In the last 24–48 hours, one dataset has quietly given marketers a clearer picture of what “AI search” really means in practice: where ChatGPT actually sends traffic, and when it decides to reach for live web data.

Semrush analyzed more than 1 billion lines of U.S. clickstream data from October 2024 through February 2026 to map how people use ChatGPT and where they go next (Semrush). The headlines are stark: outbound referral traffic from ChatGPT grew 206% year over year (January 2025 vs. January 2026), yet Google alone captures 21.6% of those clicks, and the top 10 destinations take just over 30% (Semrush).

If you run marketing for a business, this changes how you should think about SEO, “GEO,” and attribution. AI assistants are not simply stealing traffic from search. They’re becoming a new routing layer—often routing users back into Google.

What the data actually says (and why it matters)

The Semrush study surfaces five metrics every marketing leader should internalize:

  • ChatGPT’s outbound referral traffic grew 206% in 2025 (Jan 2025 vs. Jan 2026) (Semrush).
  • Google receives 21.6% of all ChatGPT referral traffic, and the top 10 domains receive just over 30% as of February 2026 (Semrush).
  • ChatGPT enabled web search on 34.5% of queries as of February 2026 (down from 46% in late 2024), meaning most responses still come from training data alone (Semrush).
  • For most of the study period, 65%–85% of prompts couldn’t be matched to any traditional keyword in Semrush’s database—because prompts look more like problem statements than keyword strings (Semrush).
  • Users are going deeper: average prompts per session rose to 1.75 by February 2026 after a long period of ~1.16–1.21 for most of 2025 (Semrush).

The strategic implication: the biggest “competition” from AI assistants isn’t necessarily other brands getting cited—it’s the assistant choosing whether to use the live web at all. If the model doesn’t trigger search, it can’t cite you, click you, or discover your newest offers.

Why “being cited by AI” is really two problems

Most marketers talk about AI visibility as one thing. It’s not. You have two separate challenges:

  • Retrieval eligibility: When does the assistant decide it needs live web data (or sources) for the prompt?
  • Selection + trust: If the assistant does pull sources, why does it choose one page (or brand) over another?

Semrush’s finding that ChatGPT enables web search for roughly one-third of queries gives you a sobering baseline (Semrush). If your category is “evergreen” (definitions, how-tos, general explanations), you’re often competing against the model’s memory—not the open web.

On the other hand, if your category is time-sensitive (prices, regulations, comparisons, new product releases), you have a better shot because the assistant needs live information. That’s why a content strategy built for AI search must include “freshness triggers” (updates, dates, versioning, new data) alongside evergreen authority.

The Google boomerang effect: AI assistants send users back to search

Many business owners fear AI assistants will eliminate search traffic. The Semrush data suggests something more nuanced: ChatGPT is frequently a pre-search step, not a replacement, and Google is the biggest beneficiary of those outbound clicks (Semrush).

This “boomerang effect” matters for measurement and planning:

  • Your SEO foundation still matters because users often jump from ChatGPT to Google to verify, compare, or navigate directly to a brand.
  • Branded demand becomes a force multiplier. If the assistant introduces your brand and the user then searches your name, you win even if you weren’t the clicked citation.
  • Attribution gets messier. Your analytics may show “Google / organic,” but the real first touch was ChatGPT. That can lead teams to underinvest in AI visibility work if they only trust last-click reporting.

Semrush puts the strategic shift bluntly: “Marketers should stop seeing ChatGPT as a destination. It is increasingly becoming a doorway to the open internet” (Semrush).

Action plan: how to make your content more AI-visible (without chasing hype)

If you want to win in GEO and AI-assisted discovery, focus on what you can control: content structure, proof, and distribution signals.

  • Publish “decision pages,” not just blog posts. Create pages that answer: who it’s for, pricing ranges, implementation timeline, alternatives, and proof points. These are the pages assistants can cite when users ask, “What should I choose?”
  • Write for extractability. Use clear subheads, short paragraphs, and specific claims tied to data or standards. If your key point can’t stand alone as a single paragraph, it’s harder to reuse in an answer.
  • Add freshness and versioning. Put “Last updated” dates, update logs, and new benchmarks where relevant. You want to be the obvious choice when an assistant needs recent information.
  • Instrument AI as a channel. Track brand lift and “assist” effects: spikes in branded search, direct traffic after content launches, and referral patterns from AI assistants where available.
  • Don’t abandon classic SEO. The data shows AI assistants often route users back to Google. If your SEO is weak, you lose the second click even if the assistant sparked interest.

One more operational takeaway from Semrush’s keyword mismatch finding: if 65%–85% of prompts don’t map cleanly to keywords, keyword research alone won’t tell you what to publish (Semrush). You need to collect real customer questions (sales calls, support tickets, onboarding notes) and turn them into structured pages that answer the full scenario.

What to do next

AI search is no longer theoretical. It’s a measurable acquisition layer with its own quirks: a limited “live web” trigger rate, heavy concentration of outbound clicks, and a surprising tendency to send users right back to Google (Semrush).

If you want Real Internet Sales to help you build a citation-ready content engine (and an SEO foundation that converts the “boomerang” traffic), we’ll map your highest-value customer questions, build the pages AI assistants prefer to cite, and track the impact across search and pipeline.

Call 803-708-5514 or visit realinternetsales.com to get started.