AI search accounts for 1% of traffic. It also accounts for 2.9%. Both figures are from credible 2026 sources, both are correct, and they are measuring different things. Almost nobody repeating them says which.
The GEO field has a citation problem of its own. Statistics travel from a study to a blog post to ten other blog posts, losing their denominator, sample size, and date along the way.
This is an audit of the numbers you will see quoted most often, what each one actually measures, and where the apparent contradictions come from.
Key Takeaways
- Most contradictions are denominator problems: share of site visits and share of global web visits are different measures.
- Sample sizes vary wildly: some widely-quoted figures rest on a few hundred runs, others on millions of citations.
- Platform scope is usually omitted: a ChatGPT-only finding gets repeated as a fact about "AI search."
- Freshness numbers measure different windows: two months and eighteen months are both cited as thresholds.
- The directional picture is consistent: the disagreements are about magnitude, not direction.
How Much Traffic Does AI Search Actually Send?
Two figures circulate: roughly 1% and roughly 2.9%. Conductor's 2026 benchmarks put AI referrals at about 1% of site traffic. SparkToro's analysis put AI tools at about 2.9% of global web visits. These are not in conflict, because they use different denominators and different populations.
Site-level referral share depends heavily on which sites are in the sample. Global visit share counts activity on AI platforms themselves, not just outbound clicks. A B2B SaaS site and a recipe blog will sit in very different places within either range.
What to do with it: quote the range, not a single number, and state the source. Anyone presenting one figure as the answer has skipped a step.
Do Top-Ranking Pages Get Cited?
The most-repeated finding is that roughly 12% of AI citations also rank in Google's top 10. This appears in Ahrefs' analysis of 1.9 million citations and matches Moz's finding that 88% of Google AI Mode citations do not appear in the organic top 10 across nearly 40,000 queries.
Two independent studies, different methods, converging results. This is one of the better-supported numbers in the field.
The caveat: convergence on a percentage does not tell you the mechanism. Whether the cause is chunk structure, freshness, entity signals, or source type varies by study and remains genuinely contested.
How Big Is the Freshness Effect?
Two thresholds circulate. One holds that content updated within roughly two months earns about 28% more citations. Another, from a study of 17 million citations, found AI-surfaced URLs run about 25.7% fresher on average than Google's organic results for the same queries.
These measure different things. The first is a citation lift tied to an update window. The second is a comparative freshness gap between two result sets. Repeating them as though they describe the same effect is a common error.
What holds: recency matters more in AI retrieval than in organic ranking. The precise window is less established than the direction, which is why our AI Citation Freshness report looks at decay patterns rather than quoting a single threshold.
The Numbers That Deserve More Skepticism
| Claim | Basis | Concern |
|---|---|---|
| Incumbents hold 64.3% of recommendation slots | 450 ChatGPT runs | Small sample, single platform, precision implies more rigor than the sample supports |
| Top 2% of brands take 78% of recommendations | Secondary reporting | Primary source rarely linked; category scope unclear |
| Review profiles make citation 3x more likely | Secondary reporting | Correlational; brands with review profiles differ in many other ways |
| Chunk size is 200-400 words | Varies by source | Also cited as 200-500 tokens and ~128 tokens; implementation-specific and not public |
None of these are necessarily wrong. All of them get repeated with a confidence their underlying evidence does not support.
Why Do These Numbers Disagree So Often?
Four recurring causes explain most apparent contradictions in GEO statistics: different denominators, different platform scope, different sample sizes, and different measurement dates in a fast-moving field.
Denominator drift
Share of site visits, share of global visits, share of queries, and share of sessions are four different measures. A figure quoted without its denominator is not usable, and the denominator is the first thing lost when a statistic gets repeated.
Platform scope collapse
A finding from ChatGPT runs becomes a fact about "AI search." This is especially misleading given how little source overlap exists between engines. Our analysis found 86% of top cited sources are not shared across platforms, which means single-platform findings generalize poorly by default.
Sample size asymmetry
Studies in this field range from a few hundred prompt runs to millions of citations. Both get quoted with the same authority and the same decimal precision. A figure like 64.3% from 450 runs carries implied precision the sample cannot support.
Date collapse
A statistic from early 2025 gets repeated in mid-2026 without a date. In a field where AI Overview coverage and platform behavior shift within quarters, an undated number is close to meaningless.
How to Evaluate a GEO Statistic
Five questions resolve most of it:
- What is the denominator? Share of what, exactly.
- Which platforms? One engine or several, and does the claim generalize?
- How large is the sample? Hundreds and millions warrant different confidence.
- What date? Anything over six months old needs re-verification.
- Is there a primary source? If the trail ends at another blog post, treat it as unverified.
Applied consistently, this filter removes a large share of what circulates as GEO fact. That is a feature. Reporting a range honestly is more useful to a decision than a precise number that does not survive checking.
What Actually Holds Up
Stripping out the contested figures, several findings rest on large samples, replicate across sources, or both:
- Organic rank and AI citation are substantially independent. Two independent studies converge near 12% overlap.
- Earned media dominates the citation pool. Muck Rack's analysis of 25 million cited links put it at 84%, stable across three editions.
- Unlinked mentions outperform backlinks. Ahrefs measured 0.664 against 0.218 across 75,000 brands.
- Platforms cite different sources. Replicated across multiple independent analyses.
- Recency matters more than in organic. Direction is well-supported; the threshold is not.
That list is shorter than most GEO content implies, and it is enough to build a strategy on. The gap between what is established and what is asserted is where most wasted budget in this field originates. Measuring your own position, as covered in tracking AI search visibility, produces more reliable input than any industry benchmark.
FAQs
1. What percentage of traffic comes from AI search in 2026?
Between roughly 1% and 2.9%, depending on measurement. Conductor's site-level benchmarks put AI referrals near 1% of site traffic, while SparkToro's analysis put AI tools at about 2.9% of global web visits. The figures use different denominators and are not in conflict.
2. Do pages that rank in Google get cited by AI?
Only about 12% of the time. Ahrefs' analysis of 1.9 million citations and Moz's analysis of nearly 40,000 queries independently converge on this figure, making it one of the better-supported findings in the field.
3. How fresh does content need to be for AI citations?
The direction is well-established but the threshold is not. One source reports a 28% citation lift for content updated within two months; a separate 17-million-citation study found AI-surfaced URLs run 25.7% fresher than organic results. These measure different effects and should not be conflated.
4. Which GEO statistics should I be skeptical of?
Any figure quoted without a denominator, platform scope, sample size, or date. Be particularly cautious with precise-looking percentages drawn from small samples, such as findings based on a few hundred prompt runs presented to one decimal place.
5. Which GEO findings are actually well-supported?
That organic rank and AI citation are substantially independent, that earned media accounts for the large majority of citations, that unlinked brand mentions correlate more strongly than backlinks, and that platforms cite substantially different sources. Each rests on large samples or replicates across independent studies.
This audit reflects published sources available as of August 2026 and describes what each study measured rather than endorsing any figure as settled. Where a primary source could not be verified, we have said so rather than repeating the number as fact.