There is a pattern to most of the more breathless claims about AI search doing the rounds at the moment. They sound plausible, they fit a narrative people are primed to believe, and they get repeated until they start to feel like settled fact. The problem is that when you actually check them against data, a fair number fall apart.
Search Engine Land has just published a useful piece by Rob Tindula at NP Digital that tests seven of the most persistent myths against real-world studies and measurement. Two of the findings are worth calling out separately because they change how you should be thinking about both traditional search strategy and AI visibility work. The rest are useful corrections to things that keep getting stated as certainties when the evidence does not support them.
Traditional search is not being cannibalised, but clicks to websites are falling
The claim that gets repeated most often is that every query answered by ChatGPT, Claude or Perplexity is a search Google never saw, and that traditional search volume is therefore shrinking as AI eats its share. The State of Search Q2 2026 report found the opposite. Traditional search and AI search are both growing, quarter on quarter, at roughly the same rate. It is not substitution, it is addition.
Where the picture gets more complicated is in what happens after the search. Google is sending fewer clicks to websites. Desktop clicks to non-Google-owned results hit their lowest point since April 2025. The issue is not that demand is disappearing. It is that more queries are being answered directly, without a click out. Search volume is up, but website visits from search are down. That is a different problem from the one most of the rhetoric assumes, and it requires a different response.
The corollary myth is that organic traffic growth is impossible in a zero-click environment. Tindula's piece includes a case study of a local business client that posted record organic traffic in July 2026, driven by nonbranded content. The caveat is that local businesses sit in a different risk category from the rest of the search landscape when it comes to AI, so this is not universally replicable. But the broader point stands. Zero-click pressure is real, but it is not evenly distributed, and a well-executed technical SEO strategy can still deliver meaningful growth in the right contexts.
Analytics undercount AI's impact by approximately 97 per cent
If you are judging AI's contribution to your traffic by looking at the AI referral line in your analytics, you are looking at a floor, not a ceiling. A study on the AI mention effect cited in the piece found that after an AI assistant mentions a brand, the user visits that brand's website at 1.5 to 2.5 times their forecasted baseline rate over the following seven days. Even after ChatGPT's May 2026 update increased citation rates, only about 2.5 per cent of those downstream visits carried a trackable referral parameter.
That means 97 per cent of visits influenced by an AI mention show up in your analytics as direct, organic or unattributed, with no indication that an AI conversation started the journey. The timing compounds the problem. Only about a fifth of those visits happen within an hour of the mention, 42 per cent within 24 hours, and the majority after day one. Attribution models built around same-session or same-day conversions will miss most of this entirely.
The practical consequence is that if you are making decisions about AI visibility work based on what you can see in a referral report, you are radically underestimating both the scale and the commercial relevance of the channel.
AI-generated content can rank, if the process is sound
The myth here is that Google and other search engines can detect and suppress AI-generated content, so anything written by a model is capped at mediocre performance. Tindula tested this directly by publishing four articles generated entirely through AI, with careful prompting and light human review but no manual rewriting. All four are still ranking months later.
The distinction that matters is not whether an LLM wrote the copy. It is whether the content adds value. Plenty of AI-generated content is low-effort, unoriginal and keyword-stuffed, and Google has improved at recognising that. But if the content starts with a real idea, genuine expertise, original data or a specific point of view, then AI can accelerate the writing in the same way it accelerates any other editorial task. Search engines do not have a "written by AI" penalty. They have a "does this add any value" filter.
llms.txt will not improve your AI visibility
The claim is that adding an llms.txt file meaningfully improves how often your site shows up in AI answers. It does not. Google has stated explicitly that it does not use llms.txt for AI search discovery. A study tracking llms.txt adoption across ten sites for 90 days before and after implementation found that most saw no measurable change in AI crawl frequency or AI-referred traffic, and the few that did also made other changes to their SEO at the same time.
The fair conclusion, as the piece puts it, is that llms.txt is infrastructure, not strategy. It will not hurt to have one, but it also will not move the needle on visibility by itself.
LLMs struggle with JavaScript, and sometimes lie about it
A widely held assumption is that modern AI crawlers are sophisticated enough to render JavaScript the way a browser does, so client-side-rendered content is visible to them in the same way it is to Googlebot. A study by Profound of AI crawler behaviour found the opposite. Emerging AI search providers do not have JavaScript rendering capabilities, and server-side rendering is crucial for AI visibility.
What makes this worse is that some LLMs will misrepresent their own capabilities. Tindula recounts asking ChatGPT to summarise an article whose content was hidden behind client-side JavaScript. ChatGPT said it could, and proceeded to provide a detailed summary. When pressed, it admitted it had not actually read the target article at all. It had found a different article on a similar topic that it could read, and returned that instead.
If your site is on a headless CMS or relies heavily on JavaScript to render page content, you cannot assume AI crawlers are seeing what users see, and you should not take an LLM's word for it if it claims otherwise.
AI finds more errors, but misses critical ones
The final myth is that SEO teams can be replaced by AI, now that tools can run audits, write briefs and execute technical fixes. An NP Digital study found that AI detects more errors overall than a human audit, catching small issues faster and at greater scale. But it still struggles with judgment calls around keyword research and search intent, and it routinely prioritises minor fixes like alt text or meta descriptions over changes that would actually move the needle on organic performance.
AI also hallucinates fixes, recommends pages that already exist, and will confidently agree with a bad idea if you phrase it assertively enough. It is a useful tool in the SEO process, but it is not a replacement for someone who understands the commercial context, knows what matters and can tell the difference between a cosmetic issue and a structural one.
What this means for your strategy
Every myth on this list has persisted because it simplifies something complicated into a single declarative sentence that fits neatly into a social media post. The reality is more textured. Traditional search is not dying, but the commercial value of a ranking is shifting. AI-generated content can perform well, but only if the underlying process is sound. AI mentions drive traffic, but your attribution model will not show you most of it unless you redesign how you measure influence.
The practical takeaway is to slow down and check the claim against data before you repeat it, and to be willing to hold two things true at once. AI is changing search quickly, and that makes it harder to verify every assertion that sounds plausible. The teams that win will not be the ones who moved fastest on the next hot take. They will be the ones who checked it first.