This post was sponsored by Ahrefs. The opinions expressed in this article are the sponsor’s own.
We’ve worked on over 50 AI research studies at Ahrefs.
Nearly every one was built to test a claim someone in the industry (or Google) was repeating, with little evidence to back it up.
Below is a culmination of the most interesting things we’ve learned, and the myths we’ve debunked, from recent AI search studies, covering 15 million data points.
1. AI Visibility: “Best-Of” Listicles Boost Your Competitor’s AI Visibility More Than Your Own
Myth busted: “Publish ‘best’ list featuring your brand and AI will recommend you.”
There’s nothing ChatGPT loves more than a “Best” recommendation list.
Across 750 ChatGPT prompts, Glen Allsopp found “best X” blogs (e.g. “The 10 best SEO tools”) made up 43.8% of all page types cited.
Unsurprisingly, many B2B companies now write their own lists and rank themselves first—but citation isn’t recommendation.
The Data
My colleague Mateusz Makosiewicz ran a controlled experiment, publishing 34 self-promotional lists on five domains and tracking 9,886 answers across ChatGPT, Gemini, Perplexity, and Copilot.
Mateusz confirmed what SEOs like Lily Ray had suspected: AI will use your article as the source of its information, but won’t actually mention your brand in its answer, and may even boost your competitor.
Here’s an example of that in the wild.
Mateusz wrote a list promoting our conference: Ahrefs Evolve. In 43% of answers, AI recommended a competitor’s event instead.
Publishing a “best” post with your name on top isn’t enough to show up in AI.
What To Do Instead
Outreach, review building, and word-of-mouth engineering (e.g. influencer campaigns, or “touring” a topic you want your brand to become synonymous with) can get your brand named across dozens of independent and, hopefully, authoritative sources. This is what AI will treat as credible long-term. We can’t say the same for a single self-authored “best of” list.
2. llms.txt: Nothing Reads Your llms.txt File. 97% Get Zero Fetches.
Myth busted: “You need an llms.txt file for AI visibility.”
In May 2026 Google gave us contradicting advice: we should be auditing our llms.txt files, even though they’re not required for AI visibility.
We decided to check the data ourselves.
The Data
We studied the server logs and bot traffic of 137,000 sites using Ahrefs Web Analytics and Bot Analytics.
Overall, 28% had published an llms.txt file, but a staggering 97% of those files were never read. Not by bots, and not by humans.
Of those that were read, 77% weren’t AI bots at all.
Most were SEO audit tools, GEO platforms, and tools built specifically to study llms.txt adoption.
It’s a tail-eating snake effect: SEOs write llms.txt files that only get read by the tools built to study them.
What To Do Instead
Skip the llms.txt file.
Put that effort into making your existing content easy to crawl, parse, and cite.
3. Schema: Adding Schema Markup Isn’t A Shortcut To Citation In AI Search
Myth busted: “Adding schema markup to a page will instantly boosts your AI citations.”
There’s been a trend of “GEO experts” attributing quick AI visibility wins to schema optimization.
The Data
We put that to the test, tracking 1,885 pages that added JSON-LD schema, matched against 4,000 control pages, measuring citation change before and after.
After 30 days, we saw no meaningful uplift on Google AI Mode or ChatGPT.
AI Overviews saw a small, real decline of 4.6%, but we couldn’t clearly pin that on schema, since both treated and control pages were already declining beforehand.
Databases like Wikidata and Google’s Knowledge Graph have absorbed schema for years.
Whether that structure actually shapes how AI models understand entities is still an open question.
But if schema does feed AI understanding at all, it’s likely to be through the Knowledge Graph; slow and indirect, not a shortcut tactic.
What To Do Instead
Use schema markup to build long-term entity associations, marking up Organization and Person data with sameAs links to Wikipedia, Wikidata, and Crunchbase, since a well-defined entity is what feeds Google’s Knowledge Graph, and a clearer Knowledge Graph entry is more likely to shape how AI models understand you.
4. SERP Positions: Ranking In Classic Search Is Still How You Get Cited By ChatGPT
Myth busted: “You don’t need to rank in classic search to appear in AI.”
By now, most of us know that AI platforms source pages from traditional search, as part of retrieval augmented generation.
But how much of a role does search ranking play in citation?
We analyzed 1.4 million ChatGPT prompts to find out.
The Data
We categorized every retrieved URL by ref_type and found that 88.46% of all citations came from the general search index; aka good old fashioned search rankings.
Specialized channels, like Reddit and YouTube, get pulled in at scale but rarely cited.
Ranking well in classic search is still doing the heaviest lifting in ChatGPT’s citation decisions.
What To Do Instead
Invest in the classic SEO work that earns ranking:
- Targeting the right keywords.
- Matching search intent.
- Fixing the technical issues that hold pages back.
5. Position 1: Ranking On Page 1 Doesn’t Guarantee AI Visibility
Myth busted: “If you rank on page one SERPs, you’re more likely to be cited in AI.”
It’s easy to assume a page one ranking gives you an outsized chance of AI visibility.
But our data suggests that’s not always the case.
The Data
We analyzed 863K SERPs and 4M citations using Brand Radar, and found just ~38% of URLs cited in AI Overviews also ranked in the top 10 for the same query, down from ~76% a year earlier.
The rest ranked further down, or didn’t rank in the top 100 at all.
Google has put more emphasis on query fan-out, pulling citations from related searches rather than the user’s original prompt. So, rankings do matter, but not for the keywords you’d expect.
What To Do Instead
Build out topical coverage across a cluster of related queries, not just your one target keyword, so you’re visible in the fan-out searches too.
6. Citations: AI Reshuffles Its Sources, But Not Its Opinion
Myth busted: “AI answers are too volatile to track reliably.”
Some SEOs question whether AI tracking is worthwhile, given how unpredictable AI answers are.
It’s true, AI answers shift constantly.
The Data
Using Brand Radar, we tracked 43,000 keywords over a month, checking each one’s AI Overview 16+ times. Wording changed 70% of the time, brands shifted 46%, and cited sources swapped 45.5%.
But the substance underneath barely shifts: when we compared meaning rather than wording, AI Overviews scored 0.95 out of 1.
Google has settled on what the answer is, it just keeps re-sourcing and rewording it.
If you win the topic, not the exact phrasing, you’ll show up more consistently.
Tracking is what lets you see that stability.
What To Do Instead
Track your AI visibility by topic and sentiment over time, since the surface noise can sometimes hide a more stable signal underneath.
7. Traffic: AI Hasn’t Killed Search. Google Sends 190x More Traffic To Sites.
Myth busted: “Google is dead. AI search has replaced it.”
Every year, SEOs enjoy writing Google’s obituary. It’s a cherished pastime that has taken off again with the growth of AI search.
But Google always lives to tell the tale and, going by our data, it will do for many more years to come.
The Data
Patrick Stox found that ChatGPT gets only ~12% of Google’s search volume for the kinds of queries people traditionally typed into Google, once you strip out non-search usage like coding help or translation.
Studying 76,000 sites hooked up to Ahrefs Web Analytics, we also discovered that Google sends 190x more traffic than ChatGPT, making up nearly 40% of site traffic, versus ChatGPT’s 0.21%.
AI hasn’t killed off Google, but it’s certainly fighting for a slice of it.
What To Do Instead
Keep prioritizing Google traffic in your strategy.
It’s still your largest channel by a wide margin, even as you build out AI visibility alongside it.
8. Clicks: AI Overviews Cost You More Than Half Your Clicks (58%)
Myth busted: “Appearing in an AI Overview lifts your clicks.”
In May 2024, Google Search head Liz Reid published a blog post claiming that “the links included in AI Overviews get more clicks than if the page had appeared as a traditional web listing for that query.”
That didn’t align with what the industry was experiencing, so Ryan Law decided to test Google’s claims.
The Data
He studied the click-through rate (CTR) impact of AI Overviews back in early 2025, and found that AI Overviews reduced clicks to top-ranking content by 34.5%.
Then, in 2026, when he noticed the AI Overview impact seemed to be getting worse, he repeated his study and found that the presence of an AI Overview cut position-1 CTR by 58%.
This finding was sadly corroborated by Seer Interactive (CTR down >65.2%), Kevin Indig (>50%), and Authoritas (47.5%).
What To Do Instead
Diversify traffic sources and track your true click loss directly.
Don’t just take Google’s claims at face value.
9. Link Building: Backlinks Matter Less Than Brand Mentions for AI Visibility
Myth busted: “Backlinks and domain authority are a big driver of AI visibility.”
Old habits die hard, and a lot of SEOs assume that backlinks and Domain Rating (DR) are the factors that matter most for AI visibility, since they work so well for search.
The Data
We studied 75,000 brands, correlating a stack of metrics against brand mentions in ChatGPT, AI Mode, and AI Overviews to understand what drives AI visibility.
Surprisingly, DR showed a very weak correlation with brand mentions in AI answers (0.266–0.326), and the number of backlinks barely registered (0.191–0.244).
What did correlate strongly was YouTube mentions (~0.737), ahead of every other factor, with branded web mentions close behind (0.656–0.709).
Links built your authority for Google. For AI, it’s whether people are actually talking about you on YouTube and across the web that counts.
What To Do Instead
Prioritize getting your brand mentioned in press coverage and forums, and build a presence on YouTube through your own channel, creator partnerships, or sponsorships.
15 Million Data Points Later: Fundamentals Beat Hacks
By studying ~15 million data points, we’ve learned a ton about AI search, including what works, what Google says works, and what is just LinkedIn hype.
At this point, you may be thinking that AI search sounds like a lot of work for very little payoff.
But the tactics that fail are mostly the shortcuts, like the quick schema markup tag-ons or the self-crowned “best” lists.
What works well are slower and more familiar strategies; ranking, earning real mentions, building content out topically, and creating a brand people talk about.
AI search isn’t unknowable, but it’s definitely allergic to hacks.
Image Credits
Featured Image: Image by Ahrefs. Used with permission.
In-Post Images: Images by Ahrefs. Used with permission.
