Hi everyone, this is Neo.

There’s a funny thing about the AI search conversation right now: everyone’s talking about it, but most conclusions are guesses.

“You need an llms.txt file!” “Add schema and AI will cite you!” “Write a ‘Best of’ list and AI will recommend you!” — you’ve heard all of this in groups, articles, and courses. But how much of it has actually been tested with data?

Ahrefs did something bold: they pooled the findings from 50+ AI search studies — 15 million data points — and systematically tested the nine most widely repeated myths in the industry. My first reaction after reading it: several conclusions directly contradict what the “mainstream advice” has been telling us.

This post breaks down all nine myths with the underlying data, then translates each one into practical moves for independent site sellers.

Myth 1: “Write a ‘Best of’ list and AI will recommend you”

Reality: AI will use your article as its source — then never mention your brand, and may even recommend your competitor.

This is the favorite play of many B2B companies: publish “The 10 Best SEO Tools,” rank yourself #1.

The data: across 750 ChatGPT prompts, “Best X” lists made up 43.8% of all cited page types — AI does love citing lists. But citation isn’t recommendation.

Ahrefs’ Mateusz Makosiewicz ran a controlled experiment: 34 self-promotional lists across five domains, tracking 9,886 answers on ChatGPT, Gemini, Perplexity, and Copilot. Results:

• AI used their articles as information sources
• But in 43% of answers, AI recommended a competitor instead of them

The most brutal example: Mateusz published a list promoting Ahrefs’ own conference, Ahrefs Evolve — and in 43% of answers, AI recommended a competitor’s event.

What to do: Don’t expect a self-authored “best of” post to buy you AI visibility. The real path is getting your brand named across dozens of independent, authoritative sources — outreach, review building, word-of-mouth, influencer collaborations. AI trusts “others say you’re good,” not “you say you’re good.”

Myth 2: “Without llms.txt, AI can’t see you”

Reality: 97% of llms.txt files are never read by anything, human or bot.

In May 2026, Google contradicted itself within a week: one team said llms.txt isn’t needed for AI visibility, then Chrome shipped an llms.txt check in Lighthouse. John Mueller’s explanation: llms.txt is “a temporary crutch to save some tokens” for AI coding tools parsing developer docs — not something regular websites need.

Ahrefs tested it with real data — server logs from 137,000 sites:

• 28% of domains publish an llms.txt file
• 97% of those files received zero requests — not from bots, not from humans
• Of the files that were read, 77% of visitors weren’t AI bots at all
• Real AI bots accounted for only 19.5% of requests
• Another 12% were tools studying llms.txt, reading each other

SEOs write llms.txt files that only get read by the tools built to study llms.txt. Perfect nesting doll.

What to do: Skip the llms.txt file. Spend that effort making your existing content easier to crawl, parse, and cite — clear titles, logical structure, complete information. That beats an index file nobody reads.

Myth 3: “Add schema and AI citations will jump”

Reality: after 30 days, citations barely moved.

“GEO experts” love this line: add Schema and your AI visibility takes off instantly. Ahrefs ran a controlled test: 1,885 pages that added JSON-LD schema vs. 4,000 control pages, tracking citation changes:

• After 30 days, no meaningful uplift on Google AI Mode or ChatGPT
• AI Overviews actually declined 4.6% — but so did the control group, pointing to an overall trend, not schema

Schema isn’t useless — it’s just not a shortcut. Databases like Wikidata and Google’s Knowledge Graph have absorbed schema for years, and whether that structure shapes how AI models understand entities remains an open question. If it feeds AI understanding at all, it’s through the Knowledge Graph: slow and indirect.

What to do: Keep using schema, but set the right expectations — it’s for long-term entity building. Mark up Organization and Person data with sameAs links to Wikipedia, Wikidata, and Crunchbase. A well-defined entity is the foundation for how AI models understand you. Don’t treat it as an instant switch.

Myth 4: “You don’t need classic rankings for AI to cite you”

Reality: 88.46% of ChatGPT’s citations come from the traditional search index.

The “AI SEO gurus” tell you traditional SEO is dead — go straight to AI optimization. Ahrefs analyzed 1.4 million ChatGPT prompts, categorizing every retrieved URL by source channel:

• 88.46% of citations came from the general search index — good old-fashioned search rankings
• Specialized channels like Reddit and YouTube get pulled in at scale but are rarely cited

Why? Because ChatGPT runs retrieval-augmented generation (RAG) — it searches first, then summarizes. Its search pool is traditional search.

What to do: Traditional SEO isn’t dead — it’s the front door to AI citations. Keep doing the work that earns rankings: content, internal links, backlinks, technical SEO. The better you rank on Google, the more likely ChatGPT cites you. It’s not either/or — it’s upstream and downstream.

Myth 5: “Rank #1 on page one and AI will cite you”

Reality: only ~38% of URLs cited in AI Overviews also ranked in the top 10 for the same query — down from ~76% a year earlier.

Wait, doesn’t that contradict Myth 4? No — and here’s the key distinction: rankings matter, just not for the keywords you’d expect.

Ahrefs analyzed 863K SERPs and 4 million citations. The rest of the cited URLs ranked further down — or didn’t rank in the top 100 at all.

The reason is Google’s query fan-out mechanism: AI pulls citations from related searches, not just the user’s original prompt. So you might rank #1 for your “target keyword” — but AI cites a page of yours sitting at position 50, because it matched one of the fan-out sub-queries.

What to do: Don’t build around one target keyword. Build topical coverage across a cluster of related queries — every sub-question under a topic deserves a real answer, because you never know which angle AI will “fan out” to your page.

Myth 6: “AI answers are too volatile to track”

Reality: the wording changes 70% of the time — but AI’s opinion barely shifts.

“AI answers are unstable, tracking is pointless” — true? Ahrefs tracked 43,000 keywords for a month with Brand Radar, checking each one’s AI Overview 16+ times:

• Wording: changed 70% of the time
• Brands mentioned: shifted 46%
• Cited sources: swapped 45.5%
• But the substance underneath barely moved

Surface drift, stable substance. AI says the same thing in different ways — but who it believes is the answer stays remarkably consistent.

What to do: AI visibility tracking is absolutely worthwhile — just don’t watch the wording. Watch position: is your brand mentioned? Where? Alongside whom? Look at monthly trends, don’t get dizzy over daily wording changes.

Myth 7: “AI search is killing website traffic”

Reality: Google sends 190x more traffic to websites than AI platforms.

This is from Ahrefs’ Patrick Stox. ChatGPT’s search volume has reached 12% of Google’s — scary headline, right? Now look at traffic:

• Google sends 190x more traffic to websites than ChatGPT
• The reason is simple: Google’s business model is “send users to websites”; ChatGPT’s is “keep users in the conversation”

The data also shows ChatGPT users aren’t mostly searchers — an OpenAI/Harvard study of 1.5 million conversations found only 24% were pure search; 51.6% were “ask intent.”

What to do: Don’t panic over “AI search share exploding” headlines. AI platforms still contribute a tiny slice of traffic for almost any site. Pay attention to AI visibility, sure — but don’t bet the farm on AI traffic. Google is still the main breadbasket for independent sites.

Myth 8: “AI Overviews are a minor impact”

Reality: AI Overviews cost you more than half your clicks — 58%.

The debate about AI Overviews eating clicks has been running since launch. Ahrefs’ quantification: for queries where an AI Overview appears, sites lose 58% of clicks.

It’s a brutal number, but there are two sides: you lose the clicks AI Overviews intercept, while gaining some brand exposure and follow-up searches. For information-heavy queries, the substitution effect is especially strong.

What to do: Understand your traffic structure. If core keywords trigger AI Overviews heavily, walk on two legs — optimize to be cited in AI Overviews (it may not bring direct clicks, but it brings brand exposure), and shift budget toward long-tail, comparison, and decision-type queries where AI Overviews are weaker and direct click intent is higher.

Myth 9: “Backlinks matter more than brand mentions”

Reality: for AI visibility, brand mentions beat backlinks.

The iron rule of traditional SEO is “backlinks are king.” But in the AI citation context, Ahrefs’ data points elsewhere: AI trusts how people talk about you, not who links to you.

In AI training data, brand mentions are everywhere — industry reports, media coverage, community discussions all shape AI’s perception of you. Meanwhile, a backlink might as well not exist as far as the model knows.

What to do: Keep building links, but elevate brand mentions to equal or higher priority. Concretely:

• Get cited in industry reports and data studies (publishing data is the shortcut)
• Earn media coverage and industry roundup inclusion
• Be genuinely discussed on Reddit, Quora, professional communities
• Do podcasts, interviews, co-created content

The Core Takeaway: Fundamentals Beat Hacks

After 15 million data points and nine debunked myths, Ahrefs’ conclusion is one sentence: in the AI search era, winners aren’t the ones playing tricks — they’re the ones with solid fundamentals.

• Classic rankings are still the main source of AI citations
• Content quality, topical coverage, and entity clarity beat any “trick file”
• Brand reputation built across independent sources is the ultimate source of AI trust

Neo’s Take

1. The biggest value of this data: turning “mysticism” into “science”

AI search is the field most starved for data. Over the past year, “AI SEO secret playbooks” have been everywhere — mostly single case studies, personal experience, or pure speculation. This set of research at least gives us a verifiable coordinate system: which investments pay off, and which are self-soothing.

2. The three most valuable conclusions for DTC sellers

First, traditional SEO isn’t dead — it’s upstream of AI. 88.46% of citations come from the traditional search index. The SEO skills you’ve built over the years haven’t depreciated — they’ve become more valuable.

Second, stop believing in “file-based” optimization. llms.txt is 97% unread, schema doesn’t move citations — these static tricks have diminishing returns. What actually works is dynamic fundamentals: content quality, topic depth, entity clarity, brand reputation.

Third, brand building just went from “nice to have” to “must have.” In the AI era, how many independent sources mention your brand directly determines AI’s trust in you. For smaller sellers this is actually an overtaking opportunity — because big companies are equally clueless about “how to be remembered by AI.” The starting line isn’t that far apart.

3. A fair word

All nine myths were tested using Ahrefs’ own research, and Ahrefs sells SEO tools — they have their own bias in data collection and interpretation (they dismiss llms.txt while selling AI visibility tracking tools). That said, the sample sizes and experimental designs are the ceiling of the AI search field right now. You can question the conclusions; it’s much harder to question the data.

4. My action list

After reading this research, I set myself four rules — sharing them with you:

• Content first: “writing one topic deeply” beats “covering more keywords”
• Topic cluster thinking: cover a whole set of sub-queries around a topic; don’t bet on a single keyword
• Treat brand mentions like backlinks: aim for at least one industry report, media, or community mention per quarter
• Track position, not wording: watch your brand’s appearance rate and position in AI answers monthly

Data is dead; how you use it is alive. The most valuable thing this research does isn’t telling you “what doesn’t work” — it helps you move limited budget from things that probably don’t work to things that probably do.

Don’t let AI search noise distract you. Build your content solidly and establish your brand — no matter how search changes, you’re still at the table.