A framework is only as useful as its measurement layer. The AHQG matrix gives you a map. This part gives you instruments.
Before getting into the how, I want to restate something about EdgeShaping — because the framing matters. EdgeShaping is not a tool for measuring AI bot traffic. That’s a byproduct. What it’s actually measuring is something closer to AI comprehension: the traces left behind when an AI system attempts to understand your content.
That distinction changes how you use the data.
Not traffic. Comprehension.
When an AI bot visits your site, the standard instinct is to log it as traffic — another row in a table, filtered out or ignored. But the access pattern itself carries signal.
AI systems don’t store pages. They compress them. A crawler retrieves your HTML, extracts meaning, and integrates a summarized representation into a model’s internal state. Which means the question isn’t just “did AI visit this page?” It’s “what did AI walk away understanding about it?”
To write content that survives compression — content whose core meaning doesn’t degrade when summarized — you need to understand how that compression works. And to measure whether it worked, you need logs that capture what AI actually accessed, in what sequence, at what frequency.
That’s what EdgeShaping tracks. Not visits. Not traffic.
The log is a record of AI comprehension.
What 404s are telling you.
Here’s a signal most analytics setups discard entirely: 404 errors from AI crawlers.
When a human lands on a 404, it usually means a broken link or a mistyped URL. When an AI crawler generates a 404, the interpretation is different. AI doesn’t follow links randomly. It constructs URLs based on what it expects to exist — what it has inferred should be there, given the content it has already processed.
A 404 from an AI crawler is a record of inference. It’s the system saying: “based on what I’ve understood about this site, a page at this path should exist.” That inference failed — but the path itself is information.
In Part 1, I described how LLMs generate answers even when no answer exists — the drive to respond producing outputs that sound correct but aren’t grounded. The same mechanism, operating at the retrieval layer, produces 404s on pages that don’t exist yet. These are traces of what AI needed but couldn’t find. Read correctly, they’re a map of latent demand: questions humans haven’t yet articulated, surfaced through the behavior of systems trying to answer them.
User-triggered retrieval: the human signal in AI traffic.
Not all AI access is the same. EdgeShaping distinguishes between two primary modes.
Learning crawlers — GPTBot, ClaudeBot, Meta-ExternalAgent — operate on their own schedules, building and refreshing training data. Their access patterns reflect institutional priorities: what these companies have decided is worth learning from.
User-triggered retrieval — ChatGPT-User, Claude-User, and similar agents — is different. These requests happen because a specific human asked a specific question, right now, and an AI system decided your content was relevant to the answer.
That’s a different kind of signal. It’s not “AI thinks your content is worth indexing.” It’s “a human asked something, AI looked here.” The user-triggered log is the closest thing we currently have to measuring AI-mediated referral traffic — a channel that GA4 cannot see.
When you overlay GA4 session data with EdgeShaping’s user-triggered access log, a more complete picture of your traffic ecosystem emerges. Some users arrived directly. Some arrived through search. Some arrived through AI — and GA4 recorded none of it.
Query Phase Analytics: AI retrieval as a leading indicator.
This is where the measurement layer connects back to the AHQG matrix.
Latent Gap content — high AI retrieval, low human search demand — tends to migrate toward Aligned over time. Human demand surfaces. The topic enters mainstream search. The question is: how much lead time do you have?
AI retrieval frequency, tracked over time, functions as a leading indicator of that migration. When EdgeShaping starts showing elevated access to a piece of content — before any corresponding movement in GSC query volume — that’s a signal worth paying attention to.
I call this Query Phase Analytics: treating AI access logs as a proxy for the pre-search phase of demand formation. The idea draws on the same logic as media mix modeling’s adstock concept — effects that lag their causes by measurable intervals. AI retrieval may be to human search what early media exposure is to eventual purchase intent.
The lag is not fixed. It varies by domain, by topic maturity, by how quickly human vocabulary catches up to AI-surfaced concepts. If you know your industry’s cycles — how long it typically takes for a latent topic to become a searched one — you can use retrieval frequency as a timing instrument, not just a measurement.
Three layers. One picture.
The full measurement stack for AHQG looks like this:
EdgeShaping — AI retrieval frequency, user-triggered access, 404 inference patterns. The AI axis of the matrix.
GSC — Human search demand, query-level intent data, impression and click trends. The human axis of the matrix.
GA4 — Confirmed human sessions, on-site behavior, conversion. Ground truth for what humans did after arriving.
No single layer gives you the full picture. EdgeShaping without GSC tells you what AI is interested in, but not whether human demand exists. GSC without EdgeShaping tells you what humans are searching for, but not how AI is processing your content. GA4 without either tells you what happened, but not why, and not what’s coming.
Together, they map both sides of the gap.