Aera doesn’t browse. It decides.
When a shopper asks an AI assistant to find the best noise-canceling headphones under $300 with two-day shipping, something unremarkable appears to happen: the AI returns a short list, the shopper picks one, and the purchase completes inside the chat window. The retailer ships the order.
What actually happened is more consequential. The shopper never visited a product page. They never saw your hero image, your lifestyle photography, or the brand story you spent three months writing. They never encountered your trust badges, your size guide, or your exit-intent offer. The AI agent evaluated your product against a dozen competitors in milliseconds, using criteria you didn’t design for, and either included you on the shortlist or didn’t.
This is not a future scenario. According to Adobe’s April 2026 data, traffic from AI sources to U.S. retail sites grew 393% year over year in Q1 2026. AI-referred shoppers now convert 42% better than non-AI traffic — a complete reversal from March 2025, when AI traffic converted 38% worse. The channel is maturing fast. The window for first-mover advantage is narrowing.
This post examines what AI-mediated browsing means for every team that touches the shopping funnel: UX, SEO, and conversion engineering. It is not a prediction. It is a diagnosis.
The Funnel Is Not Broken — It Has Been Replaced
What the AIDA model assumed about human behavior
The AIDA model — Awareness, Interest, Desire, Action — has organized marketing thinking since Elias St. Elmo Lewis codified it in 1898. For more than a century, it described something real: human beings need to discover a product, learn about it, want it, and then decide to buy it. Each stage takes time. Each stage is a surface for persuasion.
The entire apparatus of digital commerce was built on this assumption. SEO captures attention in the awareness stage. Content marketing builds interest. Product pages engineer desire through photography, copy, and social proof. Checkout flows minimize friction at the action stage. A/B testing optimizes every pixel along the path.
The model held because humans browsed. Browsing is cognitively expensive, emotionally variable, and commercially exploitable. A shopper who arrives at your product page has already invested attention. They can be influenced.
How AI agents compress Awareness → Decision into a single inference step
AI agents do not browse. They query, retrieve, rank, and recommend — in a single pass. The FTI Consulting analysis of agentic commerce describes it precisely: “As AI agents prioritize consumer intent and embed shopping into the moment of discovery, the gap between product exploration and purchase is disappearing.” The consumer journey, they write, is beginning to resemble “a loop: high velocity, compressed, personalized, non-linear and continuous.”
In practice, this means the four stages of AIDA collapse into one. Awareness, Interest, and Desire are handled by the agent’s reasoning process — invisible to the retailer, inaccessible to the marketer. What reaches the human is a curated shortlist, not a browsing environment. The decision has largely been made before the shopper is consulted.
This is not hyperbole. McKinsey’s agentic commerce research documents that AI agents are already comparing options, assembling baskets, and completing checkout via emerging payment protocols. ChatGPT now has over 800 million weekly active users. Google’s AI Overviews reach more than 1.5 billion users monthly. Commerce is flowing through AI channels at a scale that makes this a systems-level problem, not an edge case.
The “surgical shopper” and the collapse of the browsing middle
The behavioral shift is not limited to AI-native interactions. Even when shoppers arrive at retail sites through traditional channels, their behavior has changed. Quantum Metric’s 2025 Peak Benchmark found that weekly retail site traffic was down 21% year over year, while conversion rates fell 27%. Shoppers are not browsing less because they are buying less. They are arriving having already decided.
Salesforce’s Cyber Week 2025 data named this the “surgical shopper” — a consumer who is “deal-hungry, inflation-trained,” arriving at checkout with specific goals rather than open-ended intent. Foot traffic to physical stores was down across every U.S. region. Digital visits were fewer but more purposeful. The browsing middle — the consideration phase that traditional CRO was built to optimize — is compressing on both sides.
AI is the primary driver. The shopper who used to spend 40 minutes comparing vacuum cleaners across six tabs is now asking an agent to do it in 40 seconds. That 40-minute window was the opportunity for your content, your UX, and your persuasion architecture. It is shrinking by design.
The Numbers That Should Alarm Every E-commerce Team
The data from the past 12 months paints an unusually coherent picture for a market in transition. Several independent data sources converge on the same conclusion: AI-mediated commerce is growing at a rate that makes incrementalism a strategic error.
Traffic growth is exponential. Traffic from generative AI browsers and chat services to U.S. retail sites grew 4,700% year over year in July 2025, according to industry tracking data cited by SmarterArticles. HUMAN Security’s 2026 State of AI Traffic Report found that agentic AI traffic — systems that navigate and transact, not just crawl — grew 7,851% year over year. Monthly AI-driven traffic volumes nearly tripled over the course of 2025.
AI-referred visitors are higher quality. Adobe’s Q1 2026 data shows that when consumers arrive at a retail site from an AI source, their engagement rate is 12% higher than non-AI traffic. They spend 48% longer on the site and browse 13% more pages per visit. AI-driven revenue per visit was 37% higher than non-AI traffic as of March 2026. The conversion quality of this channel is now demonstrably superior to paid search and email.
But most retailers are invisible to it. The same Adobe report includes a diagnostic finding that should alarm product and marketing leadership: a substantial portion of retail product pages are not machine-readable by AI systems. The structured data required for AI agents to understand, compare, and recommend a product is absent from the majority of e-commerce product pages. Brands optimized for human browsers are invisible to the agents now driving the highest-converting traffic.
The zero-click economy is accelerating. Seer Interactive’s September 2025 analysis found that organic click-through rates dropped 61% for queries with AI Overviews, from 1.76% to 0.61%. Google AI Overviews now appear for over 13% of all queries, a figure rising steadily as the feature expands internationally. For product discovery queries, this means the majority of intent-matching impressions now resolve inside the AI interface — without a click to your site.
The strategic implication is uncomfortable: your best-performing channel may soon be one you have not yet optimized for, and your worst-performing visibility metric may be your most important one.
Who Is Deciding for Your Customer?
Understanding who the agents are, how they work, and what they favor is not optional knowledge for commerce teams. The protocol landscape is consolidating quickly.
ChatGPT’s Instant Checkout (ACP), Google’s UCP, and Perplexity’s Comet
In September 2025, OpenAI and Stripe released the Agentic Commerce Protocol (ACP), enabling ChatGPT to complete purchases from over one million Shopify merchants via Instant Checkout. OpenAI charges merchants a 4% transaction fee per completed purchase, in addition to standard Stripe processing fees. For a $100 order, total platform and processing fees run approximately $7.20.
In January 2026, Google launched the Universal Commerce Protocol (UCP) at NRF, backed by Walmart, Target, Etsy, Wayfair, and over 60 payment networks. Unlike ACP, which is optimized for a single-agent experience within ChatGPT, UCP is designed as an open, intent-based protocol that can operate across multiple AI surfaces simultaneously. As Opascope’s agentic commerce guide describes it: “UCP is an open highway system. ACP is a walled garden with an open door.”
Perplexity has taken a web-search-native approach with its Comet browser, synthesizing product information from across the internet rather than relying on direct merchant feed integrations. This makes earned media and citation-building more valuable for Perplexity visibility than feed optimization alone.
For most brands, the operative guidance is: you will need to support both ACP and UCP. They solve different problems and serve different discovery contexts. Treating them as redundant is a category error.
Amazon’s walled garden vs. the open protocol war
Amazon is conspicuously absent from both ACP and UCP. Instead, it has pursued a proprietary strategy: Rufus AI (its in-app shopping assistant with 300 million users), Alexa+ with agentic purchasing capabilities, and an aggressive legal posture against external agents. Amazon sued Perplexity in November 2025 for unauthorized purchases via the Comet browser, updated its robots.txt to block OpenAI’s crawlers — removing 600 million products from ChatGPT’s shopping results — and revised its legal terms to prohibit AI agent behavior.
TechCrunch’s reporting frames this as a structural division in the market: Amazon-first brands should optimize for Rufus, not external agents; multi-channel brands need to prepare for three distinct agent ecosystems simultaneously.
What agents actually look for — and what they ignore
Understanding agent decision-making is the most practically important thing a commerce team can know right now. According to Goodie’s analysis of ChatGPT Shopping, Google AI Mode, Amazon Rufus, and Perplexity Shopping, the factors driving product recommendation are consistent across platforms:
- Structured product data quality — completeness and accuracy of schema markup, GTINs, attributes
- Real-time inventory and pricing accuracy — stale data is a disqualifier
- Review sentiment and volume — opinion mining is a core part of agent recommendation logic
- Fulfillment metadata — shipping speed, return policy, availability by region
- Third-party authority signals — citations from trusted directories, review platforms, and editorial sources
- Feed update frequency — ChatGPT accepts updates every 15 minutes; stale feeds lose retrieval
What agents largely ignore: hero images, lifestyle photography, brand storytelling, emotional copywriting, trust badges, and UX design. These are human persuasion tools, and agents are not humans.
The Death of the Product Page (As You Know It)
This is the section that will be most uncomfortable for UX and brand teams to read, so it is worth being precise about what is and is not changing.
Agents don’t see your hero image, your lifestyle photography, or your brand story
HUMAN Security’s traffic analysis found that 77% of agentic AI activity in 2025 occurred on product and search pages — but the nature of that activity is qualitatively different from human browsing. Agents do not render JavaScript, load images, or engage with visual UX elements. Search Engine Land’s analysis of ChatGPT agent behavior found that 46% of ChatGPT bot visits begin in reading mode — a plain HTML version of a webpage with no images, CSS, JavaScript, or schema markup rendering. After landing, 63% of ChatGPT agents leave immediately. Common reasons include HTTP errors, slow load times, CAPTCHAs, and bot-blocking scripts.
The product page you have spent years optimizing is, from an agent’s perspective, a document to be parsed, not an experience to be had. What the agent extracts is structured data, semantic content, and factual attributes. What it cannot extract — and therefore cannot use — is everything that currently differentiates premium product pages from commodity listings.
The six signals agents use to evaluate and rank products
Mirakl’s analysis of how AI agents assess product pages identifies six weighted evaluation categories. In order of impact:
- Structured data — schema markup and standardized product attributes (highest weight)
- Content completeness — detailed specifications, dimensions, materials, compatibility data
- Semantic richness — use-case descriptions that match conversational natural-language queries
- Pricing competitiveness and accuracy — real-time pricing data evaluated against comparable SKUs
- Inventory reliability — accurate stock status, particularly for fulfillment timing
- Review and trust signals — third-party ratings, Q&A content, verified purchase data
A concrete example from Mirakl illustrates the shift: a traditional SEO-optimized page for running shoes repeats “comfortable running shoes” to rank for the keyword. An agent requires structured attributes: heel-to-toe drop (8mm), cushioning type (gel-based), arch support (medium), weight (9.2 oz), materials (engineered mesh), intended use (long-distance running). Without these attributes, the product cannot be matched to intent-based queries, and the agent moves on.
What “agent-readable” vs. “human-readable” actually means in practice
Agent-readability is a technical property, not a design property. A page is agent-readable when its core product information is accessible to a parser without JavaScript execution, when its schema markup is complete and valid, and when its semantic content can be chunked and extracted without ambiguity.
Adobe’s AI Content Visibility Checker — released alongside its Q1 2026 data — can diagnose any product page for machine readability. The audit reveals what LLMs can and cannot read from a given URL. For most retail teams, the first pass of this audit will surface significant gaps.
The key structural insight: agent-readability and human-readability are not in conflict. A page with complete structured data, accurate product attributes, and semantic use-case content is also a better page for human shoppers. The investments compound. The urgency is that most e-commerce product pages have not been built to either standard simultaneously.
SEO Is Not Dead — It Just Has a New Name
The emerging discipline of Generative Engine Optimization (GEO) is not a replacement for SEO. It is an evolution layered on top of it — and understanding the distinction matters for how you prioritize your roadmap.
What GEO is and how it differs from traditional SEO
Traditional SEO optimizes for a specific outcome: visibility in ranked search results, measured by position, impressions, and click-through rate. The underlying logic is keyword-to-document matching. Algorithms evaluate signals — backlinks, page authority, technical health, content relevance — and rank documents accordingly.
GEO optimizes for a different outcome: being selected as a trusted source within AI-generated answers and product recommendations. As Wordstream’s 2026 GEO guide describes it, the goal shifts from “getting traffic” to “being cited, summarized, and trusted by AI systems.” This means content must be machine-parseable, semantically unambiguous, factually accurate, and structurally clear enough for a generative model to extract and reuse without hallucination.
The Pimberly GEO vs. SEO comparison captures the operative distinction well: “SEO optimizes for discovery. GEO optimizes for comprehension.” You can rank #1 organically and still be invisible to an AI agent if your content is not structured in a way the agent can parse and trust.
For e-commerce teams, the practical implication is that product content strategy must now serve two masters simultaneously. As Salsify’s GEO guide for ecommerce puts it: “SEO: Use keywords intelligently to put your content on the first page of search results. GEO: Use EEAT content to get your site linked as a reference in AI overviews. AEO: Add contextual information like ‘How does this product work with other products? What is it used for?'”
These are not mutually exclusive. The same structured data that helps AI agents understand your products also powers rich snippets in Google search results.
Structured data as the new storefront
For product pages specifically, schema markup is the foundational investment. Complete schema for an e-commerce product includes, at minimum: product name, brand, SKU, price (with currency code), availability status, images, GTIN/UPC, and technical specifications. Review schema enables ratings data to surface in AI recommendations. Local business schema ensures fulfillment and location data are accurate for regional queries.
A 2025 study cited by structured data researchers found that GPT-4 increases its accuracy from 16% to 54% correct responses when content uses structured data. For product queries specifically, this is the difference between being on the AI’s shortlist and not existing.
Goodie’s analysis of ChatGPT Shopping adds a practical note: “ChatGPT weights structured product data highest (19 out of 100 in our analysis) because it doesn’t crawl sites. What you push is what it knows.” Feed quality is not a technical afterthought — it is the primary marketing asset for AI-mediated channels.
The llms.txt standard and robots.txt for AI crawlers
A new technical standard has emerged alongside GEO: llms.txt, a markdown-based file (analogous to robots.txt) that provides a structured way for AI agents to access and understand your most important content. Search Engine Land’s technical SEO guide for generative search identifies llms.txt as a practical edge — adoption is currently low, meaning early implementers benefit disproportionately.
Your robots.txt also requires review in the context of AI crawlers. The relevant agents — GPTBot, OAI-SearchBot, ClaudeBot, Perplexity-Bot, Google-Extended — each require explicit policy decisions. You may want training crawlers to have different access than real-time search crawlers. You may want to allow search citation while restricting training data. These are now active strategic decisions, not default configurations.
How each platform sources differently
Yext’s analysis of 6.8 million AI citations across Gemini, ChatGPT, and Perplexity reveals distinct sourcing behaviors that demand platform-differentiated strategies:
Google Gemini behaves most like a traditional search engine with elevated standards. 52% of Gemini citations come from brand-owned websites. It favors structured, factual content from a brand’s own domain — especially pages with schema markup and consistent subdomains. Brand-owned pages are your primary lever.
ChatGPT builds authority through comprehensive, well-structured content. Users arrive pre-qualified and further down the funnel. Being cited as a trusted source when ChatGPT synthesizes answers is the primary objective — which means investing in authoritative long-form content and third-party mentions.
Perplexity sources more narrowly from industry-specific directories and leans heavily on third-party reviews and ratings. In verticals like hospitality, healthcare, and consumer goods, niche directory presence and review volume drive citation frequency. Perplexity rewards specialization and third-party corroboration over brand-owned content.
The operational conclusion: AI visibility is not one-size-fits-all. Treating it as a single system risks being invisible in all three.
Conversion Engineering for the Agentic Era
Conversion rate optimization, as currently practiced, optimizes for human decision-making: reducing friction, building urgency, increasing persuasion. In the agentic era, a meaningful portion of the “conversion” decision happens before a human is involved. CRO must evolve accordingly.
Optimizing for agent selection, not human persuasion
Agent selection is a retrieval problem, not a persuasion problem. The agent evaluates candidates against a structured set of criteria and returns a ranked shortlist. Influencing that ranking requires the same tools as influencing any algorithmic ranking: clean data, strong authority signals, and precise attribute matching to likely queries.
One research finding deserves particular attention: a study on LLM agent behavior found that AI agents are “hypersensitive” to nudges in their decision environment — environmental design choices that influence recommendations without restricting options — at a significantly higher rate than human shoppers. This creates a double-edged dynamic. Agents can be influenced by well-structured product data and authority signals in ways that consistently favor well-optimized brands. But it also means that poorly structured data, inconsistent attributes, or negative review patterns can systematically suppress recommendations — with no human judgment to override the signal.
Real-time inventory, pricing accuracy, and fulfillment metadata as conversion levers
Agents evaluate fulfillment reliability as a first-class signal. A product with perfect structured data and strong reviews will be deprioritized if inventory data is stale, shipping estimates are inaccurate, or pricing is inconsistent across feeds. Goodie’s guide makes this explicit: “The factors that matter most are about discoverability and data quality, not checkout protocols. You can have the best Stripe integration in the world and never appear in recommendations because your product titles are inconsistent or your inventory data is stale.”
For operations and merchandising teams, this reframes fulfillment data as a marketing problem. The accuracy of your inventory feed is now directly connected to your share of AI-mediated recommendations.
The new CRO stack: product feeds as first-class marketing assets
The practical restructuring for conversion engineering teams centers on one principle: treat your product feeds the way you treat your advertising creative. Assign ownership, establish quality standards, create update cadences, and measure performance.
This means: complete product titles (not abbreviated SKU codes), rich descriptions (use-case scenarios, compatibility details, real-world applications), accurate GTINs and brand identifiers, and structured attributes at the variant level — not just the product level. It means monitoring feed freshness and updating inventory and pricing in near-real-time.
Retailers that utilized branded AI agents during Cyber Week 2025 grew sales 32% faster than those who did not, according to CMSwire’s Cyber Week analysis. The gap between agent-ready and agent-invisible brands is measurable. It is growing.
What This Means for UX Designers and Product Teams
The natural response to “agents bypass your product page” is to deprioritize UX investment. That response is wrong. The implications are more nuanced, and the UX opportunity is larger than it appears.
Designing for two audiences simultaneously
The agentic era creates a bifurcated audience for every product page. The first audience is the AI agent — which needs machine-readable structured data, semantic attribute content, and fast, clean HTML. The second audience is the human shopper who arrives after the agent has already pre-selected your product — which needs confirmation, trust reinforcement, and a frictionless path to purchase.
These audiences have different needs, but they are not in conflict. A product page optimized for agent readability — with complete structured data, precise attribute descriptions, and fast load times — is also a better page for the high-intent human who arrives pre-qualified. The agent-referred shopper who lands on your page has already been told you are the best option. Your UX job shifts from persuasion to confirmation.
This is a significant reframing. The effort previously spent on awareness-stage UX — attention-capturing hero sections, emotional brand statements, comparison tools — may be better allocated to post-agent arrival: clear specification displays, transparent return policy, immediate checkout pathways, and rapid trust confirmation.
Post-agent UX: what the shopper expects on arrival
Adobe’s data on AI-referred traffic behavior provides a directional signal: these shoppers spend 48% more time on-site and visit 13% more pages. They are not bouncing — they are verifying. They are cross-referencing the agent’s recommendation against the actual product page, checking stock, reading reviews, confirming the specification details the agent surfaced.
The design implication: AI-referred shoppers are confirmation-seeking, not discovery-seeking. Product page design should prioritize information density and verification affordances over narrative and aspiration. Specification tables, real-time inventory status, review summaries, and one-click purchase paths will outperform lifestyle photography and brand storytelling for this cohort.
This is not a reason to remove brand storytelling from product pages. It is a reason to restructure information hierarchy: lead with verifiable facts, follow with brand context.
The trust layer — transparency, reversibility, and the human-in-the-loop imperative
The behavioral science of AI agents introduces a design problem that has no analog in traditional UX: consumers are delegating irreversible financial decisions to systems they do not fully understand. The McKinsey agentic commerce analysis identifies this directly: “This will require transparent decision trails, fail-safes, reversibility, and regional customization of behavior and ethics to ensure agents align with user values.”
For product teams designing agent-compatible checkout flows, this creates a design imperative around transparency. Shoppers need to understand why an agent recommended a product, what criteria were used, and what happens if they want to change the selection. The PayPal agentic commerce protocol overview frames the non-negotiables: “clear consumer consent, trusted identity, reliable payments, and built-in fraud protection.”
Designing for agent-mediated commerce means designing for reversibility, auditability, and human override — not just for conversion.
An Action Plan: What to Do in the Next 90 Days
Strategic clarity is only useful if it translates into sequenced action. The following framework prioritizes investments by impact and addressability.
Weeks 1–2: Audit your machine-readability
Before investing in GEO content or feed optimization, establish a baseline. Use Adobe’s AI Content Visibility Checker to assess which product pages are currently readable by LLMs and which are not. Audit your robots.txt to understand which AI crawlers you are currently allowing or blocking. Review your existing schema markup using Google’s Rich Results Test.
The audit will almost certainly surface one of three failure modes: absent schema, incomplete product attributes at the variant level, or JavaScript-rendered content that agent crawlers cannot parse. Prioritize your highest-revenue product pages for remediation first.
Month 1: Schema, structured data, and feed quality
Implement complete Product schema markup on all priority product pages. At minimum: name, brand, SKU, GTIN/UPC, price, currency, availability, image, and description. Add Review schema where you have user-generated ratings. Validate every implementation against Google’s testing tools before deploying.
Audit your product feed against the platform-specific requirements for ChatGPT’s ACP and Google’s UCP. Correct inconsistencies in product titles, ensure variant-level attribute coverage, and establish a feed update cadence that matches the freshness requirements of each platform. For ChatGPT, this means updates at minimum daily; for Google, the Shopping Graph refreshes continuously.
Implement llms.txt — a structured markdown file pointing AI agents to your most important content. Adoption is low enough that early implementation carries disproportionate visibility benefit.
Months 2–3: GEO content strategy and off-site authority signals
Restructure product descriptions to answer conversational intent queries — the natural language questions a shopper might ask an AI agent. Add use-case scenarios, compatibility details, and contextual applications alongside traditional feature lists. Content structured around questions (“What kind of runner is this shoe designed for?”) has measurably higher citation rates in AI-generated responses.
Build or audit your presence in the third-party directories and review platforms that each AI platform sources from. For Perplexity, this means specialized industry directories. For ChatGPT, this means authoritative editorial mentions and comprehensive product reviews. For Gemini, this means structured, consistent owned-domain content.
Track your progress using AI citation monitoring tools — several have emerged in 2025–2026 that allow you to query AI platforms for your brand and measure citation frequency, accuracy, and sentiment. Establish this as a core visibility metric alongside traditional organic traffic reporting.
Conclusion: Data Beats Creativity — For Now
The most counterintuitive finding in this landscape is also the most important one. In the agentic era, the brands that win AI-mediated recommendations are not necessarily the ones with the best products, the strongest brand identities, or the most persuasive content. They are the ones whose data is cleanest, whose structured attributes are most complete, and whose feeds are most current.
Logicbroker’s agentic commerce analysis makes this explicit: “This new mode of commerce favors systems that can deliver structured product data, API-accessible inventory, fulfillment metadata, and real-time pricing. It deprioritizes traditional marketing touchpoints like website UX, lifestyle imagery, and branded messaging. In this new world, logic and data beat creativity and design.”
This is not a permanent condition. AI agents will become more sophisticated. They will learn to interpret brand signals, evaluate narrative coherence, and weight emotional resonance as a proxy for product quality. The creative advantage will return. But the teams that will be positioned to benefit from that future are the ones who have already built the data infrastructure that makes them visible to agents today.
The funnel has not disappeared. It has been internalized — moved inside the reasoning process of a machine that does not care about your homepage. Your job, for the next 90 days, is to make sure that machine can find you, understand you, and trust you enough to put you on the shortlist.
Because once you’re on the shortlist, the humans still decide.
Sources: McKinsey Agentic Commerce · Adobe AI Traffic Report · FTI Consulting · HUMAN Security 2026 · Mirakl GEO Guide · Opascope ACP/UCP · Yext AI Citation Analysis · Search Engine Land Technical GEO · Quantum Metric Peak 2025 · PayPal Protocol Overview
