What actually moves AI citations, and what doesn't
Most of this category sells on assertion. Below is every claim we make, the study behind it, the sample size and the measured effect — including the findings that are inconvenient for people selling AEO retainers.
| Finding | Measured effect | Detail | Source |
|---|---|---|---|
| Adding JSON-LD schema markup produces no measurable AI citation lift on its own. | −4.6% / +2.4% / +2.2% | Matched difference-in-differences across 1,885 pages that added schema, measured on Google AI Overviews, AI Mode and ChatGPT. All three statistically indistinguishable from zero. | Ahrefs, 2026 |
| Earned third-party media is cited far more heavily than brand-owned content. | +325% | AI search shows what the authors call a systematic and overwhelming bias toward earned media over brand-owned and social content. | University of Toronto, 2026 |
| Deep entity linking is the highest-impact single change measured. | +34% (+52% Gemini/Bing) | Adding sameAs arrays of at least eight authoritative profiles to Person and Organization JSON-LD, over a 90-day controlled test of 12 changes and ~50,000 citations. | GEO Measurement Study, 2026 |
| Gating content all but removes it from AI answers. | 14 vs 1,847 | Two gated whitepaper landing pages earned 14 citations across 90 days; open equivalents on the same topics earned 1,847. | GEO Measurement Study, 2026 |
| Quotations, statistics and inline citations measurably raise visibility. | +41% / +32% / +30% | Adding relevant quotations raised visibility ~41%, statistics ~32%, citations ~30% and fluency optimisation ~28%. | Princeton GEO paper |
| An active review profile is close to a prerequisite for being recommended. | 1% vs 75.3% | Brands with no active review profile appeared in 1% of answers. Brands that actively collected and responded to reviews appeared in 75.3%. | Onely, 2026 |
| Keyword density does nothing. | zero correlation | Varying keyword frequency by 3× across a control set of six pages showed no correlation with citation share on any engine. | GEO Measurement Study, 2026 |
| AI referral traffic is small but converts several times better than organic. | 1–6% of traffic, 4–5× conversion | AI referrals were ~1.08% of traffic in late 2025, rising to ~6.4% for B2B tech by January 2026, growing ~975% year on year. AI-referred visitors converted at 14.2% against 2.8% for Google organic; the multiplier runs from 1.3× in low-consideration ecommerce to 23× in B2B SaaS. | AirOps / Pixis, 2026 |
| Being cited no longer requires ranking on page one. | 37.9% | Of cited URLs, only 37.9% ranked in the first ten organic result blocks — down from roughly 76% in July 2025. Citation and ranking are coming apart. | Ahrefs, 863,000 SERPs, March 2026 |
| Nobody owns the answer for this category in Australia yet. | 24 firms, ~0 agreement | Four AI engines asked for the best AEO agencies in Australia named 24 different firms, with no engine agreeing on a top firm and only two or three firms appearing more than once. About 40 Australian agencies self-describe as AEO providers; five have published a formal methodology. | Bushletter AI engine test, 2026 |
Where the evidence disagrees with itself
Semrush's study of 304,805 URLs reports a +22% citation lift associated with structured data, which sits awkwardly against the Ahrefs finding of roughly zero. The difference is study design: the Semrush figure is an association — pages with schema do get cited more — while the Ahrefs figure comes from a matched difference-in-differences test that isolates the effect ofadding schema to otherwise comparable pages.
Correlation is the likelier explanation: sites that implement schema well also tend to invest in content, press coverage and brand building, and those are the things doing the work. We follow the causal design. If a better-powered causal study reverses this, we will change the page and say so.
What we will not claim
- That we can guarantee a mention, a position or a ranking in any AI answer. The systems are non-deterministic and change without notice.
- That a proprietary score out of 100 means anything. We report citation share against a fixed, published prompt set.
- That AI search has replaced Google. It is roughly 1–6% of traffic today. It converts several times better and it is growing fast, and both of those things can be true at once.