Rand Fishkin's Research on Google's AI Overviews and Mobile Search Decline
In a recent study, Rand Fishkin revealed significant insights into how Google's AI overviews have impacted mobile search volumes. According to...
4 min read
Writing Team
:
May 2, 2025 4:07:15 PM
Ever wonder if AI could replace your SEO team? Well, don't hand in your resignation letter just yet. Previsible's new AI SEO Benchmark has just dropped some fascinating insight into how large language models (LLMs) stack up against human SEO professionals. Spoiler alert: the robots aren't taking our jobs today, but they're definitely eyeing the help wanted ads.
Previsible's benchmark has given us the first comprehensive look at how various LLMs perform on SEO tasks. Claude Sonnet 3.7 emerged as the clear winner with an 83% score, beating out competitors like Perplexity (82%), Gemini 2.5 (81%), and ChatGPT 4o (79%). The benchmark tested these models across various SEO disciplines and found some fascinating performance patterns.
What's particularly interesting is how these AI models compare to human SEO professionals. According to Previsible's study, their team of human SEO experts averaged an 89% score on the same test. So while Claude might be leading the AI pack, it's still playing catch-up to human expertise. This gap becomes especially pronounced when examining performance across different SEO disciplines.
When it comes to content-related SEO tasks, LLMs performed admirably with an average score of 85%. This isn't surprising given that generative content is what these models were primarily designed for. If you're looking to use AI for content briefs, keyword research, or generating outlines, you're generally on solid ground.
However, the picture changes dramatically when we look at technical SEO. LLMs averaged just 79% on technical tasks, which might seem decent until you consider what a 21% error rate could mean for your website. Would you trust a tool that gets one in five technical SEO recommendations wrong? And that's not even the worst of it.
If Claude has a weak spot, it's definitely e-commerce SEO. The average score across all LLMs for e-commerce-specific tasks was a dismal 63%. This is particularly concerning given how mission-critical SEO is for online stores. One misplaced canonical tag or improperly structured product schema could cost thousands in lost revenue.
This performance gap highlights something important: AI might be helpful for certain SEO tasks, but it's nowhere near ready to handle specialized disciplines like e-commerce optimization. The complexity of product hierarchies, inventory management, and conversion optimization still requires human expertise and judgment.
One of the most surprising findings from Previsible's benchmark was how certain AI features affected performance. Adding an "SEO expert" persona to prompts improved results by an average of 2.8% - a small but meaningful boost that suggests properly setting context helps these models access more relevant knowledge.
However, two heavily hyped features actually hurt performance. Enabling web search capabilities resulted in a 3.2% worse performance on average. Even more shocking, the "deep research" or extended thinking modes that many AI companies are promoting led to a 5.7% performance drop. These features, despite their marketing, appear to introduce more opportunities for error rather than enhancing accuracy.
This counterintuitive finding suggests that LLMs perform best when they stick to their core training rather than attempting to introduce external information or complex reasoning chains. It also indicates that many of the flashy features being promoted by AI companies may not deliver the practical benefits they promise for SEO tasks.
David Bell, who shared these findings, points to a critical limitation of current AI technology: its probabilistic nature. Unlike humans who can verify information or admit uncertainty, LLMs are simply predicting what text should come next based on patterns in their training data. This fundamental limitation means they can sound convincing while being completely wrong.
For high-stakes SEO decisions like technical implementations or business strategy, this probabilistic approach introduces unacceptable risk. As Bell notes, you wouldn't use an alarm clock that works 83% of the time when catching a flight – you need reliability approaching 100%. The same applies to SEO tools that could potentially tank your website's visibility.
Where AI can be useful, according to Previsible's analysis, is for lower-stakes tasks where human oversight provides a safety net. Content briefs, internal linking opportunities, and other areas where AI suggestions can be verified before implementation represent the sweet spot for these tools. It's also becoming increasingly important to track your visibility in AI outputs, with tools like Peec.ai emerging to help brands understand how they appear in LLM results.
Perhaps the most telling insight from the benchmark is why human SEOs still outperform AI. Previsible notes that their team members specialize in specific disciplines like content, technical SEO, or e-commerce. No individual team member is expected to excel across all areas – instead, they assemble teams with complementary expertise to handle complex projects.
This specialization, combined with the ability to collaborate and build consensus, represents the human advantage in SEO. While Claude might score 83% across all categories, a team of specialists who each excel in their respective areas can deliver consistent excellence across the board. Add in the critical thinking, business context, and stakeholder management that SEO requires, and it's clear why human expertise remains essential.
The Previsible AI SEO Benchmark offers a fascinating snapshot of where AI currently stands in the world of SEO. Claude Sonnet 3.7 may lead the pack, but even the best AI falls short of human expertise – especially in specialized areas like technical implementation and e-commerce optimization.
Rather than viewing these results as a competition between humans and AI, the smarter approach is to look for strategic partnerships. Let AI handle the tasks it excels at, like content ideation and data organization, while human experts focus on strategy, technical implementation, and the creative problem-solving that truly drives SEO success.
Need help navigating the complex intersection of AI and SEO? At Hire a Writer, our team combines cutting-edge AI knowledge with deep human expertise to deliver SEO content that actually performs. Contact us today to discover how our experienced SEO writers can help your content rank higher while staying ahead of the AI curve.
In a recent study, Rand Fishkin revealed significant insights into how Google's AI overviews have impacted mobile search volumes. According to...
Search engines have become an integral part of our online experience, serving as gateways to information, products, and services. While Google...
Two new studies from leading marketing firms have confirmed what many SEO professionals have feared: Google's AI Overviews are taking a significant...