AI recommendations are probabilistic. A single prompt run once gives you noise. The same prompt run many times across many engines gives you signal. The AI Appearance Rate is a frequency-based measurement framework built on that insight — designed to surface who AI actually trusts in a category, not who got lucky in a single response.
AI engines are non-deterministic systems. The same prompt returns different outputs on every run. Position tracking is meaningless in this environment. Frequency across many runs is the only measurement that produces stable, actionable data.
When someone asks ChatGPT which provider to recommend, the model does not return a ranked list from a deterministic index. It generates a response influenced by training data, retrieval patterns, and probabilistic sampling. Ask the same question five minutes later and you get a different response.
Rand Fishkin's study of 2,961 prompt tests across four AI engines found that only around 30% of brands appearing in an AI response appear again on the next run of the same prompt. Run it five times consecutively and just 20% of brands persist across all five.
This is the nature of the system. Large language models are not search indexes. They do not have fixed rankings — what they have is probability distributions over possible outputs, shaped by the signals they have encountered across training and retrieval.
The implication is direct. Any tool reporting your brand's position on an AI engine is reporting a single sample from a non-deterministic distribution. That number has no predictive value, no stability, and no strategic utility. It is noise presented as signal.
The AI Appearance Rate solves this by replacing position with frequency. Instead of asking where a brand appears in one response, it asks how often a brand appears across many responses. That frequency is stable, comparable, and it reveals something position tracking never could.
After 5 runs, Brand A is clearly your real competitor. This is because AI reaches for it every single time. That is what dominant independent consensus looks like from the inside of a language model.
Total evaluations = Prompts × Engines × Runs
The more runs per prompt, the more statistically meaningful the data. A single run gives you a snapshot. Three runs gives you a direction. Five or more runs begins to reveal the dominant players — the brands AI reaches for consistently rather than randomly.
The higher the number of runs, the clearer the competitive picture. Brands appearing in 1 out of 5 runs are incidental. Brands appearing in 4 or 5 out of 5 runs have built enough independent consensus that the model has resolved them as category leaders. That distinction is the most strategically valuable output the AAR produces.
The goal is not just to measure your score. It is to increase your appearances over time — to move from incidental to consistent to dominant across the prompts that matter most to your buyers.
Each AI engine has a different retrieval architecture and relies on different signals to form recommendations. A brand can dominate on ChatGPT and be completely invisible on Perplexity. This is mostly because the AI engines are drawing on different data sources, with different recency biases, serving different user behaviours.
The AAR is therefore not a single number. It is a per-engine score. And the most strategically important question it surfaces is not just where you stand overall — it is which engine your specific buyer persona uses most, and whether you are visible on that engine in particular.
The strategic implication: identify which engine your specific buyer persona uses most heavily. A B2B SaaS targeting developers should prioritise their AAR on Claude and Perplexity. A dental practice should prioritise Gemini and ChatGPT. Dominating the right engine for your buyer is more valuable than averaging across all four.
Measuring frequency across multiple runs and multiple engines produces something more valuable than a score. It produces a competitive map. The brands appearing consistently across runs are the ones AI has resolved as category leaders. They have built enough independent consensus — reviews, earned media, editorial coverage, peer mentions — that the model reaches for them reliably rather than randomly.
Your goal is to move through three stages over time.
The free scanner generates 20 buyer intent prompts for your category, runs them across ChatGPT, Perplexity, Gemini, and Claude, audits your technical signals across seven checks, and returns your AAR alongside a view of which competitors are appearing in the prompts where you are not.