AI Broke SEO: A 50M Keyword Analysis Reveals the New Rules for Google and LLMs

⁠NinjaAI.com ⁠The world of search engine optimization is in a state of constant, rapid evolution. The rise of AI Overviews and Large Language Models (LLMs) like ChatGPT has fundamentally altered the landscape, creating a two-front war where the old rules of SEO no longer guarantee victory. Optimizing for Google's traditional search and optimizing for an LLM's knowledge base are two distinct challenges that require different strategies.This article distills the key takeaways from a recent data-driven keynote by Manick Bhan of Search Atlas, which analyzed a massive dataset of 50 million keywords. The following points are some of the most surprising, impactful, and actionable findings from the research, offering much-needed clarity in a complex new era of search.--------------------------------------------------------------------------------1. Some 'Dead' SEO Tactics Are Making a Surprising ComebackResearch based on an analysis of 15,327 websites has revealed that some supposedly "deprecated" or basic SEO fields still have a significant and measurable impact on visibility. This finding challenges long-held assumptions and proves the value of a data-first approach.The study unearthed several powerful correlations:Image Alt Text: Using image alt text, on average, improves the number of keywords a page is ranking for by a staggering 100%—it literally doubles the keyword footprint of the page.Missing H1s/H2s: If a page is missing an H1 or an H2, adding them has the biggest impact on impressions, driving an improvement of over 115%.Schema Markup: Implementing schema markup improved keyword positions by an average of 20 spots.Meta Keywords: The meta keywords tag, a field most SEOs have ignored for years, was shown to significantly improve the number of keywords a page ranks for.Canonical Links: Beyond preventing duplicate content, adding canonical links had a significant impact on impressions. This suggests, as Bhan theorizes, that canonicals may act as a direct quality signal to search engines, going far beyond simple duplicate content prevention.This underscores the importance of being a "scientist" in the field of SEO—testing what actually works rather than relying solely on old assumptions. As Bhan noted in his presentation:Look I don't make the rules i'm just looking to see what works i'm a scientist if it works on a Tuesday for me to dance outside in the rain and I get page one rankings I'm going to do it.--------------------------------------------------------------------------------2. The Authority Metric You Track is Probably WrongA fundamental conflict in modern SEO is that "authority" is measured in fundamentally different ways by Google versus LLMs like ChatGPT. Optimizing for one requires a different focus than optimizing for the other.For ranking on Google, the analysis showed that topical relevance is the most dominant factor, explaining 30% of rankings alone. The next most important signal is a traffic-based metric called "Domain Power," which has a much higher correlation with rankings than classic metrics like Domain Authority (DA) or Ahrefs' Domain Rating (DR). The reason for this shift is that Google now uses "other site metrics from Chrome to validate the value of websites and the links that they're providing." The study found a massive "+ or - 50 point gap" between DR and Domain Power, revealing that many sites with high DR scores have zero actual traffic.In contrast, for achieving visibility within ChatGPT, the classic PageRank-style metrics are the most important signals. Metrics like DR, referring domains, and trust flow hold the most weight for being sourced by the LLM.The strategic takeaway is clear: to win in the new search landscape, you must understand which engine you are optimizing for. Using the correct corresponding authority metrics is essential for an effective strategy.