The methodology, and why it is published
No vendor names in the prompts
The vendor list in every index is an output of what the engines said, never an input to what we asked. No prompt contains a company, product, or brand name. This is what separates a ranking from a survey of names we already had in mind.
Personas, not averages
A single ranking for a whole category describes nobody. An aerospace supplier under AS9100 and a fifty-person hardware startup ask about the same software and get entirely different answers back. We build eight to ten buyer personas per category and run a full prompt set against each, then publish the segment leaderboards alongside the overall one. The gaps between them are usually the most useful part.
Every engine reported separately
The major answer engines disagree with each other more than they agree. In the MES index, one product appeared in 12.5 percent of Perplexity prompts and zero percent of ChatGPT prompts. Compositing hides that, so every table reports each engine on its own alongside the combined score.
Source analysis reported
Being named and being cited are different things. We capture every source URL behind every answer and classify it, which makes it possible to tell whether a company is visible because the model learned it or because content exists and gets retrieved.
What it measures
Whether AI answer engines name a product, how often, on which engine, and to which buyer. Every ranking here is built from what the engines actually said in response to a fixed set of buying questions.
What it doesn't measure
Product quality, competitive position, revenue, or market share. A vendor can be widely deployed and rarely named, and the reverse is equally possible. In the MES index, one vendor competing actively for real deals appeared in 2 of 112 prompts. Read every placement here as a statement about AI visibility, and treat questions of fit and quality as separate work.