Machine Learning use cases for Technical SEO Automation Brighton SEO Patrick Stox Ahrefs
Download as PPTX, PDF3 likes2,748 views
The document discusses various machine learning use cases specifically tailored for technical SEO. It highlights how machine learning can enhance SEO strategies and improve website performance. The content is presented in a slide format, making it accessible for a presentation or webinar.
Machine Learning use cases for Technical SEO Automation Brighton SEO Patrick Stox Ahrefs
- 1. Machine Learning Use Cases For Technical SEO Slideshare.Net/ahrefs @patrickstox Patrick Stox Ahrefs
Editor's Notes
- #4: I’m not building custom models, just using things that already exist.
- #6: A lot of these may be able to be done better / faster with traditional means at least for now.
- #8: Large companies have to make quarterly or yearly projections of increased traffic/revenue. It's all about selling SEO to get more resources or determine allocation of resources.
- #9: Competitor forecasting can only be done with 3rd party data. I’d never seen this done but I thought it was a cool idea so I made it. Comparing against competitors is a powerful sales tool. 15 total scripts. Traffic, Traffic value, competitor traffic, competitor traffic value, competitor page traffic, Making performance predictions during a core update
- #11: Google already rewrites them. Why wouldn’t you just grab that list?
- #12: Google already rewrites them. Why wouldn’t you just grab that list?
- #13: Google already rewrites them. Why wouldn’t you just grab that list?
- #15: No one likes to do redirects
- #16: Why?
- #17: Not accurate if URL formats changed.
- #18: Slower but more accurate. This is the first time I’m releasing this publicly and there’s nothing out there like it. I made it really easy to use. This takes the data from the best by links report in Ahrefs sorted to 404 and data from the top pages report. It crawls current pages and archive.org for the broken pages and matches them. Weaknesses of this method currently are it relies on archive.org which may not have the content and the crawling takes a while. Could also be used to fix broken links. There's a version of the similarity matching I used that uses BERT now so it should theoretically be a little more accurate
- #19: content hashes are easier when you have access to both pages, so it doesn’t really make sense for duplicates/canonicals
- #20: Maybe with the right dataset.
- #21: I was curious which titles Google had rewritten the most.
- #23: maybe if it handles multiple intents, but I don’t like INT models in general, every search is an intent and you can get more specific intents with clustering
- #26: Could add SERP similarity to the keyword clustering for better results. This can also be used for disambiguation.
- #28: I’ve always wanted to build this. You could just output it as a sitemap and be done with it.
- #29: Rankings, traffic, Page changes, Link changes. Alerts are probably just as good in most cases.
- #31: if you can identify it, so can search engines
- #32: Questionable accuracy
- #35: Early hints, A/B testing, edit header responses
- #36: Find error > write rule to fix error. Change noindex to index, nofollow to follow, add titles/descriptions you generated, etc. You could fix a lot of the issues needed for faster core web vitals Disclaimer that you might want to fix things in your own systems
- #37: Match redirect > write redirect Flatten redirect chains Disclaimer that you might want to fix things in your own systems