Scrape Any Website and Feed It to GPT Using n8n (My Real Workflow)

This one’s been sitting in my backlog for a while now, but I finally got around to putting it together the way I actually use it. If you’ve ever wanted to scrape content from a website and then feed that into GPT for analysis, rewriting, or summarizing, this post will walk you through the full thing.

The exact flow I use, with real techniques, different options depending on your setup, and a downloadable JSON file.

Why I Use This Workflow (and When You Might Too)

I first built this setup to analyze one of my own blog posts. I wanted to:

• Check structure and SEO automatically
• Feed the content to GPT for summary + improvement suggestions
• Log it somewhere (Notion, Slack, etc.)

But this works for much more:

• Summarizing competitor blog posts
• Pulling product descriptions
• Extracting pricing pages for internal research
• Auto-generating summaries or Q&As from docs and help pages

Basically, if the content is on the web and not behind a login, you can grab it 😉.

What the Workflow Does

Here’s what we’re going to build:

1. Get the latest blog post URL (from an RSS feed or direct input)
2. Fetch the full HTML
3. Extract clean text using Cheerio
4. Send it to GPT (with a tailored prompt)
5. Output a markdown SEO/structure summary
6. Push it to Slack or save it in Notion/Docs (your choice)

The Scraping Techniques (Pick What Works for You)

Depending on the site you’re working with, you’ve got a few options:

1. Basic: HTML + Regex Cleanup

const html = $json["body"];
const clean = html.replace(/<[^>]*>/g, '').replace(/\s{2,}/g, ' ').trim();
return [{ json: { content: clean } }];

• Works for clean, minimal pages
• Can break badly if the page has lots of nested tags or JS

2. Accurate: Use cheerio (My Pick)

const cheerio = require('cheerio');
const $ = cheerio.load($json['body']);
const text = $('article').text();
return [{ json: { content: text.trim() } }];

• Extracts structured content by tag (like article, .main, etc.)
• Lets you target specific elements (like h1, .post-body, etc.)

If you’re self-hosting n8n on Railway (like me), you can install cheerio via npm and use it in Function nodes. I wrote about that setup here.

3. For JavaScript-Heavy Sites

Some pages render content dynamically with JS, so a simple HTTP request won’t cut it.

Options:

• Use an external Puppeteer script
• Use browser automation like Playwright (triggered via webhook)
• Or just use a headless browser scraping API (ScraperAPI, Browserless, etc.)

I usually avoid this unless absolutely necessary.

Feeding It to GPT (Prompt + Config)

Once we have the cleaned content, the next step is to get GPT to do something smart with it.

Here’s a prompt I use to generate a markdown-style SEO report:

You are an expert blog editor and SEO consultant. Here is a blog post:

{{content}}

Please analyze the following:
1. Summary (3 lines max)
2. On-page SEO issues
3. Internal/external links
4. Suggested keywords
5. Structural problems
6. Suggested improvements

Respond in clean markdown format.

You can modify this to:

• Rephrase content
• Generate tweet threads
• Extract action items or product features
• Build FAQs from the page

Output Options (Pick One or Chain Them)

I’ve sent the GPT response to:

• Slack → for real-time review
• Notion → for logging SEO audits
• Google Docs → for sharing with a client

Just add a Notion or Google Docs node at the end and pass {{$json["text"]}} as the content.

Want the Workflow?

Download the n8n JSON file here → includes:

• RSS feed reader
• HTML fetcher
• cheerio text extractor
• GPT-4 node
• Slack message output

You can plug in Notion instead if that’s your style.

Stuff I’ve Learned Doing This

Here are a few things worth noting that don’t always get mentioned:

• Don’t scrape aggressively — many sites have rate limits or bot protection
• Check your selectors — some blogs don’t use <article>, so you might need .post, .content or .entry
• Always clean the text before feeding to GPT — it saves tokens and avoids noisy input
• GPT does better when you’re specific — don’t just say “analyze this,” say “find SEO issues”
• Add timestamps to logs so you can compare improvements over time