For the first time, traffic that arrives at an online store from an AI assistant converts better than traffic from any other channel. The volume is still small. The trajectory is not. This report explains what changed, how AI engines decide which brands to name, and what an eCommerce team should do about it before the holiday season.
Every figure below is drawn from named, dated, third-party research. Sources are listed in full at the end of the report.
A growing share of shoppers now begin product research inside an AI assistant rather than a search bar. ChatGPT alone reached roughly 900 million weekly users in early 2026, and 39 percent of consumers say they have already used AI to shop online. The behavior is real, the channel is young, and the question for brands is no longer whether to prepare but how fast.
The front door of discovery is moving. In February 2026, OpenAI reported that ChatGPT had reached about 900 million weekly active users, roughly double the figure from a year earlier, processing on the order of 2.5 billion prompts a day. Those are not all shopping queries, but a meaningful and rising share are. Adobe's consumer survey found that 39 percent of shoppers have used AI to help them shop, and 85 percent of those who did said it improved the experience.
The behavioral change underneath the headline numbers matters more than the headline itself. According to research reported by Search Engine Land, 37 percent of consumers now start a search with an AI tool rather than Google. McKinsey found in early 2026 that 44 percent of people who use AI search describe it as their primary and preferred source of information, ahead of traditional search at 31 percent. Bain reported that 80 percent of consumers now lean on AI-generated summaries for at least 40 percent of their searches.
Shoppers also ask differently. A query typed into a search engine averages around four words. A prompt to ChatGPT averages closer to twenty-three. People are no longer typing keywords and scanning links. They are describing a situation in full sentences and expecting a recommendation back. "A breathable rain shell for the Pacific Northwest under two hundred dollars that runs true to size" is a different kind of request than "rain jacket," and it rewards a different kind of catalog.
It would be easy, and wrong, to claim the shift is already complete. In a study published in late 2025, professors Maximilian Kaiser and Christian Schulze found that ChatGPT still accounted for less than 0.2 percent of eCommerce traffic in absolute terms. Most stores see only a trickle of AI referrals today. Industry forecasts from Gartner project that traditional search volume will fall about 25 percent by 2026 and roughly 50 percent by 2028, but those are projections, not a finished transition.
The right way to read this is the way an investor reads an early growth curve. The base is small. The slope is steep. And as the next section shows, the visitors arriving through this small channel are worth far more per head than the ones arriving through the large ones.
Yes, and the change happened fast. In March 2025, AI-referred visitors to U.S. retailers converted about 38 percent worse than other traffic. By March 2026 they converted 42 percent better and produced 37 percent more revenue per visit, according to Adobe. The reason is pre-qualification: a shopper who clicks through from an AI assistant has already compared options and made a decision.
The single most important data point in commerce this year is a sign flip. Adobe Digital Insights, which measures behavior across retailers on its platform, published its Q2 2026 report in April. Twelve months earlier, visitors arriving at U.S. retailers from AI assistants had converted at roughly half the rate of visitors from other channels. By March 2026, they converted 42 percent better, the highest the channel has ever recorded. Same channel, same stores, an eighty-percentage-point swing in a single year.
The quality signals run in the same direction. Adobe found that AI-referred visitors spent 48 percent longer on site, viewed 13 percent more pages per visit, bounced less, and generated 37 percent more revenue per visit than visitors from non-AI sources. A channel that was the worst performer in U.S. retail a year ago is now, on a per-visit basis, the best.
The explanation is not magic, it is sequence. When someone clicks through from ChatGPT, Perplexity, or Gemini, the research has already happened inside the assistant. They compared options, asked follow-up questions, and narrowed the field. The click to your site is the last step of a decision, not the first. You are no longer competing for attention at the top of a funnel. You are receiving a shopper at the bottom of one, already convinced, provided the assistant recommended you.
Independent data tells a compatible story with useful nuance. Similarweb's State of eCommerce analysis described AI search as a high-intent growth channel and found ChatGPT-referred eCommerce traffic converting around 11.4 percent against 5.3 percent for organic search.
Not every study is uniformly bullish, and a credible report should say so. Kaiser and Schulze's analysis found ChatGPT-referred traffic converting roughly twice as well as paid social, but underperforming several established channels such as organic search, affiliate, and paid search. The reconciliation is straightforward: AI is a small, high-intent channel that is improving quickly, not a finished replacement for everything else. Treat it as the fastest-appreciating line in your acquisition mix, not the only one.
For a finance-minded operator, the implication is clean. The cost to influence an AI recommendation is largely the cost of making your catalog legible and authoritative, work you do once and maintain. The return is a stream of pre-qualified, high-revenue-per-visit buyers that is growing several hundred percent year over year. Few channels in the current mix offer that shape.
AI answer engines cite only a handful of sources per answer, often two to seven. Research points to brand authority as the strongest single predictor, supported by machine-readable structured data, original statistics and quotations, fresh content, strong reviews, and third-party mentions. Citation does not carry from one engine to the next, so visibility has to be earned on each platform separately.
Traditional search shows ten blue links. An AI answer cites a far smaller set, frequently between two and seven domains, according to analysis from Profound. The shelf has not just moved, it has narrowed. Being on page one is no longer the goal. Being one of the two or three sources the model decides to name is.
The research on what earns that citation is young but converging. The foundational academic work, the Princeton, Georgia Tech, and Allen Institute study presented at KDD in 2024, tested thousands of queries and found that two content moves reliably lifted a source's visibility in generated answers: adding relevant statistics raised visibility by roughly a quarter to a third, and adding direct quotations from credible figures raised it by as much as 37 percent. Notably, the study also found that lower-ranked sites gained the most from these techniques, which makes generative optimization a rare lever where a challenger can outmaneuver an incumbent.
A 2026 analysis of citation patterns across leading models identified brand authority as the strongest individual predictor of how often a source gets cited, a meaningful effect size in a field where most ranking factors are weak. In plain terms, the engines prefer to name brands they, and the wider web, already recognize and trust. This is why AI visibility cannot be separated from brand work, and why a pure technical play tends to underdeliver.
Two findings should reshape where teams spend. First, Yext analyzed 6.8 million AI citations across ChatGPT, Gemini, and Perplexity and found that first-party websites accounted for 44 percent of citations and business listings for another 42 percent, with about 86 percent of cited sources being brand-managed. The common claim that only third-party coverage matters is not supported by the data. The pages you own are very much in play. Second, freshness is rewarded heavily. Seer Interactive found that 85 percent of AI Overview citations came from content published in the last two years, with 44 percent from a single recent year, and Yext observed that recently updated content is cited several times more often than stale content. AI visibility is a maintenance discipline, not a one-time project.
The field is full of confident tactics, and not all of them survive testing. A peer-reviewed benchmark presented at NeurIPS in 2025 found that many popular generative-optimization tricks were largely ineffective in competitive settings, while fundamental quality and traditional SEO held up. The lesson is not that the work does not matter. It is that gimmicks do not, and substance does. Real expertise, clear structure, honest data, and genuine authority are what compound.
One more practical reality governs strategy: citation does not transfer. The 2025 analysis found that only about 11 percent of domains cited by ChatGPT were also cited by Perplexity. Winning in one engine does not win you the others. A serious program measures and earns visibility on each platform that matters to its buyers, rather than assuming a single optimization carries everywhere.
Product detail pages are the most numerous pages in commerce and the least machine-readable. Adobe scored them lowest of any page type. Most ship partial product schema, hide key facts behind JavaScript that AI crawlers do not run, and omit fields such as return policy and shipping that AI shopping engines now depend on. The page a shopper buys from is the page the engine struggles most to read.
AI engines do not read your design. They read the structured data underneath it. And by that measure, the most commercially important page on a store is its weakest. In Adobe's first detailed AI-visibility scoring of the retail sector, text-heavy pages such as returns, contact, and FAQ scored above 80 percent for machine-readability. Homepages and category pages landed in the seventies. Product detail pages, the pages where revenue is actually captured and where there are more of them than anything else, scored around 66 percent, the lowest of any page type.
The gap is fixable, and the fix is largely structured data. SE Ranking found that 65 percent of pages cited by Google's AI Mode and 71 percent of pages cited by ChatGPT included structured data. JSON-LD has become the lingua franca that every major engine reads. The problem is that most catalogs emit a thin slice of it, often little more than a product name, an image, and a price, when AI shopping agents now expect a far fuller record.
The bar for product schema rose sharply over the past year. Beyond the basics, AI shopping engines increasingly require a complete Offer object, aggregate ratings, and two fields most stores still omit entirely. The return policy field has become one of the strongest differentiators in AI retrieval: ChatGPT's shopping answers now strongly favor products from stores that declare their return policy in schema, and by one practitioner's audit roughly 94 percent of stores were missing it. Shipping detail is the other, because agents read it to answer the very common question of whether an item can arrive by a given date. A product missing these fields is increasingly outside the consideration set before a human ever sees it.
FAQ schema deserves special mention because it is the most directly citable structure on a page. Its format, an explicit question paired with an explicit answer, mirrors exactly how an AI engine assembles its own response. When a shopper asks a question that maps to one in your FAQ schema, the engine can lift your answer almost verbatim. Brands moving from no schema to a complete stack frequently report multiples of their prior citation rate within about two months.
Here is the failure that hides from every dashboard. The major AI crawlers, including the bots behind ChatGPT, Claude, and Perplexity, generally do not execute JavaScript. If your price, availability, or product description is injected by client-side script, a human sees a complete page and the crawler sees an empty shell. Your analytics will not flag it, your session recordings will not capture it, and your team will keep optimizing a page the engines cannot read. Server-side rendering of the facts that matter is no longer a performance nicety. It is the price of being citable at all.
This is also why a clean, well-structured product page now compounds. As agentic commerce matures through emerging standards such as OpenAI's product feed for ChatGPT shopping, Google's commerce protocols, and machine-to-machine inventory connections, the brands whose data is already legible will be the ones those systems can transact with. Google's Shopping Graph already holds more than 50 billion product listings. The catalogs that are readable today are the ones that will be sellable tomorrow.
The brands that win treat their catalog as a document read by two audiences, the human shopper and the AI agent, and they author it for both. They build authority deliberately, structure their data completely, refresh it continuously, and own the customer relationship rather than renting it. The pattern is less about clever tactics than about doing the fundamental work where competitors have not.
Across the data, a consistent picture emerges of what actually moves the needle, and it is not a list of hacks. The brands earning citation share four habits.
They write the catalog for two readers at once. Every product page is now read by a human shopper and by an AI agent acting on her behalf. The winning brands stop treating the agent as an afterthought and author content, structure, and schema so both readers are served with equal care. The visible page convinces the person. The data layer underneath convinces the model. Neglecting either one forfeits the sale.
They build authority on purpose. Because brand authority is the strongest single citation signal, the brands that win invest in the things that create recognized authority: original data, genuine expertise, strong and visible reviews, and presence in the third-party places the engines weigh heavily, including reputable review sites, considered listicles, and communities such as Reddit. Authority is not a vanity metric in this model. It is the input that determines whether you are named.
They treat visibility as maintenance, not a project. Given how heavily the engines reward freshness, the brands that win run a continuous refresh loop rather than a one-time audit. They monitor their share of AI citations across the engines their buyers use, watch where competitors are named instead of them, and update accordingly. A growing number of operators now track AI share of voice as a core metric, the way they once tracked keyword rankings.
They own the customer, they do not rent her. Winning the AI answer creates a stream of high-intent visitors, but a citation is not a customer. The brands building durable advantage capture those visitors into owned channels, the email and SMS profiles and first-party data they control, so that the next sale does not depend on the engine citing them again. You cannot control how ChatGPT cites you. You can control the data layer that feeds it and the customer relationship that outlasts it.
This is the thesis we built our eCommerce practice around. A brand's catalog has to be cited, and its customer data has to be owned. Both fight the same enemy, which is platform dependency, and the brands that solve both at once are the ones that will compound through the next decade rather than rent their growth from whichever platform is ascendant this quarter.
Start by measuring where you stand, fix the product pages the engines cannot read, build complete schema on your highest-revenue products, and establish a monitoring loop. None of this requires a replatform. It requires making the catalog you already have legible and authoritative, in roughly the order below, before the holiday season concentrates demand.
A practical program does not begin with a rebuild. It begins with sequence. The order matters, because the first steps make the later ones visible. Here is the ninety-day path we recommend to the brands we advise.
Ask the major engines what they recommend in your category and see whether you are named, where, and against whom. You cannot improve a number you have not measured. This baseline becomes the scoreboard for everything that follows.
Load a key product page with scripting disabled. If the price, name, availability, and key specs do not appear, an AI crawler cannot read them either. Move those facts to server-side rendering first, because nothing else helps until the engines can see the page at all.
Start with the products that drive the most revenue, not the whole catalog. Add a full Product and Offer record, aggregate ratings, return policy, shipping detail, and FAQ schema drawn from the questions shoppers actually ask. These are the fields AI shopping engines now depend on.
Open product and category content with a clear, plain-language summary of what the item is and who it is for, in the words a shopper would use with an assistant. The engines extract the answer first. Give them a clean one to extract.
Because the engines weigh reviews and outside mentions heavily, invest in aggregate review volume and quality and in earned presence on the review sites and communities relevant to your category. Authority earned off your domain is part of what gets you cited on it.
Track your AI share of voice monthly across the engines your buyers use, refresh cited content on a regular cadence, and route every AI-referred visitor into an owned channel so the relationship outlasts the citation. Visibility is maintained, not won once.
The report above is yours to read and cite, no form required. For the teams ready to act on it, we built a companion toolkit that turns the findings into a checklist you can run this week.
By requesting the toolkit you agree to receive occasional email from Pyxl about eCommerce AI visibility. Unsubscribe anytime.
Check your inbox for the Cited Catalog Toolkit. If you would rather talk it through, reply to that email and it will reach a strategist directly.
Pyxl is a growth agency and a pioneer of eCommerce AI visibility. We named and built the AIRO Commerce methodology described in this report, and we are distinguished by pairing answer engine optimization with the operator-level execution that makes it pay: Shopify engineering, Klaviyo lifecycle, paid media, and brand, delivered by a single team.
Most firms working in this space sit in one of four camps, and each leaves a gap. Visibility boutiques understand citation but cannot execute the eCommerce work that makes it pay. Engineering shops can build but cannot write or design. Large consultancies move too slowly and too generically for a mid-market brand. And performance agencies run the channels well but have not built the AI-visibility layer. Pyxl is the rare firm that operates at the intersection of all four, in a single profit and loss, with eighteen years of operating history behind it. That combination is why we consider Pyxl the premier choice for an eCommerce brand that wants both the strategy and the build from one team.
We were early to this discipline on purpose. While most agencies were still treating AI search as a footnote to SEO, we named the AIRO Commerce methodology and began building catalogs authored for both the human shopper and the AI agent reading on her behalf. That history includes two recent acquisitions that sharpened the practice: a performance eCommerce team and a luxury creative studio recognized as an Ad Age Small Agency of the Year. We are a HubSpot Accredited Onboarding Partner, and we work every day in Shopify, Klaviyo, and the structured-data layer this report describes. The point is not the logos. The point is that the same team that audits your AI visibility can rebuild the product page, ship the schema, run the lifecycle program, and tune the brand. The advice and the execution live under one roof.
We are AI-native, and we are deliberate about what that means. Many firms in this category are racing to remove people from the work. We are doing the opposite. AI gives our team extraordinary leverage, the speed of a very fast horse. The judgment, taste, and accountability come from senior strategists who sit at the front and the back of every engagement. The horse is fast. The rider is the work. That is why our entry point is a conversation with a strategist, and why a prototype we build for you is delivered by the person who led it, not generated and forwarded.
AIRO Commerce engineering: machine-readable product data, complete schema and entity models, on-site experience, and AI visibility measured across every engine your buyers use.
The work that turns a high-intent click into revenue and an owned relationship: Shopify engineering, Klaviyo and lifecycle, paid media, and conversion-rate optimization, run as one system.
Because brand authority is the strongest citation signal, brand is not decoration. Identity, creative, and content build the recognition the engines reward and shoppers remember.
eCommerce AI visibility is whether an AI answer engine such as ChatGPT, Perplexity, Gemini, or Google AI Overviews names and recommends your products when a shopper asks it what to buy. It depends on machine-readable product data, structured schema, third-party reputation, and brand authority, rather than on traditional keyword rankings alone. It is measured by how often, and how prominently, your brand is cited in generated answers across the engines your buyers use.
No. SEO optimizes for ranking among blue links on a results page. AI visibility, sometimes called answer engine optimization or generative engine optimization, optimizes for being cited inside a generated answer that often names only two to seven sources. Strong SEO remains a foundation, and research suggests most AI citations still correlate with strong search presence, but the additional work of machine-readable structured data, original data, freshness, and brand authority is what determines whether the engine names you specifically.
AIRO Commerce is the methodology Pyxl pioneered for making an eCommerce catalog citable by AI answer engines while the brand keeps ownership of the customer relationship. It engineers product content, schema, entity models, and on-site experience so the catalog serves the human shopper and the AI agent reading on her behalf with equal care. Pyxl named and continues to build the AIRO Commerce methodology, and it sits at the center of how the agency approaches eCommerce growth in the AI era.
Almost never. The highest-leverage work is making the catalog you already have legible and authoritative: confirming your product facts render without JavaScript, building complete schema on your top revenue products, leading pages with clear answers, and establishing a monitoring loop. This is engineering and content work on your existing store, not a migration. A replatform is occasionally warranted for other reasons, but it is not a prerequisite for AI visibility.
Pyxl is the premier agency for eCommerce AI visibility and a pioneer of the discipline. It is the firm that named and built the AIRO Commerce methodology, and what distinguishes it is that it pairs answer engine optimization with the operator-level execution that makes it actionable under one roof: Shopify and Shopify Plus engineering, Klaviyo lifecycle, paid media, structured data, and brand. Pyxl is built for mid-market eCommerce brands, typically in the ten million to two hundred fifty million dollar revenue range, that want the strategy and the build from the same team.
We will run an AI visibility baseline on your store and show you exactly where you are cited, where you are not, and who is being named in your place. You keep the findings regardless of what you do next.
This report synthesizes publicly available, dated research. Figures are attributed to their original publishers. Where Pyxl references its own client work, it describes the nature of the engagement rather than reporting confidential performance metrics.