An SEO agent on n8n: how to automate monitoring of three websites

⚡ The gist in 5 seconds - The gist: Anastasia Nikulina, a head of content, built an SEO agent on n8n that automatically collects data for three websites, stores history, and delivers short reports to Telegram. - Availability: A detailed case study with architecture and examples has been published on Habr. - Limitation: The agent doesn't replace an SEO specialist — it's a tool for analytics, monitoring, and hypothesis testing. A language model is connected selectively, so expensive requests aren't run on every page. ### 🔍 What was found Anastasia Nikulina, who heads the content department at an IT company, faced a sharp drop in search traffic in 2025: over several months, the metrics nearly rolled back to 2024 levels, with a decline of almost 40%. Contractors suggested expanding the semantic core and buying links, but that didn't solve the business problem. So the author built her own SEO agent on n8n. The system pulls data from Screaming Frog, Yandex Metrica, Webmaster, and page content, normalizes it, matches URLs, and removes technical noise. History is stored in Google Sheets, which record the current state, changes, and manual decisions. Then rules come into play: thresholds, signals, and a scoring system. Only for classifying new pages and deep analysis of a few candidates is a language model connected. The output is three short reports in Telegram: a technical one (crawl errors, redirects, missing meta tags), a summary of the site's overall temperature (visits, traffic, rankings), and a content breakdown (which SEO materials are growing or declining, which hypotheses to test). Before the crisis, the flagship product received up to 30 leads per month from organic search. ### 💡 Why it matters The main value of the case study lies in smart resource management. The author emphasizes that powerful neural networks cost money, so she designed the architecture so the model only works where text analysis and hypotheses are needed, while all the routine work — normalization, filtering, URL matching — is handled by rules and plain code. It's a practical example of how to stop drowning in manually compiled reports and get a ready-made list of what truly requires the editorial team's attention, delivered straight to a messenger. This workflow can be adapted to your own projects. ### 🧩 Context Until mid-2025, SEO work was going great: blogs were growing, and so was search traffic. In the summer, everything changed abruptly: an industry crisis, declining software demand, seasonality, and the rise of AI-generated answers, which took a share of informational queries, all hit at once. Contractors couldn't explain the causes, and familiar methods like rewriting meta tags stopped working. Anastasia dove into SEO herself, learned about search intent and user behavior, and ultimately created a tool that lets her see the full picture weekly — from technical errors to content hypotheses.

🔗 Read on habr.com