A new analysis argues AI visibility budgets should be defended by testing dashboard scores against real customer behavior. ChatGPT's referral traffic remains a fraction of Google's organic traffic, which reframes how marketers should measure AI exposure.
ChatGPT sends roughly 0.62% of the organic traffic Google does, and that number is the whole argument for rethinking how AI visibility gets measured. Search Engine Journal’s analysis of four datasets on AI visibility makes a case that the category is not a sham, but that most of the ways people measure it are.
What the analysis actually claims
The piece frames AI visibility as a budget question. Teams are being asked to fund work aimed at getting cited or surfaced inside AI answers, and those budgets need to survive scrutiny. The recommendation is to test dashboard scores against customer behavior rather than treating a vendor’s visibility number as ground truth.
It breaks the measurement problem into three things worth tracking: exposure, recall, and consideration. Exposure is whether your brand shows up in AI-generated answers at all. Recall is whether people remember seeing you there. Consideration is whether that exposure moves anyone toward a decision. Those are different questions, and a single composite score from a monitoring tool answers none of them cleanly.
The 0.62% figure is the anchor. ChatGPT’s referral traffic to websites is small relative to Google organic, which is the honest context for anyone pitching an AI visibility retainer right now.
Why the traffic number cuts both ways
Read the 0.62% as a reason to ignore AI visibility and you’ll probably be wrong within a year or two. Read it as a reason to spend like AI traffic is already material and you’ll be wrong today. Both errors are common, and they tend to come from the same place: treating a single number as a verdict instead of a snapshot.
Here’s the part that matters for site owners. Referral traffic is the easiest AI metric to capture and the least representative of AI’s actual influence. Someone who reads an AI-generated answer, absorbs a brand name, and then types that brand into Google or goes direct never shows up in a ChatGPT referral report. The click is attributed to organic or direct. The influence came from the AI answer. That gap is structural, not a temporary measurement bug, and no dashboard is going to close it by counting sessions.
So the 0.62% is real and it is also a floor on influence, not a ceiling. Anyone quoting it as proof that AI visibility is overhyped is quoting a traffic metric to answer a brand-awareness question.
The dashboard problem is a procurement problem
AI visibility tools have proliferated because the underlying task is easy to automate: run a set of prompts against several models, check whether a brand appears, score it, chart it over time. The output looks like a rank tracker, which is exactly why it gets bought. It’s familiar, it’s quantifiable, and it produces a number that goes in a slide.
The trouble is that prompt-based scoring measures what a model says when asked a synthetic question, not what a customer experiences when they ask a real one. Prompt sets drift. Models change. A score that moves from 34 to 41 between months may reflect a model update, a prompt list edit, or noise, and the dashboard won’t tell you which.
That’s why the exposure-recall-consideration framing is more useful than it first appears. It forces the conversation away from “what’s our AI visibility score” and toward “did anyone see us, did it register, and did it change anything.” Those are harder to measure and much harder to fake. They also map to budget conversations that finance teams actually understand.
What to watch next
Two things will decide whether this category matures or collapses into another line item nobody can justify. First, whether AI platforms start exposing impression-level data to publishers and brands the way search consoles do. Without that, everyone is inferring from referral clicks and synthetic prompts, and the inference will stay rough. Second, whether the tooling vendors start reporting recall and consideration alongside exposure, or keep selling a single score because it’s easier to demo.
Until then, the practical move for most teams is unglamorous. Keep tracking AI referral traffic because it’s free and real. Run periodic manual checks on the prompts that actually match your buying journey rather than a vendor’s generic set. And when someone asks you to defend an AI visibility budget, lead with what customers do after they encounter the brand in an AI answer, not with the score on the dashboard. The score is the input. The behavior is the evidence.
The 0.62% will move. What won’t move quickly is the measurement gap between what AI answers influence and what analytics can see. Plan around the gap, not the number.
Source: searchenginejournal.com