Answer Engine Visibility, and Being Straight About What It Can Be Measured To

Challenge

A client wanted to know whether the AI answer engines were sending them anything, and whether it was worth anything. Neither question has a native answer. Analytics platforms have no channel for AI traffic, so it scatters across direct, referral and unassigned, and the honest state of most reporting on this subject is that nobody actually knows.

Solution

The visibility work itself was ordinary disciplined technical search work. Crawl and indexation faults corrected, on-page structure fixed, accurate structured data deployed across service and location templates. No AI-specific trick was involved, and we think that is the most useful thing about the case: what the answer engines reward is a site they can read cleanly.

The measurement was the build: a Blue Glass Insights first-party-built tool. We read the referrer and the landing page parameters on every page load, match them against a maintained list of answer-engine hostnames, and write the result as a first-party value we own, registered as a custom dimension and grouped into its own channel so it appears in standard reporting instead of being reconstructed by hand each month. Separately, we sample the engines directly against a defined question set to measure how often a client is named in the answer and how often they are actually cited with a link. Those are not the same number, and the gap between them is usually where the work is.

The limits, stated plainly, because a measurement you cannot bound is not a measurement. Some AI traffic arrives with no referrer at all and cannot be attributed by anyone, ourselves included. Someone who reads an answer today and searches the brand next week converts as organic search, and no available method separates that from ordinary demand. We label the first as unattributable and the second as inference rather than counting either as a win. Clients are better served knowing which parts of a number are solid.

Result

The client now has a report of actual AI-to-sales conversion, including a $45,000 order tracked through in the analytics, not reconstructed afterward from a customer’s recollection.