AI Rankings Don't Exist. Here's What Your Business Should Actually Track.

A new category of marketing tool wants to sell you your "AI ranking position." There is a problem with this: the number changes almost every time someone runs the same prompt. What the research actually shows is more useful and more familiar than a ranking. It looks like brand building.

Rand Fishkin and Patrick O'Donnell at SparkToro published the most rigorous test of AI brand recommendations to date in January 2026. Their team ran 2,961 prompts across ChatGPT, Claude, and Google AI Overviews. Six hundred volunteers. Twelve different recommendation prompts, from chef's knives to cancer care hospitals.

The core finding: less than a 1-in-100 chance that two identical prompts return the same list of brands. For position order specifically, less than 1 in 1,000. The AI does not rank brands. It draws from a pool, differently, every time.

That single finding breaks the entire premise of a growing industry worth hundreds of millions of dollars.

The $745-Per-Month Problem

The AI visibility tracking market exploded through 2025 and into 2026. Tools like Scrunch AI ($300/month), Profound ($499/month), Ahrefs Brand Radar ($699/month), and Semrush's AI Toolkit ($745/month) now compete for agency budgets. The combined search volume around "AI visibility tools" exceeds 1,860 monthly searches at CPCs of $29 to $52.

What these tools report is your brand's position in one specific AI response at one specific moment. Run the same prompt the next morning and you might be second. Or absent. Or first. Fishkin put it plainly: "any tool that gives a 'ranking position in AI' is full of baloney."

This is not a small methodological quibble. These tools apply search engine logic to a system that does not work like a search engine. In traditional SEO, positions are relatively stable. You can track them weekly. You can correlate changes to actions. In AI recommendations, the position is essentially a random draw from a consideration pool. Recording one draw and calling it a trend is like tracking a single hand of poker and calling it a strategy.

The Consideration Pool Is Real

Here is the nuance from Fishkin's data that most commentary missed. While position is random, presence frequency is not.

Certain brands appeared consistently across hundreds of runs of the same prompt. City of Hope hospital appeared in 97% of ChatGPT responses about West Coast cancer care. Sony and Bose appeared in 55 to 77% of headphone recommendation responses. They were not always first. But they were almost always present.

The AI maintains something that functions like a consideration set for each category: a pool of brands it draws from when generating responses. Whether your brand is in that pool depends on your brand's footprint across the sources AI models index. Position within the pool, on any given response, is noise.

Byron Sharp and Jenni Romaniuk at the Ehrenberg-Bass Institute described this exact mechanism in human buying behaviour. Sharp's research across 130+ brands in "How Brands Grow" found that brands grow by being easy to think of when a buying moment arises, not by being differentiated or deeply loved. He called this mental availability: a brand's propensity to come to mind across the range of situations that trigger a category purchase.

The AI consideration pool is a digital implementation of mental availability. The brands appearing consistently are not necessarily the highest-rated or most differentiated. They are the ones with the broadest, most consistent presence across the sources AI models treat as authoritative: review platforms, industry directories, trade press, and web content that earns reference from others.

Why Breadth Beats Depth

Romaniuk's work on Category Entry Points (CEPs) explains why some brands appear across a wider range of AI prompts than others. CEPs are the specific needs, occasions, and questions that trigger a purchase decision. A trade business that appears across multiple contexts, emergency repairs, scheduled maintenance, insurance work, commercial fit-outs, builds a broader footprint than one optimising for a single phrase.

For AI visibility, this translates directly. The models pattern-match across a range of contexts. A brand linked to multiple relevant problems gets drawn into more response pools. A brand tightly associated with one keyword gets drawn into fewer.

Les Binet and Peter Field's research on advertising effectiveness adds the investment lens. Their analysis of the IPA Databank, covering thousands of campaigns, found that brand-building activity (broad reach, emotional, memory-building) drives long-term market share growth, while activation activity (targeted, rational, response-driven) drives short-term sales. The optimal split for most categories sits around 60% brand to 40% activation.

AI visibility sits squarely on the brand-building side of this equation. You cannot activate your way into an AI consideration pool. There is no prompt to bid on, no keyword to buy. The brands that appear consistently got there through cumulative brand presence built over time. Treating AI visibility as a performance channel, something to optimise week by week with tracking tools, applies activation thinking to a brand problem.

This is the same mistake Binet and Field have documented across traditional media for a decade: businesses chasing short-term metrics at the expense of the long-term brand building that actually drives growth. Understanding why measuring every marketing dollar can limit growth matters here.

What Lily Ray's Warning Means for Your Strategy

Lily Ray, one of the most respected voices in SEO, delivered a pointed warning at Affiliate Summit West 2026: several popular "generative engine optimisation" tactics are already being treated as spam by Google and Microsoft. Self-promotional listicles, comparison pages designed to game AI citations, and prompt injection techniques are being flagged.

Her broader point is more important than the tactical warning. AI models trust corroboration. They draw from sources that other credible sources also reference. Third-party reputation, not self-published content volume, is what drives citation. A mention in a trade publication carries more weight than ten blog posts on your own site.

This aligns with what the data shows about which brands consistently appear in AI recommendations. They are not the ones who "optimised for ChatGPT." They are the ones who built genuine brand presence across the web over years.

What that looks like in practice:

The Foundation Still Matters More

Google AI Overviews now appear on roughly 13 to 26% of US searches depending on the data source, with informational queries triggering them 39% of the time. That is significant. But Google Search still drives the vast majority of website traffic for most Australian SMEs.

AI Overviews pull predominantly from top-ranking organic pages. Strong traditional SEO directly feeds AI Overview inclusion. Your Google Ads capture intent that already exists and produce leads with a measurable cost per acquisition. AI visibility is worth building toward, but it sits on top of a functioning search and paid media foundation, not instead of one.

If someone is suggesting you redirect your Google Ads or SEO budget toward AI visibility tools, they are getting significantly ahead of the evidence. The right approach: continue what produces measurable results through search and paid media. Build the brand presence that feeds both traditional and AI visibility simultaneously. Track AI mention frequency as a secondary signal, not a primary KPI.

How to Measure This Yourself

You do not need a $500-per-month tool. Take five prompts an Australian prospect might use to find a business in your category. Run each one 20 times in ChatGPT. Record how often your brand appears across those 100 total responses. That is your presence frequency baseline.

Check it quarterly. If that number is growing, something in your brand building is working. If it is flat despite investment, you have a brand footprint problem that no ranking tool will fix. If it is zero, your first priority is building the review coverage, credible citations, and clear positioning described above, not subscribing to a tracking dashboard.

The AI visibility game is a brand game. Anyone selling it as a ranking game is applying a framework from a channel that worked differently to one that does not work that way at all. The mechanism that builds AI consideration set inclusion is the same one Sharp described in 2010: be easy to think of, across many relevant contexts, consistently over time. The channel changed. The mechanism did not.

Further Reading

• SparkToro: AIs Are Highly Inconsistent When Recommending Brands - Fishkin's full research methodology and data
• Lily Ray: The State of AI and SEO in 2026 - Why GEO tactics are being flagged as spam
• Google AI Overview Statistics 2026 - Current data on AI Overview reach and traffic impact
• Search Engine Journal: AI Recommendations Change With Nearly Every Query - Industry analysis of SparkToro's findings
• Best AI Visibility Tools Compared - Current market pricing and capabilities