AI discovery
What does AI visibility actually measure?
Keep recommendations, citations, crawler requests, referral visits, and revenue as separate observed signals.
Learn
These guides explain what to calculate, which systems supply the evidence, and where the answer must remain unknown. No magic score, benchmark theatre, or token total pretending to be profit.
First principles
AI discovery
Keep recommendations, citations, crawler requests, referral visits, and revenue as separate observed signals.
Decision guide
Use traces to debug model behavior and customer economics to debug the business.
Define the unit, the revenue, the variable model cost, and the contribution profit.
Keep the numerator, denominator, time window, and customer population aligned.
Price cached and uncached tokens, failures, retries, and streaming usage correctly.
Cost needs observed revenue and a trustworthy identity link before it becomes customer profit.
Use each metric for the decision it can actually support.
Join payer identity and model usage without spreading unknown cost across every account.
Server recipe
Measure final streamed usage by account without collecting prompts or responses.
Server recipe
Keep cache reads, five-minute writes, and one-hour writes separate by customer.
Methodology
Audit the effective price, billing mode, and fail-closed boundary behind every cost.
Revenue recipe
Join Stripe customer revenue to AI usage through a stable internal account—not email.
Revenue recipe
Connect store-scoped Lemon orders, refunds, and AI usage without guessing identity.
Practical methodology
Reconcile OpenAI usage and Stripe revenue by customer with tested SQL. Keep refunds, mixed currencies, duplicate joins and unknown costs explicit.
Practical methodology
Verify browser visits, server AI usage, payer identity, and data boundaries with a practical GOD SDK checklist. GA4 is optional, not a prerequisite.
Prefer the live formula?