●THE AI CITATION STUDY · FINAL REPORT

Why AI cites a company, keeps citing it, and says its name.

Most AI-visibility studies check once and count who got cited. We did the opposite. For twelve weeks we tracked ten B2B domains across all four major AI engines — 1,431 buyer questions, 59,645 stored answers, two famous publishers held back as untreated controls — and took every citation apart week by week: what got cited, what was different about it, and why it held, dropped, or came back. The full report is ready.

The finding that changed how we think about AI search: a citation isn’t a lottery ticket. It’s a durable, measurable property of your domain — and you can engineer it.

Get the full report

Enter your email and we’ll send the complete study — all 28 figures and the methodology behind them — straight to your inbox.

How to get content cited by AI Search
How to know if AI bots are looking at my content
What are the gaps in my content
How to create a process to maintain AI Visibility

No spam. Just the report, plus the occasional finding as the next study runs. Unsubscribe anytime.

●HOW IT WAS RUN

Other studies take a snapshot. We ran a twelve-week controlled study.

Ten company domains. Four AI engines — Google AI Overviews, ChatGPT, Claude, and Perplexity. 1,431 tracked buyer questions, re-checked every week for twelve weeks, with 59,645 raw answers stored. Eight domains ran the VizzEx approach; two of the most famous names in marketing were held back, untreated, as controls. Every citation was joined to the exact post that earned it — so we could see not just who got cited, but why.

Hundreds of studies count citations. None could say why a company got cited, why it stayed, why it came back, or why the answer said its name. This one was built to.

59,645
Stored AI answers
1,431
Tracked buyer questions
4
AI engines, re-checked weekly
12
Weeks · 10 domains, 2 untreated controls
●WHAT THE DATA SHOWS

The results overturn the standard playbook.

These patterns held across engines and across checkpoints. The report shows the data behind each one — and what to do about it.

FINDING 01

Citations hold — and they come back.

Between 39% and 58% of the citations these companies won held four weeks or more. When one was lost, it returned 73% of the time — whether the post was edited or left untouched. AI visibility behaves less like weather and more like an asset.

FINDING 02

What keeps you cited is measurable.

Citations that held shared a short, measurable list of traits — connected posts answering the question, other engines citing you, FAQ sections, semantic links. We rank every one, with the lift it carried. Chasing high-volume questions, it turns out, worked against you.

FINDING 03

The domain keeps the win.

When one post stopped being cited, another from the same company usually took its place — 79% to 94% of the time on questions a site could answer from several posts. The unit that wins AI visibility isn’t a page. It’s your domain.

FINDING 04

Cited isn’t the same as named.

Across the study, companies were cited in 7,304 answers and named in just 337 of them. Being used and being credited have different causes — and the report shows exactly what moves each, and where to put your name so the answer says it.

●WHAT’S INSIDE

28 figures. One operating picture of AI visibility.

  • The traits that predicted a citation holding — ranked, with the lift each one carried, measured question by question.
  • Why structure wins head-to-head — treated posts out-structured the page they were up against in 89% of 1,163 comparisons; the famous controls, far less often.
  • What the losses that came back had in common — turning ‘volatility’ into something you can engineer.
  • Who takes the answer when you don’t — the pages AI cites instead, and which losses are actually worth chasing.
  • How Google’s updates moved citations — week by week, which losses returned after each update and which didn’t.
  • Cited vs. named — the 337-of-7,304 gap, and where to place your brand so the answer credits you by name.
  • The full methodology — every engine, schedule, control, and the 59,645 stored answers, so you can judge the findings and measure your own content the same way.
●A LOOK INSIDE

A glimpse of what lands in your inbox.

Twenty-eight figures, built from 59,645 stored answers. A few of them:

Key findings: 39 to 58 percent of won citations held four or more weeks, 73 percent of lost citations came back, 79 to 94 percent of losses stayed with the same site, 337 of 7,304 cited answers named the company.
Bar chart ranking what predicted a citation holding: connected posts, cross-engine citation, FAQ sections, being named, semantic links, structural symmetry.

What predicted a citation holding — ranked.

Chart showing treated posts outscored the competing page on structural symmetry in most head-to-head comparisons.

Treated posts out-structured rivals in 89% of match-ups.

Bar chart of won positions that held four or more weeks, broken out by AI engine.

Won positions that held four weeks, by engine.

Bar chart of lost citations that returned, broken out by AI engine.

Lost citations that came back, by engine.

●WHO IT’S FOR

Written for B2B marketing leaders and the teams who run their sites — anyone deciding what to publish, what to fix, and what to leave alone as they engineer their AI visibility.

Get the report while the findings are fresh.

The next study accepts new participant sites. This report is where you decide whether you want your content measured the way these ten were.