When AI Search "Kills" Your Traffic, It's Telling You Something True

I've been studying and practicing SEO for more than 20 years. I've watched Google roll out Panda, Penguin, Hummingbird, RankBrain, BERT, MUM, the Helpful Content Update, and now AI Overviews and AI Mode. Every one of these was supposed to "change SEO forever." Every one sent agencies into a panic about reinventing their playbook.

Here's the reality I keep coming back to: not much has actually changed.

The tactics shift around the edges. The fundamentals don't move. And right now, with traffic down 20%, 40%, even 60% across countless sites, I'm watching the industry tell two stories that are mostly wrong.

Story one: AI is destroying SEO. Google AI Overviews and ChatGPT and Perplexity are eating the open web. Adapt or die.

Story two: You need to "optimize for AI" with whole new disciplines — GEO, AEO, AIO, LLMO. Pay someone to rebuild your strategy from scratch.

Both stories sell well. Both miss what's actually happening.

The data nobody wants to put plainly

Here's what the numbers actually show. AI Overview prevalence in Google searches grew from roughly 6% in January 2025 to over 50% by October 2025. Click-through rates on top-ranking pages dropped 58% when an AI Overview is present, according to Ahrefs research published in February 2026 — nearly double the 34.5% decline measured a year earlier. Some publishers have lost 20%, 30%, even 90% of their traffic, as documented by AdExchanger in their January 2026 reporting on the AI search reckoning. The travel blog The Planet D told Bloomberg it lost half its traffic in the months after Google launched AI Overviews in May 2024.

Those numbers are real. They're scary. And they only tell half the story.

Look at what's underneath them.

A 2026 analysis from Wellows examined 2,400 AI citations and found 96% of them came from sources with strong E-E-A-T signals — Google's framework for Experience, Expertise, Authoritativeness, and Trustworthiness. The same analysis tested how various ranking factors correlate with whether AI engines actually cite your content. Domain Authority, the metric most agencies still chase, correlates at r=0.18. E-E-A-T correlates at r=0.81. That's not a small difference. That's the difference between "barely matters" and "the dominant factor."

Pages ranking #6 through #10 with strong E-E-A-T are cited 2.3 times more often than #1-ranked pages with weak E-E-A-T. Read that twice. The AI doesn't care about your rank. It cares whether you're a trustworthy source.

What AI search actually changed

Old search worked like a phone book. Google showed you ten links. You picked one. The site that ranked first won most of the traffic, even if its content was thin, generic, or rehashed from somewhere else. Volume and links could carry mediocre content a long way.

AI search doesn't work like a phone book. It works like a reference librarian. When you ask a question, the AI synthesizes an answer from sources it judges credible, then cites the ones it leaned on. If your site isn't credible enough to be the source, you don't appear in the answer. There is no "page two." You're either cited or invisible.

AI search isn't killing SEO. It's exposing the SEO that was always weak.

The sites losing traffic are the ones that were never solid to begin with

Think about who's hurting most. Recipe blogs that bury a single recipe under 1,500 words of personal narrative to game the algorithm. Affiliate sites that paraphrased the same product roundup as fifty other affiliates. Aggregator pages that scraped Wikipedia and added some keywords. Generic "what is X" pages that any decent LLM can write better in three seconds.

These sites built traffic on the gap between what searchers wanted and what Google could deliver. AI closes that gap. The sites had no real moat — no original research, no first-hand experience, no proprietary data, no actual authority. They were filler that ranked because Google had no better option.

Now Google has a better option. The filler isn't ranking anymore.

What still works (the same things that always worked)

The brands keeping their traffic and earning AI citations are doing what good SEO has always required:

• They publish original work. First-hand experience, original research, proprietary data, case studies with measurable outcomes. Things AI literally cannot summarize from someone else because nobody else did the work.
• They have visible, named authors with real credentials. Anonymous content gets ignored. Content with a real expert behind it — LinkedIn profile, qualifications, track record — gets cited.
• They're transparent. Clear pricing, accurate sourcing, consistent business information across the web, real contact details. AI systems explicitly verify these signals before deciding whether to trust your site.
• They're focused. Sites that try to cover everything end up authoritative on nothing. The brands surviving the AI shift have tight topical focus — they own a specific niche deeply rather than skimming dozens.
• They get cited elsewhere. Brand mentions across Reddit, LinkedIn, industry publications, and other authoritative sources signal authority better than backlinks ever did. Ahrefs analyzed 75,000 brands in 2026 and found that brand mentions correlate with AI Overview visibility three times more strongly than backlinks (correlation of 0.664 versus 0.218).

None of this is new. This is E-E-A-T. This is what Google's quality guidelines have asked for since 2014. The difference is that before AI, you could fake it with volume and clever optimization. Now you can't.

What I'd actually do if my traffic dropped

If I were watching my traffic decline right now, I wouldn't panic-buy a "GEO strategy" from someone who renamed their SEO services last quarter. I'd ask harder questions:

• Is the content I'm publishing actually better than what an AI can generate from training data? If not, why would anyone cite it?
• Do my pages have a named, credentialed author — or are they ghost-written and unsigned?
• When AI engines do cite my niche, who do they cite? What do those sources have that I don't?
• Am I producing original work that someone would have to come to my site to get? Or am I rephrasing what's already out there?

The brands that answer those questions honestly and act on the answers will be fine. Better than fine — they'll consolidate. The flood of generic AI content has made authentic expertise scarcer and more valuable than it's been in decades.

The rest will keep blaming AI for problems that were always there.