In July 2026, the numbers behind AI search visibility finally exposed an uncomfortable truth: getting cited by AI once tells you almost nothing about whether you will be cited again. In the 2026 State of AI Search report from AirOps and Kevin Indig, the standout figure is brutal — only 30% of brands stay visible across back-to-back answers to the same question, and just 20% survive across five consecutive runs. Ask ChatGPT the identical query twice and there is a 70% chance the brand it named the first time is gone the second. AI citation is not a ranking you climb; it is closer to a lottery you re-enter every time.
What happened to this category over the past 12-18 months? "Get cited by AI" went from a slogan to a funded industry. ChatGPT reached 900 million weekly active users by February 2026, 2.25x the 400 million a year earlier, and by some measures accounts for the large majority of AI referral traffic. As spend chased that attention, vendors sold "AI visibility scores" as if they were as stable as Google rankings. The AirOps data says the ground underneath those scores is shaking: a single measurement is closer to a coin flip than a KPI.
Who is measuring, and where is the category heading? Semrush just shipped an expanded 2026 AI Visibility Index built on 126 million AI search prompts; Profound, Scrunch, and Otterly all sell cross-engine tracking. The entire category is converging on one realization — you cannot manage AI visibility from one snapshot, you have to measure a distribution over many runs. That reframes the whole problem from "am I cited" to "how often, how stably, and where in the answer."
Does this hit SMBs? Directly — and it is good news disguised as bad. Below we lay out the full numbers, what each of three reader types should do this week, a comparison of AI-visibility tools, the two things vendors will not tell you about volatility, and a no-SaaS way to measure your own citation stability with a spreadsheet and GA4.
The Detail: Volatility Is the Metric
The core AirOps numbers: across repeated queries, 30% of brands persist from one answer to the immediate next, and 20% stay present across five runs. A second finding reframes on-page tactics: 44.2% of all LLM citations are pulled from the first 30% of a page, so where a claim sits on the page materially changes whether it is quoted. And the academic backbone, the Princeton / Georgia Tech / IIT Delhi GEO paper, showed visibility in generative answers can rise up to 40% by adding cited sources, statistics, and direct quotations. Read together: AI does not reward a fixed "authority score" — it reassembles an answer each run, favoring content that is quotable, front-loaded, and evidence-dense. Who benefits? Brands that publish specific, citable facts high on the page. Who loses? Brands that bought a one-time "AI optimization" and a single flattering screenshot.
Immediate Actions for Three Reader Types
For brand owners / SMB founders:
- Stop trusting any single "we got you cited by AI" screenshot. Ask the vendor for the same query run 5-10 times and the percentage of runs you appeared in — that is the only honest number.
- Budget for consistency, not a one-off spike: publishing citable statistics and getting mentioned across third-party sites is what raises the odds on every re-roll.
For marketers / SEO practitioners:
- Front-load the quotable claim. With 44.2% of citations from the first 30% of the page, move your key statistic, definition, or answer into the opening, not the conclusion.
- Track a distribution, not a rank: log appearances across many runs and report an "appearance rate," the way you would report conversion rate, not position #3.
For developers / agencies:
- Get clients' structured data right (Organization, Product, FAQ, Article) so engines can parse quotable facts cleanly.
- Build a lightweight sampler: a scheduled job that fires a fixed prompt set at the engines N times and stores appearance rate over time — a deliverable monthly report, not a single screenshot.
SaaS Tool Comparison
| Tool | Positioning | Price tier | Best for |
|---|---|---|---|
| Profound | Enterprise AI visibility / AEO analytics, cross-engine | High (enterprise monthly) | Mid-large brands with a marketing team |
| Semrush AI Visibility (Index) | Large-scale prompt index + brand tracking | Mid-high (add-on to Semrush) | Teams already on Semrush wanting scale data |
| Scrunch AI | AI search monitoring + citation tracking with dashboards | Mid-high | Brands needing repeatable run-based reporting |
| Otterly.AI | Lightweight AI mention / ranking monitoring | Mid (SMB-friendly) | Small teams starting AI tracking |
| DIY (scripted prompts + GA4 + sheet) | Self-built appearance-rate sampler | Low (labor only) | Budget teams willing to invest hours |
What They Won’t Tell You
- A visibility "score" from one run is marketing, not measurement. If 70% of brands vanish between two consecutive answers, any tool showing you a single number is hiding the variance. The honest unit is appearance rate over N runs with a date range; anything else is a screenshot dressed as a KPI.
- Volatility means you can win without "authority," and lose despite it. Because each answer is re-assembled, a small brand with one sharply-quotable, front-loaded stat can surface on a given run over a bigger competitor — and the reverse also happens. That cuts both ways: exciting for challengers, unnerving for incumbents, and fatal to the idea that you buy a score once and keep it.
A No-SaaS SMB Alternative
- Fix a list of 20-40 buyer queries you want to be found for, and run each one 5-10 times per month through ChatGPT, Perplexity, Claude, and Google AI Mode.
- In a spreadsheet, log 1/0 for "did we appear" per run, then compute an appearance rate per query — that single percentage is your real AI-visibility metric.
- Rewrite your top pages to front-load a citable statistic or definition in the first 30% of the content, then re-sample and watch the rate move.
- Use GA4 referral reports to track real traffic from chatgpt.com and perplexity.ai alongside the appearance rate, so you connect citations to actual visits.
FAQ
Why did the brand AI mentioned yesterday disappear today?
Because AI re-assembles each answer rather than serving a fixed ranking. AirOps found only 30% of brands persist between back-to-back answers. It is expected variance, not a penalty — which is why you measure appearance rate over many runs, not a single result.
What is a good "appearance rate"?
There is no universal benchmark yet, so set your own baseline: sample a query 10 times, record how often you appear, and treat improvements against your own prior month as the win. Consistency across runs matters more than one lucky hit.
Does on-page placement really change citation odds?
Yes. The 2026 data shows 44.2% of LLM citations come from the first 30% of a page, so moving your key statistic or answer to the opening measurably raises the chance it gets quoted.
Do small brands stand a chance against big ones?
More than in classic SEO. Because answers are re-assembled per run, a sharply-quotable, front-loaded fact can surface a small brand on a given run. Volatility is a door for challengers as much as a risk for incumbents.
My Take
The mainstream sells "AI visibility scores" as the new search rankings. I read the AirOps data the opposite way: there is no stable score to sell. When 70% of brands vanish between two consecutive answers, a single number is not a metric, it is a screenshot — and the whole business model of "we raised your AI score" is measuring noise. What actually exists is a distribution: appearance rate over many runs, plus front-loaded, quotable content that improves the odds on each re-roll. For a shop like ScriptWalker the implication is direct: do not resell someone else's opaque "score." Sell two concrete, verifiable things instead — a structured-data + front-loaded-content build, and a monthly appearance-rate report that samples each query 10 times so the client sees a real percentage move. When everyone else is selling a lucky screenshot, the one who sells a measured distribution is the one who keeps the retainer.
Sources
- AirOps × Kevin Indig — The 2026 State of AI Search (first-party report: 30% / 20% / 44.2%)
- Semrush — Expanded 2026 AI Visibility Index (126M prompts) (first-party press release)
- Aggarwal et al. — GEO: Generative Engine Optimization (Princeton / Georgia Tech / IIT Delhi) (first-party research)
- Google — Structured Data documentation (official authority)
- Google — GA4 Analytics (official product, referral tracking)