If you’ve been working on your AI visibility for any length of time, you’ve probably noticed something unsettling: it doesn’t hold steady.
You show up in ChatGPT one week, then you don’t. You’re cited in Perplexity, then you’re gone. You check your Google AI Overview appearances and they’ve shifted again.
This isn’t a sign you’re doing something wrong. It’s how AI search works, and understanding why it fluctuates changes how you should approach building a presence in it.
The research confirms what practitioners already suspected.
In January 2026, Rand Fishkin and Patrick O’Donnell at SparkToro published the most rigorous study to date on AI recommendation consistency. They ran 600 volunteers through 2,961 prompts across ChatGPT, Claude, and Google AI, recording every response.
The finding was stark: there is less than a one in a hundred chance that ChatGPT or Google AI, asked the same question a hundred times, will give you the same list of brands in any two responses. For ranking order specifically, it drops to less than one in a thousand.
The SparkToro study describes AI tools as probability engines. They are not designed to give consistent answers. They are designed to generate statistically plausible answers, and that process produces something different almost every time.
Position Digital’s 2026 AI SEO statistics roundup confirms this independently: AI recommendations are highly inconsistent by design, with less than a one in one hundred chance of the same brand list appearing twice across repeated identical queries.
It’s not just inconsistency. The platforms behave differently from each other.
Even if you’ve cracked visibility on one AI platform, that doesn’t translate to another. A coach cited regularly by Perplexity may be completely absent from ChatGPT’s responses to an identical query. What works for Google AI Overviews has only a 13.7% citation overlap with AI Mode results, according to Ahrefs December 2025 research. The platforms are drawing from different sources, using different retrieval logic, and applying different weighting to what they consider authoritative.
ChatGPT currently dominates with roughly 79% of generative AI web traffic. But Goodie’s 2026 B2B AI Search Traffic Report shows that ChatGPT held 89% of B2B AI referrals in August 2025 and had dropped to 63% eight months later, with Claude growing from 1.4% to 18.5% in the same period. The market is fragmenting, and the platforms your clients use are shifting underneath you.
The Similarweb 2026 Generative AI Brand Visibility Index treats cross-platform consistency as a key signal of durable AI presence for exactly this reason: a brand that appears across all four major platforms has a fundamentally stronger position than one that dominates only one. Visibility is volatile even for established brands. Nike leads in fashion AI visibility but is on a negative trend. The brands gaining momentum are ones with broader, more distributed presence.
Why the volatility happens
Several things drive the instability, and they compound each other:
LLMs are probabilistic by design
The same prompt run twice doesn’t retrieve the same answer because the model is sampling from a probability distribution, not retrieving a fixed result. The temperature settings on most consumer-facing AI tools ensure variety in outputs. As I’ve written preciously on Medium, a coaching brand can get clicks and referrals from AI tools one week, go dry the next, then reappear.
Training data gets updated
The web that AI models trained on six months ago is different from the web they’re training on now. If a competitor publishes a strong series of well-cited posts, or earns coverage on a platform AI tools weight heavily, that can shift the model’s associations in a topic area without any change on your end.
The models themselves get updated
ChatGPT moved to GPT-5.3 Instant as its default model and immediately cited 20% fewer domains than before. GPT-5.4 changed fan-out search behaviour significantly. A model update can alter your visibility without warning, in either direction.
The source hierarchy shifts
Reddit dropped out of ChatGPT’s top ten cited sources entirely in late 2025 after a series of platform updates, then started creeping back. LinkedIn surged from the 11th most-cited source to roughly 5th between November 2025 and February 2026. Any visibility strategy built around a single source type is exposed to exactly this kind of shift.
Personalization layers on top of all of this
Depending on the user’s history, location, and browsing context, AI tools will increasingly return different answers to the same question from different people. This is only becoming more pronounced as personalization capabilities mature.
Your competition does not sleep
AI visibility is, just like SEO, somewhat a zero-sum game. If the coaches that compete with you produce content and gain mentions online, they are very likely to start getting cited instead of you. To make sure you’re doing everything right, check our AI visibility guide for coaches or just get in touch with us for a free consultation.
What this means for coaches specifically
For health coaches building an AI presence, the instability has a practical consequence.
If your entire visibility strategy depends on one type of content, one platform, or one type of citation source, you are exposed to every shift in how AI tools’ behavior.
A coach who has only a strong website but no press mentions, no YouTube presence, no LinkedIn activity, and no community presence in relevant health forums will have fragile AI visibility.
The moment the model’s weighting shifts, or a training data update reduces the prominence of website-only sources, that coach disappears from conversations.
Ahrefs’ December 2025 research found that YouTube brand mentions are the single strongest correlating factor with Google’s AI Overview visibility among all signals studied.
But YouTube citations barely overlap with what earns you Perplexity mentions, where community-based sources and live-retrieval results dominate.
And neither of those fully predicts Claude visibility, which according to Onely’s 2026 research prefers academic and E-E-A-T heavy content above almost everything else.
The SparkToro research also shows something relevant here: the size of the pool of potential recommendations in your niche directly affects how much variance you’ll see.
In a tight niche with few specialists, AI tools tend to be more consistent about who they recommend. In a broad niche with hundreds of options, the variance is much higher.
Health coaching sits somewhere in the middle, which means coaches with distributed, multi-source presence are consistently more likely to appear than those relying on a narrow footprint.
The practical response: spread far and wide
The answer to AI visibility instability is not obsessing over any one channel or trying to rank in a specific AI tool. Tracking your AI visibility rigorously is not the answer, either.
It’s building a presence across enough surfaces that you remain in the probability distribution regardless of which model update, source reweight, or platform shift happens next.
Stacker’s research found that distributing content to a wide range of publications can increase AI citations by up to 325% compared to publishing only on your own site.
That’s the clearest empirical case for breadth over depth in AI visibility strategy.
What this looks like in practice for a health coach:
- consistent cornerstone content on your own site,
- regular off-site placements on health and wellness publications,
- a YouTube or podcast presence that generates transcripts AI tools can index,
- LinkedIn activity since LinkedIn is now a top-five citation source in ChatGPT,
- community presence in relevant forums and health discussion spaces,
- press coverage and directory listings,
- and a clearly defined entity with consistent name and niche description across all platforms.
None of this alone guarantees stable visibility. The SparkToro research is honest about this: AI recommendations are probability engines, and you’re working to increase your probability of appearing across a broader range of queries and contexts, not to lock down a position.
But as the research concludes, visibility percentage across many prompts run many times is a reasonable metric. The coaches with the broadest, most consistent footprint show the highest and most durable visibility percentages.
The right mental model is less “how do I rank in AI” and more “how do I make sure my name is deeply embedded in the fabric of my topic area online.”
The fluctuation doesn’t go away. But it stops threatening coaches who have done the work across enough surfaces to weather any single shift.
If you want to understand how distributed your current AI footprint is and where the gaps are, get in touch.
Leave a Reply