A marketing manager opens their analytics on a Monday morning. Organic clicks are flat or down. Direct traffic is mysteriously up, including to deep, specific pages no one should be finding by typing a URL. Paid looks fine, but the overall story does not add up, and the report due to leadership is going to be hard to defend.
The data is not broken. The problem is that the model used to read it was built for an internet that no longer exists. A growing share of buyers now research inside AI tools like ChatGPT, Perplexity, Gemini, and Google's AI Overviews, then arrive on a website with no usable trace of where they came from. Standard analytics drops them into buckets like Direct and Unassigned, and last-click reporting gives the credit to whatever happened last instead of the AI conversation that actually drove the decision.
This is the AI attribution gap, and it is widening every quarter. Here is what is actually happening, why it matters for budget decisions, and the practical steps to take now, whether you market globally or in a specific market like Japan.
What the AI attribution gap actually is
When someone clicks a Google result, the browser usually passes a referrer so analytics knows where the visit came from. AI tools frequently break that chain. Many send no referrer at all, especially from mobile apps, and the ones that do are inconsistent.
The result is that genuinely valuable visits get misfiled. An analysis of more than 446,000 visits, reported by Clickport, found that roughly 70 percent of AI-referred traffic arrives without referrer headers and lands in the "Direct" bucket, the same place analytics dumps bookmarks and typed URLs. Separately, Workshop Digital's analysis of 181.6 million sessions found that a large share of ChatGPT and Perplexity sessions arrive with no classifiable medium, so they show up as "Unassigned," a label that tells you nothing.
The platforms behave very differently. Perplexity and desktop ChatGPT in search mode are relatively well-behaved and pass referrer data, and ChatGPT appends a utm_source tag to citation links. Most mobile apps strip the referrer entirely. That matters because the apps are enormous: ChatGPT alone has tens of millions of monthly mobile downloads, which means a large slice of its traffic is structurally invisible to any analytics tool.
Why this got worse in May 2026
Two recent changes pushed the problem from a slow leak to something marketers are noticing on their dashboards right now.
First, Google removed the method that let sites identify clicks coming from AI Overview results. As ROI Revolution reported, that signal disappeared in early May 2026 with no confirmed replacement, so for now AI Overview presence has to be tracked through rank-tracking tools rather than traffic analytics. Because a click inside an AI Overview carries a normal google.com referrer, it is essentially indistinguishable from a regular organic click. With AI Overviews appearing on a large and growing share of searches, that is a meaningful chunk of traffic where attribution is structurally impossible.
Second, the click-through math underneath rankings has shifted. Research cited in this week's marketing trend roundup from B2The7 found that pages holding top-three positions saw click-through-rate declines in the range of 18 to 34 percent once AI-generated answers appeared above them, even when rankings and impressions held steady. The ranking did not move. The clicks did. If your reporting only watches rankings, that loss is invisible until the traffic number drops.
There is a small piece of good news, announced just this week. Google has begun testing dedicated AI Search performance reports in Search Console, showing how often your pages appear inside AI Overviews and AI Mode. The catch, as Search Engine Journal notes, is that these reports show impressions but not clicks, they exclude Gemini, and they are rolling out to a subset of UK sites first. It is progress on visibility, not a fix for attribution.
Why this is a budget problem, not just a reporting one
It would be easy to file this under "analytics housekeeping." That underestimates it, because the traffic hiding in those buckets is not average traffic. It tends to convert better than almost anything else.
Multiple independent studies point the same direction, though the exact numbers vary by methodology and industry, so treat them as a consistent pattern rather than precise universal figures. Adobe Analytics tracked the 2025 holiday season and found AI referrals converting meaningfully better than non-AI sources, with a much larger lift on peak days. An analysis of 94 e-commerce stores reported by Search Engine Land found ChatGPT visitors converting roughly 31 percent higher than non-branded organic search, with higher revenue per session. Ahrefs published its own data showing AI search visitors made up a tiny fraction of traffic but drove a wildly disproportionate share of signups, an extreme case explained by the fact that people who reach an SEO tool through ChatGPT are already deep in buying mode.
The behavioral reason is simple. When someone arrives from a Google search, they are still in research mode. When someone arrives from an AI answer, the tool already did the research phase for them. They land pre-qualified, closer to a decision. The click is the end of the funnel, not the beginning.
Put those two facts together and the danger is clear. Your highest-intent visitors are the ones most likely to be filed as anonymous Direct traffic, and last-click reporting then hands the credit to a later branded search or a direct visit. As analyst Rand Fishkin has put it, attribution has not died, it just stopped telling the truth. Brands that make budget decisions on last-click data alone end up systematically underfunding the very channel that is quietly influencing the most valuable buyers.
What this means for marketers in Japan
The mechanics of the gap are global, but the mix of platforms is local, and Japan is a useful example. A meaningful share of search in Japan still runs through Yahoo! JAPAN alongside Google, each developing its own AI answer features, and generative AI adoption among consumers has been climbing fast. For brands targeting Japanese audiences, that means the invisible-traffic problem does not arrive only through Google and the global AI tools. It arrives through a more fragmented set of surfaces, which makes clean first-party tracking and consistent UTM discipline more important, not less.
The practical takeaway is the same everywhere, just with an extra reminder for multi-market teams: do not assume the channel mix that hides AI traffic in one market is identical in another, and make sure measurement is set up per audience rather than copied across them. We get into the broader Japan platform picture in our guide to paid advertising in Japan.
What to actually do now
You cannot fully close this gap, because some of the signal was never sent. But you can recover a large part of it and stop making decisions on numbers you know are incomplete. Here is the practical order of operations.
Start by sizing the problem honestly. Look at your Direct and Unassigned traffic over the last year. If Direct has been climbing without a matching rise in brand awareness, email volume, or app usage, a portion of that growth is almost certainly misfiled AI traffic. You are not trying to get an exact number. You are trying to confirm the gap is real for you, which it almost certainly is.
Build a custom AI channel in your analytics. In GA4 you can create a custom channel group that classifies sessions whose referrer or source matches known AI platforms such as chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com. Google's own documentation supports this approach. It only catches the roughly 30 percent of AI visits that arrive with referrer data, but that 30 percent is far better than treating all of it as generic Direct, and it gives you a real channel to watch over time.
Tighten UTMs and first-party tracking. Make sure every link you control that might be surfaced by an AI tool, citation, or partner carries clean, consistent UTM parameters. Where you can, capture referrer and landing data server-side, since server-side capture is more resilient than client-side tags that mobile apps and consent rejections quietly break.
Add a self-reported attribution question. The single most effective low-tech fix is asking "How did you hear about us?" on your lead form or at checkout, with an explicit option for AI assistants like ChatGPT. This catches influence that no referrer ever will, and it is one of the few ways to see the buyers who were persuaded inside an AI conversation but converted later through a branded search. Feed those answers into your CRM as an AI-influenced segment.
Shift the reporting conversation from last click to influence. When you present to leadership or clients, separate "what closed the sale" from "what influenced the decision." Treat brand visibility inside AI answers as a real performance signal alongside clicks and conversions, and use the new Search Console AI impression data, once it is available to you, as one input rather than the whole answer.
Make sure your brand is the source AI tools cite. The defensive work above recovers measurement. The offensive work is making sure you are actually present in those AI answers in the first place, which is the core of answer engine and generative engine optimization: clear, citable, well-structured content that AI systems can understand and recommend. Measurement tells you AI is sending high-intent buyers. Visibility work is how you get more of them.
Common mistakes to avoid
Three errors are common as teams react to this shift.
Trusting last-click reporting as if nothing changed. If your channel decisions still assume every conversion's true source is the final click, you are underfunding AI-influenced demand by design.
Trying to "fix" it with one tool and declaring victory. No single analytics product sees the whole picture today. The realistic goal is a combination: a custom AI channel, self-reported attribution, server-side capture where possible, and AI-visibility tracking. Anyone promising a complete one-click solution is overselling.
Ignoring the gap because it is hard to measure. Difficulty is not a reason to keep optimizing for a number you know is wrong. Even a rough, partial view of AI influence leads to better budget decisions than a precise view of the wrong thing.
How we help
We help businesses get their measurement honest before they make budget calls on it. That means building proper AI channel classification, tightening GA4 and GTM tracking and UTM discipline, adding self-reported attribution into the lead and sales process, and connecting it to a clear view of how each channel actually influences buyers. It connects directly to our work on AEO and GEO services and on search engine marketing for the Japanese market, because seeing AI traffic and earning more of it are two halves of the same problem.
If your Direct traffic has been climbing and you are not sure how much of it is really AI, that is a quick and worthwhile thing to check before your next budget cycle.
FAQ
How much of my "Direct" traffic is actually from AI?
There is no exact figure, because the core of the problem is that this traffic arrives without identification. But analysis reported by Clickport found that around 70 percent of confirmed AI visits land in Direct. A practical signal: if your Direct traffic has grown without a matching rise in brand awareness, email, or app usage, some of that growth is misfiled AI traffic.
Why is AI traffic so hard to track?
Many AI tools, especially their mobile apps, send visitors to your site without a referrer, so analytics cannot tell where they came from. Clicks inside Google's AI Overviews carry a normal google.com referrer, making them indistinguishable from regular organic clicks. And Google removed the signal that previously helped identify AI Overview clicks in early May 2026, with no replacement confirmed yet.
Does this mean GA4 is useless now?
No. GA4 still tracks clicks and conversions reliably for traffic that arrives with proper referrer and UTM data. The issue is that it has no native AI Search channel and cannot see the large share of AI traffic that arrives without identifiers. The fix is to add a custom AI channel, tighten UTMs, and supplement with self-reported attribution rather than abandoning GA4.
What is the single most useful thing I can do quickly?
Add a "How did you hear about us?" question to your lead form or checkout, with an explicit option for AI assistants. It is low-tech, it catches influence no referrer can, and it gives you a real signal for the buyers who were persuaded inside an AI conversation but converted later.
Is the AI attribution gap relevant outside e-commerce?
Yes. Much of the published conversion data comes from e-commerce and SaaS because revenue is easy to measure there, but the underlying behavior applies to any business with a considered purchase, including B2B, professional services, and high-value categories like real estate. Anywhere buyers research before they commit, AI is increasingly part of that research, and the same tracking gap applies.
Will Google fix this in Search Console?
Partly, and slowly. Google began testing dedicated AI Search performance reports this week, showing impressions inside AI Overviews and AI Mode. But those reports show impressions rather than clicks, exclude Gemini, and are rolling out to a limited set of sites first. They improve visibility into where you appear, but they do not close the attribution gap on their own.
Ready to see how much of your traffic is really AI?
We help businesses build honest measurement across GA4, GTM, and first-party tracking, classify AI traffic properly, and turn AI visibility into a channel you can actually report on. Contact us if you want a quick read on how big your AI attribution gap is before your next budget decision.
