This submit was sponsored by Ahrefs. The opinions expressed on this article are the sponsor’s personal.
We’ve labored on over 50 AI research studies at Ahrefs.
Almost each one was constructed to check a declare somebody within the trade (or Google) was repeating, with little proof to again it up.
Beneath is a fruits of probably the most fascinating issues we’ve realized, and the myths we’ve debunked, from latest AI search research, protecting 15 million information factors.
1. AI Visibility: “Finest-Of” Listicles Enhance Your Competitor’s AI Visibility Extra Than Your Personal
Delusion busted: “Publish ‘greatest’ record that includes your model and AI will suggest you.”
There’s nothing ChatGPT loves greater than a “Finest” suggestion record.
Throughout 750 ChatGPT prompts, Glen Allsopp discovered “greatest X” blogs (e.g. “The ten greatest web optimization instruments”) made up 43.8% of all web page sorts cited.

Unsurprisingly, many B2B corporations now write their very own lists and rank themselves first—however quotation isn’t suggestion.
The Information
My colleague Mateusz Makosiewicz ran a managed experiment, publishing 34 self-promotional lists on 5 domains and monitoring 9,886 solutions throughout ChatGPT, Gemini, Perplexity, and Copilot.
Mateusz confirmed what SEOs like Lily Ray had suspected: AI will use your article because the supply of its data, however gained’t truly point out your model in its reply, and should even increase your competitor.
Right here’s an instance of that within the wild.

Mateusz wrote a listing selling our convention: Ahrefs Evolve. In 43% of solutions, AI really helpful a competitor’s occasion as an alternative.

Publishing a “greatest” submit along with your title on prime isn’t sufficient to indicate up in AI.
What To Do As an alternative
Outreach, assessment constructing, and word-of-mouth engineering (e.g. influencer campaigns, or “touring” a subject you need your model to grow to be synonymous with) can get your model named throughout dozens of impartial and, hopefully, authoritative sources. That is what AI will deal with as credible long-term. We will’t say the identical for a single self-authored “better of” record.
2. llms.txt: Nothing Reads Your llms.txt File. 97% Get Zero Fetches.
Delusion busted: “You want an llms.txt file for AI visibility.”
In Might 2026 Google gave us contradicting recommendation: we ought to be auditing our llms.txt files, despite the fact that they’re not required for AI visibility.
We determined to verify the information ourselves.
The Information
We studied the server logs and bot traffic of 137,000 sites utilizing Ahrefs Web Analytics and Bot Analytics.
Total, 28% had revealed an llms.txt file, however a staggering 97% of these recordsdata have been by no means learn. Not by bots, and never by people.

Of people who have been learn, 77% weren’t AI bots in any respect.
Most have been web optimization audit instruments, GEO platforms, and instruments constructed particularly to review llms.txt adoption.
It’s a tail-eating snake impact: SEOs write llms.txt recordsdata that solely get learn by the instruments constructed to review them.
What To Do As an alternative
Skip the llms.txt file.
Put that effort into making your current content material easy to crawl, parse, and cite.
3. Schema: Including Schema Markup Isn’t A Shortcut To Quotation In AI Search
Delusion busted: “Including schema markup to a web page will immediately boosts your AI citations.”
There’s been a pattern of “GEO specialists” attributing fast AI visibility wins to schema optimization.
The Information
We put that to the take a look at, tracking 1,885 pages that added JSON-LD schema, matched towards 4,000 management pages, measuring quotation change earlier than and after.
After 30 days, we noticed no meaningful uplift on Google AI Mode or ChatGPT.
AI Overviews noticed a small, actual decline of 4.6%, however we couldn’t clearly pin that on schema, since each handled and management pages have been already declining beforehand.

Databases like Wikidata and Google’s Data Graph have absorbed schema for years.
Whether or not that construction truly shapes how AI fashions perceive entities remains to be an open query.
But when schema does feed AI understanding in any respect, it’s prone to be by means of the Knowledge Graph; gradual and oblique, not a shortcut tactic.
What To Do As an alternative
Use schema markup to construct long-term entity associations, marking up Organization and Person information with sameAs hyperlinks to Wikipedia, Wikidata, and Crunchbase, since a well-defined entity is what feeds Google’s Data Graph, and a clearer Data Graph entry is extra prone to form how AI fashions perceive you.
4. SERP Positions: Rating In Basic Search Is Nonetheless How You Get Cited By ChatGPT
Delusion busted: “You don’t must rank in traditional search to seem in AI.”
By now, most of us know that AI platforms supply pages from conventional search, as a part of retrieval augmented generation.
However how a lot of a task does search rating play in quotation?
We analyzed 1.4 million ChatGPT prompts to seek out out.
The Information
We categorized each retrieved URL by ref_type and located that 88.46% of all citations got here from the overall search index; aka good quaint search rankings.

Specialised channels, like Reddit and YouTube, get pulled in at scale however not often cited.
Rating effectively in traditional search remains to be doing the heaviest lifting in ChatGPT’s quotation selections.
What To Do As an alternative
Spend money on the traditional web optimization work that earns rating:
- Concentrating on the correct key phrases.
- Matching search intent.
- Fixing the technical points that maintain pages again.
5. Place 1: Rating On Web page 1 Doesn’t Assure AI Visibility
Delusion busted: “For those who rank on web page one SERPs, you’re extra prone to be cited in AI.”
It’s simple to imagine a web page one rating offers you an outsized likelihood of AI visibility.
However our information means that’s not at all times the case.
The Information
We analyzed 863K SERPs and 4M citations utilizing Brand Radar, and located simply ~38% of URLs cited in AI Overviews additionally ranked within the prime 10 for a similar question, down from ~76% a yr earlier.

The remaining ranked additional down, or didn’t rank within the prime 100 in any respect.
Google has put extra emphasis on query fan-out, pulling citations from associated searches quite than the consumer’s unique immediate. So, rankings do matter, however not for the key phrases you’d anticipate.
What To Do As an alternative
Construct out topical coverage throughout a cluster of associated queries, not simply your one goal key phrase, so that you’re seen within the fan-out searches too.
6. Citations: AI Reshuffles Its Sources, However Not Its Opinion
Delusion busted: “AI solutions are too unstable to trace reliably.”
Some SEOs query whether or not AI monitoring is worth it, given how unpredictable AI solutions are.
It’s true, AI solutions shift consistently.
The Information
Utilizing Model Radar, we tracked 43,000 keywords over a month, checking every one’s AI Overview 16+ occasions. Wording modified 70% of the time, manufacturers shifted 46%, and cited sources swapped 45.5%.
However the substance beneath barely shifts: once we in contrast that means quite than wording, AI Overviews scored 0.95 out of 1.

Google has settled on what the reply is, it simply retains re-sourcing and rewording it.
For those who win the subject, not the precise phrasing, you’ll present up extra constantly.
Monitoring is what permits you to see that stability.
What To Do As an alternative
Observe your AI visibility by matter and sentiment over time, because the floor noise can typically cover a extra steady sign beneath.
7. Visitors: AI Hasn’t Killed Search. Google Sends 190x Extra Visitors To Websites.
Delusion busted: “Google is lifeless. AI search has changed it.”
Yearly, SEOs get pleasure from writing Google’s obituary. It’s a cherished pastime that has taken off once more with the expansion of AI search.
However Google at all times lives to inform the story and, going by our information, it should do for a lot of extra years to come back.
The Information
Patrick Stox discovered that ChatGPT gets only ~12% of Google’s search volume for the sorts of queries folks historically typed into Google, when you strip out non-search utilization like coding assist or translation.
Learning 76,000 websites hooked as much as Ahrefs Web Analytics, we additionally found that Google sends 190x extra visitors than ChatGPT, making up almost 40% of website visitors, versus ChatGPT’s 0.21%.

AI hasn’t killed off Google, however it’s actually preventing for a slice of it.
What To Do As an alternative
Preserve prioritizing Google visitors in your technique.
It’s nonetheless your largest channel by a large margin, at the same time as you construct out AI visibility alongside it.
8. Clicks: AI Overviews Price You Extra Than Half Your Clicks (58%)
Delusion busted: “Showing in an AI Overview lifts your clicks.”
In Might 2024, Google Search head Liz Reid published a blog post claiming that “the hyperlinks included in AI Overviews get extra clicks than if the web page had appeared as a standard internet itemizing for that question.”

That didn’t align with what the trade was experiencing, so Ryan Law determined to check Google’s claims.
The Information
He studied the click-through price (CTR) affect of AI Overviews again in early 2025, and located that AI Overviews lowered clicks to top-ranking content material by 34.5%.
Then, in 2026, when he observed the AI Overview affect gave the impression to be getting worse, he repeated his research and located that the presence of an AI Overview cut position-1 CTR by 58%.

This discovering was sadly corroborated by Seer Interactive (CTR down >65.2%), Kevin Indig (>50%), and Authoritas (47.5%).
What To Do As an alternative
Diversify visitors sources and observe your true click loss straight.
Don’t simply take Google’s claims at face worth.
9. Hyperlink Constructing: Backlinks Matter Much less Than Model Mentions for AI Visibility
Delusion busted: “Backlinks and area authority are a giant driver of AI visibility.”
Previous habits die onerous, and loads of SEOs assume that backlinks and Area Ranking (DR) are the components that matter most for AI visibility, since they work so effectively for search.
The Information
We studied 75,000 brands, correlating a stack of metrics towards model mentions in ChatGPT, AI Mode, and AI Overviews to know what drives AI visibility.
Surprisingly, DR confirmed a really weak correlation with model mentions in AI solutions (0.266–0.326), and the variety of backlinks barely registered (0.191–0.244).

What did correlate strongly was YouTube mentions (~0.737), forward of each different issue, with branded internet mentions shut behind (0.656–0.709).

Hyperlinks constructed your authority for Google. For AI, it’s whether or not individuals are truly speaking about you on YouTube and throughout the net that counts.
What To Do As an alternative
Prioritize getting your model talked about in press protection and boards, and construct a presence on YouTube by means of your personal channel, creator partnerships, or sponsorships.
15 Million Information Factors Later: Fundamentals Beat Hacks
By finding out ~15 million information factors, we’ve realized a ton about AI search, together with what works, what Google says works, and what’s simply LinkedIn hype.
At this level, you might be considering that AI search seems like loads of work for little or no payoff.
However the ways that fail are principally the shortcuts, like the short schema markup tag-ons or the self-crowned “greatest” lists.
What works effectively are slower and extra acquainted methods; rating, incomes actual mentions, constructing content material out topically, and making a model folks discuss.
AI search isn’t unknowable, however it’s undoubtedly allergic to hacks.
Picture Credit
Featured Picture: Picture by Ahrefs. Used with permission.
In-Submit Pictures: Pictures by Ahrefs. Used with permission.
#Search #Myths #Debunked #Million #Information #Factors
