ChatGPT Sends 20% of Its Traffic Straight Back to Google. The "AI Kills Search" Story Just Collapsed.

One in five people who used ChatGPT ended up on Google next. The share climbed from 14% in late 2024 to 21.6% by February 2026, according to new research from Semrush. The search analytics company analyzed more than a billion lines of clickstream data from a panel of 200 million users.

ChatGPT and Google fed each other. The theory had floated around marketing circles for months without data behind it. Semrush put hard numbers on it. For brand teams that spent 18 months debating whether to defund SEO or double down on AEO, the finding rewrote the question.

The Funnel Flattens Into a Loop

The Semrush data described a loop with three moving parts.

Exploration came first. A buyer opened ChatGPT and typed something like "best CRM options for a 50-person B2B company." That kind of conversational, detail-loaded phrasing accounted for 65 to 85% of ChatGPT usage. The model answered. The buyer got oriented.

Validation came second, on Google. The buyer searched the specific vendor names that surfaced in the ChatGPT response, then worked through review sites, case studies, pricing pages, and comparison content to check what the AI had told them.

Follow-up came third, back in ChatGPT. Something like "Compare Vendor X and Vendor Y for a team that uses HubSpot." The cycle repeated, sometimes across days, often inside a single session. Queries per ChatGPT session jumped 50% in late 2025, which meant buyers stayed in the tool longer and asked sharper questions the further in they went.

Every turn through the loop reinforced the last.

Buyers Stopped Typing Keywords and Started Writing Briefs

The average Google search ran 3.37 words. The average ChatGPT prompt ran 23 words, and some stretched past 2,700. Prompts at that length carried context, constraints, and criteria in one shot. They functioned as briefs.

The contrast was stark. A Google user typed "best project management software." A ChatGPT user typed "best project management software for a 12-person remote marketing team that uses HubSpot and needs Gantt chart views with client-facing portals." The second buyer was already in purchase mode. That kind of specificity gave away intent that short Google queries never did.

Content that earned a citation had to match that level of detail. Generic listicles got skipped. The citations went to opinionated, use-case-specific writing. Case studies with real numbers. Comparison pages that walked through real integration scenarios. Documentation that answered the exact question asked.

A company that knew exactly who it served, what problem it solved, and how it differed from its nearest alternative could produce content that matched real buyer prompts. Vague positioning produced vague content, and the model had no reason to cite it. Every CMO I talked to saw the shift, but few had turned it into content operations yet.

Win The Handoff From AI To Google

The playbook split into two. Get cited in the AI answer. Hold up in the Google result. The gap between the two was where deals quietly broke.

The first move was the AI audit. The winning teams ran their top buyer prompts through ChatGPT, Perplexity, and Gemini to see whether they showed up, how the sentiment read, and which competitors surfaced when they didn't. HubSpot's AEO product automated part of the work. A manual pass delivered most of the signal in an afternoon.

The second was the validation audit. When a buyer heard a vendor name in a ChatGPT response and searched it on Google, the results had to tell the same story. The website had to be clear about who the company served. G2 reviews and comparison pages had to hold up under a skeptical click. The AI citation was the introduction. The Google result was the second impression. When those two contradicted each other, the deal went away.

The third was content built for the way buyers actually prompted. A 23-word question deserved an answer written with the same specificity. Only the brands that matched that depth earned the citation the next time around.

Search Grew. The Old Playbook Didn't.

The "AI replaces search" framing made a good headline. It didn't make a good roadmap. AI expanded the search market instead of eating it. Total search volume across engines and LLMs grew 26% worldwide in the same period, and Google held its spot as the single largest referral destination on the internet. ChatGPT opened another door into that traffic.

The SEO-versus-AEO debate was the wrong binary. The harder assignment was earning a place in a loop that ran in both directions. The brands that pulled it off compounded visibility no paid channel could replicate, and they took share without announcing it. Their pipeline did the announcing for them.