Methodology · v2026.1
Version 2026.1 · last updated June 25, 2026 · stamped on every measurement
We measure Discovery Visibility — whether AI assistants name a company in their answers to name-free buyer questions (for example, "best business insurance provider for mid-size companies"). We report this separately from Branded Accuracy (questions that already name the company, which AI answers trivially), so the headline number reflects genuine discovery rather than a brand lookup.
We measure four assistants: ChatGPT (OpenAI), Claude (Anthropic), Google Gemini, and Perplexity. Every engine's results are reported separately — the engines differ dramatically for the same brand and questions, so a blended number would hide the finding. For signed engagements, the order form names the engine subset that contractual acceptance criteria attach to.
Each engagement fixes a set of buyer questions a real prospect would ask before choosing a provider — phrased without naming any company. Questions are agreed with the Client and frozen for the measurement window so results are comparable across runs.
Every question is run against all four measured assistants — ChatGPT, Claude, Google Gemini, and Perplexity (contract-grade engagements add repeated samples per question). A question is counted as present when the company is surfaced on at least one measured engine — and because engines differ dramatically, every engine's own result is reported alongside, never blended away. Detection is deliberately conservative: a borderline or ambiguous mention is not counted as present.
Measurement uses provider APIs and logged sessions — OpenAI, Anthropic, Google, and Perplexity channels. These approximate, but are not identical to, the consumer apps, whose answers vary with personalization, memory, model version, and time. Results are a consistent, reproducible proxy — not a claim about any one user's app output on a given day.
Every raw AI answer is stored with a timestamp in a tamper-evident, append-only, hash-chained audit trail. Each reported result is traceable to the exact captured answer, and the chain can be re-verified independently — so any number can be tied back to its evidence.
NextAIForge, LLC is independent. We are not affiliated with, endorsed by, or sponsored byany insurance carrier, third-party administrator, or AI provider named or measured. Company, product, and engine names are trademarks of their respective owners and are used only to describe what was measured. A measurement describes an AI system's behavior on a stated date — it is not a statement about the quality of any company.
Measurements reflect specific prompts captured on stated dates and can change as AI systems change. If you believe a result is inaccurate, or you want the underlying evidence for a claim about your company, email info@nextaiforge.com and we will review it against the captured audit trail.
This methodology is version 2026.1. Each measurement is stamped with the methodology version in force when it was captured. Material changes increment the version; prior results remain valid under the version under which they were measured.
See the remedy & guarantee terms for how this measurement supports each product's acceptance criteria and deliverable remedy. This page is informational and is not legal advice.