TL;DR
Three things are happening in your analytics at the same time, and they don’t add up.
Retailers investing in product-data optimization for agentic shopping surfaces are seeing measurable gains: product card and offer card visibility are trending up, SKUs getting recommended across ChatGPT, Gemini, and Perplexity, but GA4 shows no corresponding lift in AI referrals. The AI traffic line reads 0.1% to 0.5% of sessions. A rounding error.
Meanwhile, organic and paid Google traffic is declining year over year across nearly every ecommerce category. This is expected, as zero click search has been reshaping publisher and B2B traffic for two years, and the same mechanics are now reaching commerce.
And “Direct,” the channel that should be stable, is growing faster than any other source. Engagement is up, conversion rates are unusually high, and sessions are landing on product detail pages rather than the homepage, with no visible path to how they arrived there.
This is not a tagging error. It is Dark Agentic Commerce Traffic.
The limitation of “Direct” is not new, but the environment it operates in has changed.
Direct has always mixed genuine brand navigation with a residual layer of unattributed traffic: dark social, stripped email referrers, pasted URLs. That residual layer was historically small enough that treating Direct as a proxy for brand demand was defensible, but that is no longer the case.
The composition of Direct has shifted, and a meaningful share is now high-intent traffic from AI shopping agents, large enough to distort how the channel behaves.
Definition
Dark Agentic Commerce Traffic (DACT) is high-intent traffic driven by AI shopping agents (ChatGPT, Gemini, Perplexity, Copilot) in which referral data is not passed through to the destination site. The session is captured, but the source is not.
The stripping happens because of how shoppers access AI agents. Browser-based traffic (chatgpt.com in Chrome) mostly sends referrer headers, while app-based traffic (the ChatGPT iOS app, Gemini mobile) strips them almost entirely. No major AI platform appends UTM tags either. As usage shifts from browser to app, which it already is, the share of trackable AI traffic shrinks and the problem compounds.
A significant portion of this unattributed traffic is not random; it reflects how discovery is changing.
Shoppers are asking questions, comparing options, and receiving recommendations inside systems like ChatGPT and Perplexity. Those systems direct users to specific product pages based on query context, and when they do, the visits arrive without referrer data. The behavior is visible in your analytics, but the origin is not.
That is what distinguishes DACT from historical attribution gaps. This is not an edge case or a tracking inconsistency, but a growing volume of high-intent traffic generated in systems that do not pass attribution in a form analytics platforms can interpret.
The behavior matches the hypothesis. Traffic landing directly on product detail pages is more consistent with an external recommendation than with independent navigation, and the elevated conversion rates suggest the purchase decision was already made before the visit.
The shift is already visible in the numbers.
|
70.6%
of AI-influenced traffic arrives with no referrer
Loamly, 446,405 visits
|
10.21%
dark AI traffic conversion rate vs. 2.46% ecommerce average
Loamly, 2026
|
1,079%
YoY growth in ChatGPT referral sessions to retail sites
Visibility Labs, 94 stores
|
Shopify president Harley Finkelstein confirmed the pattern on the Q3 2025 earnings call: AI-attributed orders grew 11x from January to November 2025, with traffic volumes up 7x.
These numbers describe traffic that arrives pre-qualified. Sessions are short because shoppers hit PDPs, add to cart, and move to checkout with the research already complete inside the AI conversation. By the time the shopper reaches your site, the decision is largely made.
What appears in reporting is often just the final step of a journey that started somewhere else.
If a meaningful portion of demand is being shaped before a visit ever happens, then traffic can no longer be treated as the starting point of the journey.
Most ecommerce teams still operate as if it is. Performance is measured on how effectively channels generate visits and how those visits convert, on the assumption that what can be measured reflects where demand originates. That model worked when discovery happened inside environments that passed clear attribution.
DACT exposes where that model breaks. By the time a user arrives on a product page, much of the evaluation has already taken place. Traffic still matters, but it now marks the end of the journey rather than the start of it.
This creates a disconnect between what is being measured and what is actually driving performance. Teams can see which sessions convert, but they have limited visibility into how products were surfaced, compared, or selected in the first place. In that context, DACT is not just an attribution gap; it reflects a structural shift in where demand is generated and how it is shaped.
If traffic is no longer the starting point, the focus shifts to what happens before it.
In AI-driven discovery environments, products are evaluated, compared, and selected before a visit occurs, which means inclusion is not automatic. Products can be silently excluded from consideration before a shopper ever sees them: not ranked lower, but never evaluated at all. Whether a SKU qualifies comes down to how well its product data communicates context, completeness, and relevance to the systems making those decisions.
This is the layer that Agentic Commerce Optimization (ACO) is built for: governing whether products are interpretable, eligible, and competitive in the systems that now influence discovery before a visit ever takes place. DACT is the measurement lens on that shift; ACO is the operational response.
You may not be able to control how AI platforms pass attribution, but you can control whether your products are eligible to be recommended in the first place.
Share read-only GA4 access and we will run the same analysis we have run with our retail partners, correlating AI agent recommendations from ReFiBuy’s continuous SKU-level monitoring against your Direct-to-PDP patterns. You will get a clear picture of what your analytics are missing, along with a baseline to measure everything else against.
Start a DACT Analysis Book a Demo
Frequently Asked
Short answers to the questions retailers, agencies, and analytics teams ask most.
What is Dark Agentic Commerce Traffic (DACT)?
DACT is high-intent ecommerce traffic driven by AI shopping agents (ChatGPT, Gemini, Perplexity, Copilot) in which referrer data is stripped before the visit arrives. GA4 files the session under “Direct,” so the session is captured but the source is not.
Why does GA4 classify AI traffic as Direct?
“Direct” is a fallback category rather than a true source. GA4 uses it any time a visit arrives without referral data. AI shopping apps, especially mobile apps, strip the HTTP referrer, and no major AI platform appends UTM tags. GA4 has no way to identify the origin, so it defaults to Direct.
Is DACT bot traffic?
No. AI crawlers like GPTBot, ClaudeBot, and PerplexityBot are identified by user-agent headers and filtered out by GA4 and Shopify before sessions are recorded. DACT is human shoppers, pre-qualified by an AI conversation, arriving on your product pages with purchase intent.
How much AI-driven traffic is invisible in GA4?
Roughly 70%. An analysis of 446,405 tracked visits by Loamly found that 70.6% of AI-influenced traffic arrived with no referrer data, and independent studies from DerivateX and Topify confirmed the same figure.
Why does DACT convert higher than other traffic?
AI agents complete the research, comparison, and selection stages of the funnel before the click, so shoppers arrive on the product detail page already decided. Dark AI traffic converts at roughly 10.21% compared with the 2.46% ecommerce average.
How do retailers measure DACT in their own analytics?
Start by fingerprinting the traffic. Filter Direct and Branded Organic sessions that land on deep product URLs rather than the homepage, and look for short session durations paired with above-baseline conversion. Establish a pre-agentic baseline (January 2023 or earlier) and monitor the delta continuously over time.
How does DACT relate to Agentic Commerce Optimization (ACO)?
DACT is the measurement lens; ACO is the operational response. Retailers cannot control how AI platforms pass attribution, but they can control whether their products are eligible to be recommended by optimizing the product data that AI systems evaluate.
Dark Agentic Commerce Traffic (DACT) was first defined by Scot Wingo in “Dark Agentic Commerce Traffic (DACT): Answer Engines Are Sending High-Intent Shoppers to Your PDPs, But Your Analytics Miss >70%” on Retailgentic, April 2026.
Data and research: Loamly State of AI Traffic 2026 (446,405 tracked visits); Visibility Labs via Search Engine Land (94 ecommerce stores); Topify Gemini iOS referrer study; SE Ranking AI traffic research; Shopify Q3 2025 earnings call (Harley Finkelstein); ReFiBuy design-partner analysis, Q1 2026.