Methodology

Rankry uses a blind testing methodology designed to produce honest, unbiased data about how AI models perceive and recommend brands.

Blind Testing

We never ask AI "What do you think about [your brand]?" — that would produce biased, predictable answers. Instead, we ask real buyer-intent questions the way your customers would: "What's the best project management tool for remote teams?" or "Compare CRM options for B2B startups." We then analyze whether your brand appears in the response naturally.

This blind approach means the data reflects reality. If ChatGPT recommends you when a user asks a genuine question, that's a real signal. If it doesn't, that's valuable too.

Semantic Core

For each project, we generate a semantic core of 100 unique prompts. These cover different stages of the buyer journey — from awareness ("what types of tools exist for X") to comparison ("best A vs B vs C") to decision ("which tool should I use for specific use case"). The prompts are tailored to your industry, your competitors, and your use cases.

Live Web Search

All AI models are tested with live web search enabled. This is critical because in the real world, users interact with AI that has access to current information. Testing without search would give you data about the model's training data, not about what your customers actually see.

Response Parsing

Every AI response is parsed through our proprietary extraction pipeline. We identify which brands are mentioned, in what order, and extract the specific claims AI makes about each brand — both positive ("reliable uptime," "great for beginners") and negative ("limited features," "expensive renewal"). This structured data feeds all four of our metrics.

Weekly Updates

Reports refresh every week via batch processing. This captures changes in AI behavior over time — model updates, new training data, shifts in recommendations. Your dashboard tracks these trends so you can see whether your AI visibility is improving, declining, or stable.

Updated on: 02/03/2026