AI visibility is often discussed as if it were a traditional rank-tracking problem: choose a keyword, record a position, repeat every day.
That mental model is convenient, but it misses the most useful question for many founders and marketing teams:
When a buyer asks an AI assistant for solutions in my category without naming my brand, does the assistant recommend me, recommend competitors, or skip me entirely?
That is organic AI visibility. It is closer to a structured recommendation audit than a search-engine position check.
This article lays out a practical way to measure it without pretending that probabilistic AI answers behave like a stable list of blue links. Start with the decision you want to understand
A useful prompt represents a real buying or evaluation decision. It should contain enough context for an assistant to make a recommendation, but it should not lead the model toward your brand.
This measures what the model says after you name the brand. It does not measure organic discovery.
This gives the assistant room to assemble its own shortlist. The resulting answer reveals which brands it associates with the need, how it describes them, and which criteria appear to drive the recommendation. Build a small prompt set around buyer intent
You do not need hundreds of prompts for an initial snapshot. Start with 10 to 20 prompts covering a few distinct intents: Category discovery: "What are the best tools for X?" Use-case fit: "What should a team use to solve Y?" Constraint-based evaluation: "Which options work for a small team, limited budget, regulated market, or technical buyer?" Comparison intent: "What are the strongest alternatives for this type of workflow?" Implementation intent: "How should a company set up or improve this process?"
Keep the market, language, and audience consistent. Record the exact prompts. If you later repeat the exercise, methodological consistency matters more than adding a large number of loosely related questions. Classify the answer before interpreting it
For each answer, capture a small set of observable facts: Was your brand recommended? Which competitors were recommended? Where did each brand appear in the answer? What reasons, use cases, or differentiators were attached to each brand? Were sources or citations provided? Did the answer express uncertainty or qualify the recommendation?
A simple classification works well: Recommended: your brand appears as a genuine solution to the unnamed need. Mentioned: your brand appears, but not as a clear recommendation. Competitors only: the model recommends alternatives and excludes your brand. No usable recommendation: the answer is generic or avoids naming products.
This avoids a common measurement error: treating every textual mention as equivalent visibility. Separate presence from positioning
Suppose an assistant names your product but describes it as suitable only for enterprises, while your actual target customer is a five-person startup. That is visibility, but the positioning is inaccurate.
For every meaningful mention, record: the category the model places the product in; the audience it thinks the product serves; the strengths and limitations it repeats; the competitors it groups with the product; the evidence or sources behind those claims.
This turns a vague "AI mentioned us" observation into a concrete content and positioning backlog. Treat the result as a snapshot, not a permanent rank
Generative answers can vary because of model updates, sampling, prompt phrasing, geography, and changing source material. A single run should therefore be presented as a directional sample.
Avoid statements such as "we rank number three in ChatGPT." There may be no stable third position to own.