Date.
5. Merge date & links joins the two lists by position, and Select posts of last 7 days keeps the items whose Date falls in the last 7 days. A date that JavaScript cannot read drops the item.
6. HTTP Request1 fetches each kept post, and Extract individual posts picks its title and its content. Merge Content with Date & Link adds the date and the link back.
7. Summary asks OpenAI for a summary of under 70 words. Keywords asks for the three most important technical keywords. Rename Summary and Rename keywords keep one field each, and Merge joins them.
8. Merge ChatGPT output with Date & Link joins them with the title, date and link, and NocoDB news database creates one row for each post.

Change the look

• How far back it looks: the number in the code of Select posts of last 7 days. Keep it equal to the schedule.
• The day and time: the options of Schedule Trigger each week.
• What the model writes: the prompts in Summary and Keywords, such as a different length or more keywords.
• The columns: the six field values in NocoDB news database. Any other output works, such as an email or a spreadsheet.

Good to know

• Nothing runs until you fill in the address and the selectors. Each site is built differently, so expect to adjust them.
• It reads one list page and one page for each new post, once a week. Check the site's terms and its robots rules before you use it on a site you do not own.
• Each post costs two OpenAI calls, one for the summary and one for the keywords.
• The text of every kept post goes to OpenAI, and the results go to your NocoDB.
• If your NocoDB is older than v0.200.0, set API Version in NocoDB news database back to the older choice.
• The JSON holds no credential.