Essays from an index page summarized by OpenAI

Essays from an index page summarized by OpenAI

An n8n workflow that reads a list of essays on a website and summarizes the newest three. It fetches the index page, picks out the essay links, fetches the first three essays, and has an OpenAI model summarize each one. The run ends with one item for each essay: its title, its summary and its address.

As the file stands, it reads the essay index at paulgraham.com. Point it at another site by changing the address, one selector and the base address of the links.

What is inside

• workflow.json: the workflow as n8n exports it, 13 nodes beside 2 sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

Mindket made two changes:

• The chat model named gpt-4o-mini. OpenAI's deprecations page, read on 12 November 2026, says it shuts down on 14 January 2027 and names gpt-6-luna as the replacement. Mindket's copy names gpt-6-luna. The node sets no temperature, so nothing else needed to change.
• The file had no name. Mindket gave it one. Its two sticky notes now say Fetch the latest essays and Summarize them with OpenAI.
• No credential is inside.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• Nothing else. The site it reads needs no account.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in OpenAI Chat Model1.
3. Press Execute workflow on the manual trigger.
4. Open Clean up to read the three results.

How a run goes

1. When clicking "Execute Workflow" starts the run by hand.
2. Fetch essay list fetches http://www.paulgraham.com/articles.html, which the site answers with a redirect to its secure address.
3. Extract essay names picks the link of every anchor inside a table that sits inside another table, as the selector table table a, and keeps its href. On the live page, the first links it picks are essays.
4. Split out into items makes one item for each link, and Limit to first 3 keeps the first three.
5. Fetch essay texts fetches each essay from http://www.paulgraham.com/ plus the link.
6. Extract title takes the text of the page's title tag. Summarization Chain reads the whole page as a document. Default Data Loader passes every item's JSON, tags included, to Recursive Character Text Splitter, which cuts it into pieces of 1000 characters with no overlap.
7. The chain summarizes each piece with OpenAI Chat Model1 and then joins the partial summaries into one. It uses n8n's Map Reduce method and its default prompts.
8. Merge joins each title with its summary by position, and Clean up keeps three fields: title, summary and url.

Change the look

• How many essays: the number in Limit to first 3.
• Another site: the address in Fetch essay list, the selector in Extract essay names, and the base address in Fetch essay texts and in Clean up.
• The style of the summary: the options of Summarization Chain, such as its prompts.
• The model: OpenAI Chat Model1.
• Where the summaries go: add a node after Clean up, such as an email or a spreadsheet row.

Good to know

• It reads three essay pages and one index page for each run, and nothing is stored. The results sit in the run's output.
• The chain summarizes the raw page, HTML tags included, in pieces of 1000 characters. A page of 30,000 characters makes about 30 pieces, so one essay can cost about 30 model calls plus one to join them.
• Essays are the site owner's writing. Keep the summaries for your own reading unless you have the right to share them.
• The selector and the link handling fit this one site's layout. A site whose links are absolute, or whose index has another layout, needs its own selector and its own base address.
• The text of each essay goes to OpenAI.
• The JSON holds no credential.