In the first four months of 2026, 68.01% of US Google searches ended without a click — up from 60.45% in 2024, according to SparkToro's analysis of Similarweb clickstream data. That figure excludes the Google mobile app, where zero-click behavior is more aggressive, so the true rate is higher.
The standard response has been to invent new acronyms and chase the citation. That is worth doing. But it misses the more consequential half of what is happening, which is that the visits still arriving are a different population than they used to be — and that is a conversion problem, not a search problem.
The Acronyms, Briefly
The terminology is genuinely muddled, and most of the overlap is real rather than a distinction anyone should agonize over.
| SEO | AEO | GEO | |
|---|---|---|---|
| Competing for | A ranked position | Being the extracted answer | Being a cited source |
| Surface | Results page links | Snippets, answer boxes | AI-generated responses |
| Main lever | Relevance and authority | Content structure | Brand entity strength |
| Mostly determined by | Your site | Your page | Pages you do not control |
| Measured in | Rankings, sessions | Snippet ownership | Citation share, mentions |
AEO is largely on-page: state answers first, keep passages self-contained, use tables and lists, mark up your data. It is within your control and it is mostly good writing discipline.
GEO is largely off-page. A model decides whether to cite you based on what it has read about you across the whole web — comparison articles, roundups, community threads, third-party profiles. You influence that; you do not own it.
The practical implication: if you are doing one thing, do AEO, because it is controllable and it compounds with everything else. GEO is a longer, less direct game.
The Part Nobody Optimizes For
Zero-click search does not remove visits uniformly. It removes a specific kind: the early, informational, low-intent one. "What is X." "How does Y work." "X vs Y, briefly." Those questions now get answered in place.
What survives the filter is the visitor who needed something a summary could not provide — a price, a limit, a demo, a specific integration, proof of a claim. The composition of your traffic shifts toward intent even as the volume falls.
This produces a pattern that is easy to misdiagnose:
- Sessions fall 25%
- Conversion rate rises
- Total conversions fall much less than 25%, or hold flat
A team watching a sessions dashboard reads catastrophe. A team watching conversions reads a manageable decline. A team watching conversion rate might even think something improved. All three are looking at the same underlying change: the low-intent visits stopped arriving.
The corollary is uncomfortable but freeing. If traffic growth is no longer available as a lever, conversion rate is what is left. For most sites, the achievable response to a structural decline in clicks is not to recover the clicks. It is to get more out of the ones that still come.
Why This Is the Low-Traffic Problem Again
Anyone who has run CRO on a low-traffic site will recognize the constraint immediately. Fewer sessions means longer tests, and a 25% traffic decline extends every test timeline by roughly a third at the same minimum detectable effect.
So the honest framing is: your conversion work matters more than it did, and your ability to run conventional experiments got worse at the same time. The tactics that address this are the same ones that always applied when volume is scarce:
- Test bigger changes. A 2% lift you cannot detect is worth less than a 15% lift you can. Page-level rewrites beat button-color tests when sample size is the binding constraint.
- Use variance reduction. CUPED and similar methods extract more signal from the same traffic.
- Prefer Bayesian or sequential approaches, which let you act on accumulating evidence rather than waiting for a fixed horizon.
- Use adaptive allocation. A multi-armed bandit spends less traffic on losing variants, which matters more when traffic is the scarce input.
What to Actually Do
Write for extraction, not just for ranking. Lead each section with the answer. Keep passages self-contained enough to quote. Use tables for anything comparative. This is AEO and it costs nothing beyond editorial discipline.
Publish the specifics that survive summarization. A model can paraphrase your positioning. It cannot invent your pricing, your limits, your integration list, or your methodology. Concrete, checkable, first-party facts are both what gets cited and what the surviving visitor came for. The same content does both jobs.
Fix the landing experience for high-intent arrivals. The visitor who clicks through post-summary wants confirmation and a path to act, not an introduction to the category. This is covered in detail in CRO for AI search traffic.
Change what you report. If your dashboard leads with sessions, it will describe the next two years as a disaster regardless of business performance. Lead with conversions and revenue; keep sessions as a diagnostic.
Do not chase lost volume with more content. Publishing more pages that compete for queries which no longer produce clicks is the most common and most expensive wrong response.
The Bottom Line
Two-thirds of searches ending without a click is a structural change, not a downturn to wait out. GEO and AEO are worth doing — AEO especially, since it is controllable and compounds — but neither restores the traffic mix that existed before.
What they do produce, along with the shift itself, is a smaller and better-qualified stream of visitors. Whether that is a crisis or an improvement depends almost entirely on what happens after the click. That has always been the conversion team's job; it is now a larger share of the total opportunity than it has ever been.
If fewer sessions means your tests no longer reach significance in a useful timeframe, that constraint is exactly what Surface AI was built around — continuous adaptive optimization that keeps improving without waiting on fixed-horizon sample sizes.