What Actually Changed

"AI is killing SEO" is already a saturated take, and most of what's written about it is punditry rather than data. The narrower, more useful question for anyone who actually builds and runs websites for clients is: what's concretely different about where traffic comes from, what's now hammering servers, and what "optimize for search" even means when the searcher is a bot summarizing your page for someone who never clicks through.

Three things are happening at once, and they compound rather than existing independently: Google's own results page has stopped sending most clicks to anyone; the AI chatbots that were supposed to pick up the slack haven't; and the same AI companies scraping the web to build their answers are hitting servers harder than ever, often ignoring the one mechanism site owners have always used to opt out.

Google's Own Results Page Stopped Sending Clicks

Pew Research Center tracked 900 US adults and 68,879 real Google queries in March 2025: when an AI Overview appeared, users clicked a traditional result only 8% of the time, versus 15% without one. Only 1% clicked a link inside the AI summary itself, and session abandonment — no click at all — jumped from 16% to 26%. Pew: Google users are less likely to click on links when an AI summary appears.

That study is worth a fairness caveat, and not just because Google publicly disputed it. Independent ex-Bing analyst Duane Forrester raised real, separate issues: the 68,879-query sample is roughly 0.0000134% of Google's ~500 billion monthly queries, Pew's own disclosed margins of error run as wide as ±13.7 points for some age groups, and the study compared live March queries against researcher-run April re-queries — a problem, since AI Overviews are known to vary session-to-session even for an identical query. Search Engine Journal: Validity of Pew Research on Google AI Search Results Challenged.

The direction of the finding still holds up independently of Pew's exact numbers, though. Ahrefs (December 2025, 863,000 keywords and 4M+ AI Overview URLs) measured a 58% click-through-rate reduction for top-ranking pages when an AI Overview is present, up from 34.5% in April 2025 — via MediaNama. Separately, Search Engine Land / Kevin Indig found a roughly 50% desktop CTR drop with an AI Overview present, using its own independent dataset. And SparkToro/Datos put the bigger picture in context: as of early 2026, under a third of Google searches still send a click at all — Less than one third of Google searches still send a click.

The Chatbots Aren't Making Up the Difference

This is the part that undercuts the "just optimize for ChatGPT instead" narrative. Google Search and Discover referrals fell 34% and 15% respectively over a year; small publishers (1K-10K daily views) lost up to 60% of traffic, larger ones (100K+) around 22% — AdExchanger: The AI Search Reckoning.

All AI platforms combined — ChatGPT, Perplexity, Claude, the rest — account for roughly 1% of publisher traffic. Google alone is still around 40% of total site traffic; ChatGPT is about 0.21%, with a search-style CTR roughly 96% lower than Google's. 9to5Google: AI links are "less than 1%" of traffic, Chartbeat via Nieman Lab. For texture: The Planet D, a travel blog running since 2008, lost half its traffic after AI Overviews launched and folded entirely; Business Insider's organic search traffic fell 55% over three years, followed by a 21% staff cut — TechCrunch on ChatGPT referrals.

The net effect isn't "traffic moved from Google to AI chatbots." It's that a meaningful share of traffic simply stopped existing as a referral at all.

The Same Companies Are Scraping Harder Than Ever, Rules or No Rules

This is the piece most relevant to running the actual infrastructure behind a site, not just its marketing. Cloudflare Radar (June 2026) found that automated requests now make up 57.5% of HTML traffic versus 42.5% human — bots hold the majority for the first time. Cloudflare: Content Independence Day.

robots.txt is a voluntary convention with no technical enforcement behind it, and that gap isn't hypothetical. Cloudflare directly accused Perplexity of ignoring it — rotating IPs and spoofing a Chrome user-agent to keep scraping sites that had explicitly blocked PerplexityBot, across tens of thousands of domains. Perplexity's defense was that "user-driven fetching" isn't "crawling"; Cloudflare de-listed them as a verified bot afterward. Cloudflare: Perplexity is using stealth, undeclared crawlers, TechCrunch.

The practical response has moved from "add a robots.txt line" to actual infrastructure: Cloudflare's AI Crawl Control (Allow / Charge / Block per bot, pay-per-crawl at $0.01+ per request) and a new Content-Signals directive distinguishing Search, Agent, and Training bots. Cloudflare AI Crawl Control docs. Sites that blocked AI crawlers outright saw traffic drop as much as 75% — useful as a sense of scale, not a recommendation either way. This is a rate-limiting and WAF problem now, squarely in a dev shop's lane: the real engineering question isn't "AI bad," it's distinguishing legitimate bots (declared user-agent, respects Disallow) from the adversarial ones (IP rotation, UA spoofing) at the server level.

What "SEO Deliverable" Now Includes

The industry has landed on two overlapping terms for optimizing content so an LLM cites or recommends it, rather than optimizing a page to rank in a list of links: Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO). Seer Interactive: What is GEO?, CXL: AEO comprehensive guide. LLMs are trained on scraped corpora like Common Crawl, so backlinks and mentions — including a solid Wikipedia entry — increase how often a brand surfaces in the training set itself, independent of whatever live web-search tool a model also has access to. There's research quantifying this, not just SEO-industry claims: a "Conditional Monopoly" effect where a well-known brand gets recommended 100% of the time when competing products are spec-identical, though that dominance evaporates with as little as a +0.1-star rating advantage for a competitor. arXiv: Incumbent Advantage — Brand Bias in LLM Recommendation Systems.

HubSpot is a clean illustration of how seriously this shift is being taken. Co-founder Brian Halligan coined the term "inbound marketing" in 2005-2006, and blogging for organic search traffic was its foundational tactic. HubSpot's own blog is now telling marketers to pull investment out of exactly that kind of top-of-funnel informational content — the kind an AI Overview answers directly, zero-click — and reinvest in comparison content, proprietary research, and tools or calculators, while tracking AI-citation frequency alongside organic sessions rather than instead of it. HubSpot: SEO trends for 2026 — how search and AI are changing, HubSpot: Zero-click searches and the future of your marketing funnel. The company that popularized the old playbook is now publicly changing it, which is a more reliable signal than any single traffic study.

What This Means for Client Sites

None of this is "the web is dying." It's a plumbing change: fewer clicks per search, no meaningful replacement channel, more automated load on the server, and a wider definition of what counts as being found. For a small shop building and maintaining client sites, three things follow directly:

• Rate-limiting and bot-management decisions belong in the infrastructure conversation now, not as an afterthought — the majority-bot traffic numbers above make that a server-cost question, not just a security one.
• "SEO work" has quietly expanded to include GEO/AEO-relevant signals — structured data, clear citations, a credible external footprint — whether or not a client has asked for it by name.
• Traffic reporting for clients needs a caveat it didn't used to need: organic sessions are a shrinking fraction of how a page actually gets read, and a flat or declining number doesn't necessarily mean the content stopped working.