AI Search Answers Without Referring: 9.4% Search Displacement, 5.2% Outbound Clicks

Key takeaway: The web has long run on a bargain — publish content, and search sends visitors back. A study released in July 2026 quantifies how AI search weakens that bargain structurally: wider ChatGPT Search access cut traditional search use by 9.4%, and only 5.2% of ChatGPT conversations produced a click to an external site (versus 31.1% for Google searches). If you only watch visit counts, you miss how your brand is being handled inside the answers themselves (as of 2026-07-15; pre-peer-review preprint).

---

TL;DR

• A drop in search traffic may be structural change, not SEO failure. Broader ChatGPT Search access alone cut traditional search use by 9.4% — and reporting on the study says the decline widened to 17% after 20 weeks.
• AI answers without sending users out. Only 5.2% of ChatGPT sessions produced an outbound click — roughly one sixth of Google's 31.1%.
• So the metric changes. In a world of fewer clicks, "does the AI answer mention and cite us" comes before visit counts as your exposure metric.

What the study found — three numbers

The arXiv preprint "Answering Without Referring: How AI Search Rewrites the Web's Economic Bargain" (Qiaoni Shi, Kai Gu, Kai Zhu, July 2026) analyzed URL-level Comscore clickstream data from U.S. desktop users and used ChatGPT Search access expansions as a natural experiment to estimate how much AI search displaces traditional search (arXiv). It has not yet been peer-reviewed — but it is third-party behavioral data, not a platform's self-reported figure, which makes it worth taking seriously.

Metric	Figure	Meaning
Traditional search use	−9.4%	Weekly decline caused by wider ChatGPT Search access (arXiv)
After 20 weeks	−17%	Displacement deepens over time (SEJ coverage)
ChatGPT sessions with an outbound click	5.2%	Versus 31.1% for Google searches — about one sixth (arXiv · SEJ)

"ChatGPT produces outbound clicks in only 5.2% of conversation sessions." — Shi, Gu, Zhu, Answering Without Referring (arXiv, 2026)

Interpretation — the "traffic in return" bargain is weakening

The heart of this study is not traffic volume but deal structure. Traditional search brokered information demand to websites: in exchange for letting engines index content, sites got visitors back. AI search settles information demand inside the answer. When the user is satisfied, they go nowhere. That is precisely what the paper's title calls "answering without referring."

In practice, this structural shift invites two misreadings.

• Misreading 1 — "Traffic fell, so our content got worse." Even with unchanged content quality, inbound traffic falls once users start asking AI instead of searching. Doubling down on classic SEO without diagnosing the cause can mean investing more into a shrinking pipe.
• Misreading 2 — "No clicks, so AI search can be ignored." The opposite. If decisions conclude inside the AI answer, whether your brand appears in that answer is the exposure. And the remaining 5.2% of clicks are not to be discarded — visitors arriving from an AI answer come pre-contextualized, and multiple observations report higher conversion quality for AI referrals (related analysis).

What to do — shift from visit metrics to answer metrics

The starting move is shifting your measurement axis one notch: from "did they visit our site" to "are we inside the AI answer." Four recommendations:

1. Measure your presence in AI answers on a schedule — For the questions your customers actually ask, check whether your brand is mentioned across ChatGPT, Perplexity, Gemini, and others — and in what context (recommended, compared, negative). AI visibility measurement tools such as RanketAI automate this loop; RanketAI's brand visibility analysis tracks mentions, citations, and framing separately.
2. Check mentions and citations as separate metrics — Being cited as a source and being named are different things. A ghost citation — cited but never named — is especially costly in an environment where clicks are disappearing.
3. Instrument AI referrals separately — Within shrinking total traffic, AI-driven referrals are often growing. Separating AI referrers in GA4 and using Search Console's Generative AI reports reveals the "small but dense" stream that aggregate numbers hide.
4. Publish in citation-friendly form — Content built for extraction — direct answers, statistics, named sources — earns AI exposure better than long reads that assume a click. And since the decline deepened over 20 weeks, waiting widens the gap.

Frequently asked questions

Should we abandon SEO if search traffic is falling?

No. The study shows search shrinking, not disappearing — a 9.4% decline also means over 90% remains. The realistic posture is dual-track: maintain SEO while adding exposure in AI answers (GEO/AEO) as a separate, tracked metric.

Isn't 9.4% still a small change?

The slope matters more than the snapshot. Per the same coverage, the decline reached 17% twenty weeks after access expanded. The figure also isolates ChatGPT alone — effects from Google AI Overviews, Gemini, and other answer surfaces are not included.

If there are no clicks, why does AI exposure matter?

Because the decision concludes inside the answer. When a user asks "which tool is good?" and the AI names three brands, a brand outside that list never even enters the comparison. Exposure and recommendation still happen without a visit — measuring that is what AI brand visibility measurement is.

How do I check my site's AI referrals?

Start by separating AI-service referrers (chatgpt.com, perplexity.ai, and others) in GA4. On the Google side, Search Console's Generative AI performance reports now officially cover AI Mode and AI Overviews exposure. The stage before the visit — mentions and citations inside answers — requires separate measurement tooling.

Does this apply to markets outside the U.S.?

The study is based on U.S. desktop clickstream data, so the exact figures do not transfer directly. But the direction — growing ChatGPT usage and expanding answer surfaces — is shared across markets, so treat it as directional evidence and verify magnitudes with local data.

Related reading

• Why AI search traffic converts higher — the flip side of fewer clicks: why the remaining stream is denser.
• What a ghost citation is — cited without being named: the costliest loss in a click-starved environment.
• Google Search Console's Generative AI reports — AI search exposure measurement becoming an official tool.
• The AI search trust paradox — positioning a brand in a market where adoption is up and trust is down.