AI Shopping Assistants in 2026:
a field guide for enterprise consumer brands.

This guide is for VPs and Directors of ecommerce, digital shelf, and ecommerce content at enterprise consumer brands — leaders planning content investment across multiple AI surfaces and trying to make sense of what's the same, what's different, and where to focus first.

For most of the last two years, "AI shopping assistant" was shorthand for Amazon Rufus. That has changed. As of mid-2026, four major AI shopping assistants are operating at meaningful scale, each with a different mechanic, a different evaluation logic, and a different commercial model behind it. Walmart Sparky is now embedded across the Walmart app, ChatGPT, and Google Gemini. ChatGPT runs the largest AI-native shopping channel by users. Perplexity has shipped agentic checkout to all U.S. users for free. Rufus has crossed 300 million users.

The category is no longer emerging. It is established, multi-player, and reshaping discovery in real time. Salesforce reported that during 2025 Cyber Week alone, AI and agents influenced $67 billion in global sales — 20% of all orders worldwide — with traffic from third-party AI agent channels like ChatGPT and Perplexity tripling year-over-year and converting at 8x the rate of social media.

This piece is the field guide. What each of the four assistants does, how each evaluates content, where each is gaining ground, and what enterprise consumer brands need to think about across all of them — not just the one that happens to be biggest today.

AI shopping assistants in 2026 — no longer emerging, now default

The shift can be stated concisely. AI shopping assistants are now default discovery surfaces, not optional add-ons. Three data points define the moment:

• Amazon's Rufus, embedded in the main Amazon app and search bar, was used by more than 300 million customers during 2025 — up from 250 million at Q3 — with monthly active users growing 149% year-over-year. Amazon reported nearly $12 billion in incremental annualized sales attributed to Rufus in Q4 2025 results.
• Walmart's Sparky, launched in June 2025, has been tried by roughly half of Walmart app users per Walmart's Q4 FY26 earnings call — and Sparky users have ~35% higher average order value than non-users. Sparky is now operating inside ChatGPT and integrating into Google Gemini.
• ChatGPT, with shopping research and Instant Checkout layered on top of OpenAI's roughly 900 million weekly active platform users, is now positioned as a major AI-native commerce surface — though OpenAI has not disclosed how many of those users actively shop through it.
• Perplexity launched its free agentic shopping experience to all U.S. users in November 2025, with PayPal-powered Instant Buy and merchants spanning Wayfair, Abercrombie & Fitch, Ashley Furniture, Adorama, and NewEgg.

The category-level signal: Adobe Analytics reported AI-driven traffic to U.S. retail sites surged 693.4% year-over-year during the 2025 holiday season, with AI referrals converting 31% more than non-AI traffic.

For enterprise consumer brands, this is no longer a question of whether to optimize for AI shopping assistants. The future of ecommerce is being shaped right now across four very different surfaces, and the question is how to do so coherently across all of them.

The three audiences your product page now serves

Before going through the four assistants, it's worth naming the structural shift this whole landscape sits inside. In 2026, every product page is read by three fundamentally different audiences with three different evaluation rubrics — the deeper version of this argument lives in the digital shelf optimization piece, but the short form:

Human shopper — still 85% of traffic

Browsing and evaluating independently. Wins on keyword-rich, benefit-led copy that ranks in retailer search and converts the click. The audience your descriptions have always been written for, and the foundation everything else builds on.

AI-assisted human — Rufus, Sparky, ChatGPT, Perplexity

Around 10–15% of shopping interactions and rising sharply. The shopper is still human, but the interface is conversational and the recommendation is filtered by an AI assistant. This is where all four of the assistants surveyed below operate. What this audience needs from your content is depth of question coverage, persona signals, and claims grounded enough to be cited.

Autonomous AI agent — Buy for Me and the agentic horizon

Less than 1% of traffic today, emerging fast. Amazon's "Buy for Me," Perplexity agentic, and the agent layer being built into Shopify Agentic Storefronts can select and complete a purchase without human review at the point of decision. Their evaluation is programmatic — structured attribute completeness, no contradictions, parity across surfaces. The line between AI-assisted and autonomous is dissolving (Rufus itself now takes agentic actions; Perplexity's agentic experience is now free and live), and content strategy has to anticipate both modes.

The four assistants below all sit in the AI-assisted human persona — but each is built differently, and that matters for how content has to be structured.

Amazon Rufus

For a deeper treatment, see the Amazon Rufus piece in this cluster — this section is a field-guide summary.

Scale and trajectory

Rufus is the largest AI shopping assistant by paid commercial impact today. Amazon's Q4 2025 earnings, reported in February 2026, confirmed:

• More than 300 million customers used Rufus during 2025
• Monthly active users grew 149% year-over-year, total interactions up 210%
• Nearly $12 billion in incremental annualized sales (exceeding the $10 billion Q3 pace)
• Shoppers using Rufus are about 60% more likely to complete a purchase

Andy Jassy described Rufus as "the AI agent" on the Q4 call, signaling that Amazon now positions Rufus as agentic, not just conversational. November 2025 brought 50+ technical upgrades — account memory, automatic cart adds, price alerts, auto-buy at target prices, and the October 2025 launch of "Help Me Decide" (single-product recommendations with AI-generated explanations).

What Rufus evaluates

Rufus evaluates listings on the depth and specificity of how they answer real shopper questions, not on keyword density. Specifically, Rufus weighs:

• Q&A coverage — does the listing's content directly answer common shopper questions?
• Persona signals — explicit mentions of who the product is for and the use cases it fits
• Claim citability — specific, grounded claims an assistant can lift verbatim
• Cross-surface consistency — title, bullets, A+ content, FAQs, and structured attributes telling the same story
• Engagement signals — dwell time, A+ scrolls, review depth, and post-engagement behavior, layered with Amazon's "downstream impact" measurement (a seven-day rolling attribution model)
• Comparison-readiness — sharpened by "Help Me Decide," which lifts language from listings to explain why one product is the pick

What's distinctive

Rufus is the most behaviorally rich of the four assistants — it sits on years of Amazon catalog, review, and shopping behavior data, and increasingly on cross-Amazon signals (Kindle, Prime Video, Audible memory was announced for the months following November 2025). Content that wins on Rufus is descriptive, review-resonant, and dense with answers — not just structured data.

Implications for content strategy

Rufus rewards depth of question coverage and citable claims. PDPs need 6–10 substantive FAQs covering use case, comparison, compatibility, and safety; A+ content with explicit comparison blocks against adjacent SKUs; and language specific enough that an assistant can quote it without paraphrasing. The AI product descriptions piece in this cluster goes deep on the how.

Walmart Sparky

Scale and trajectory

Walmart launched Sparky in June 2025 as a customer-facing GenAI shopping assistant inside the Walmart app. By Walmart's Q4 FY26 earnings call, roughly half of Walmart app users had tried Sparky, and customers using Sparky had average order values about 35% higher than non-users. Walmart's CEO has publicly framed Sparky as the primary vehicle for discovery, shopping, and post-purchase management.

Sparky's reach is no longer limited to the Walmart app. As of late March 2026, Sparky began operating inside ChatGPT, with Google Gemini integration also rolling out. Walmart's stated stance is "openness" — Sparky is being designed to interoperate with other agents rather than lock customers into Walmart only. In fall 2025, Walmart Connect also began testing advertising formats tied to Sparky as agentic shopping experiences scale.

What Sparky evaluates

Sparky's evaluation is more attribute-driven than Rufus. It maps shopper intent to catalog attributes — size, material, use occasion, ingredients, ratings — and pulls from structured data fields. Specifically, Sparky weighs:

• Structured attribute completeness — incomplete attribute fields are a hard surface-out condition
• Listing Quality Score — Walmart's proprietary content quality benchmark, which Sparky leans on heavily
• Price and value signals — Sparky users skew toward value-conscious shoppers, and discount/value content surfaces accordingly
• Multi-step task structure — Sparky is designed for goal-driven queries ("plan a cookout for eight"), so listings that fit a planning task surface more readily
• Multimodal inputs — Sparky processes text, images, and voice, with audio and video in the pipeline

What's distinctive

Sparky is the most agentic of the four assistants by design. Where Rufus answers, Sparky plans, reasons, and acts across Walmart's retail stack — catalog, inventory, pricing, and fulfillment. It is built for multi-step shopping workflows ("plan a unicorn-themed birthday party," "fix this leaky faucet"), not just single-product Q&A. The Kellogg School of Management published a case study on Sparky in early 2026 titled "Agentic AI and the Future of Shopping" — an academic signal that Sparky is being studied as a category-defining agentic system, not a chatbot variant.

Implications for content strategy

For brands selling on Walmart, the unlock is structured-attribute completeness — every filterable field populated, every spec verifiable, every claim attribute-grounded. The content that wins on Rufus is descriptive and review-driven; the content that wins on Sparky is structured and attribute-complete. Brands optimizing only for Amazon and assuming Walmart will follow are systematically losing surface area inside Sparky.

ChatGPT shopping

Scale and trajectory

ChatGPT sits on the largest AI-native platform user base of the four assistants — OpenAI confirmed roughly 900 million weekly active users in February 2026. The shopping experience is exposed to that full user base, though OpenAI has not separately disclosed how many actively shop through ChatGPT. The shopping path has evolved fast through late 2025 and into 2026:

• September 29, 2025: Instant Checkout launched, allowing single-item purchases inside ChatGPT, with Etsy as the first marketplace partner
• October 28, 2025: PayPal partnership announced, opening tens of millions of additional small businesses to ChatGPT commerce in 2026
• Late November 2025: Shopping research feature launched — buyer's guides with multi-product comparisons drawn from web sources
• February 16, 2026: "Buy it in ChatGPT" expanded to all U.S. users (Free, Plus, Pro tiers), with over 1 million Shopify merchants in the onboarding pipeline (Glossier, SKIMS, Spanx, Vuori, and more)
• March 24, 2026: OpenAI revamped the shopping experience after Instant Checkout's initial traction underwhelmed, leaning into discovery and comparison rather than direct purchase

The Instant Checkout transactional experience has had a bumpier ride than the discovery experience — but discovery itself has surged. AI-referred traffic from channels like ChatGPT and Perplexity tripled year-over-year during 2025 Cyber Week and converted at 8x the rate of social media traffic.

What ChatGPT evaluates

ChatGPT does not sit on a retailer catalog. It pulls from the open web, retailer pages, brand sites, third-party reviews, and structured product feeds where they exist. Specifically, ChatGPT weighs:

• Web visibility — content beyond the retailer page (brand site, review sites, comparison content) heavily shapes what ChatGPT surfaces
• Structured product schema — Shopify and Etsy integrations route structured catalog data into ChatGPT directly via the Agentic Commerce Protocol (co-developed with Stripe)
• Comparison and review depth — ChatGPT's shopping research feature builds buyer's guides, so cross-product comparison content matters
• Free-tier accessibility — the experience is available to free users as well as paid, so the surface is broad

What's distinctive

ChatGPT is the only one of the four assistants that isn't anchored to a single retailer. That cuts both ways: brands with strong web-wide presence can surface even when their retailer rankings are weaker, but brands that have over-invested in Amazon and under-invested in their own brand site will be visibility-disadvantaged. The Agentic Commerce Protocol is also a meaningful infrastructure bet — it's the open standard OpenAI and Stripe co-developed to let any merchant become discoverable inside ChatGPT, and PayPal has adopted it.

Implications for content strategy

The brand site matters again. Content investment in ChatGPT means investment in a structured "agent feed" on the brand site (canonical product URLs, attributes, FAQs, last-updated timestamps), strong third-party review presence, and comparison-ready content that sits beyond the retailer wall. Brands with mostly-empty brand microsites — Hellmann's, Rice Krispies, and similar — have meaningful surface area to develop here that isn't available through retailer-only optimization.

Perplexity

Scale and trajectory

Perplexity moved into shopping as a flagship use case through 2025 and 2026:

• May 14, 2025: PayPal partnership announced to power agentic commerce on Perplexity Pro
• November 19–25, 2025: "Buy with Pro" features expanded to all U.S. users for free, with Instant Buy launching alongside Black Friday week
• Merchant base: Wayfair, Abercrombie & Fitch, Ashley Furniture, Fabletics, Adorama, NewEgg, plus 5,000+ additional merchants via BigCommerce and Shopify integrations
• Comet browser: Perplexity's AI-native browser is the launching surface for the agentic experience on desktop and mobile web

Perplexity's commerce ambition is explicit. CBO Dmitry Shevelenko described the model as "the agentic part is the seamless purchase right from the answer." Perplexity wants to be the place people make decisions, with the transaction layer integrated rather than handed off.

What Perplexity evaluates

Perplexity is the most citation-driven of the four assistants. Its core product is an answer engine, and shopping inherits that DNA. Specifically, Perplexity weighs:

• Citable source credibility — Perplexity surfaces sources alongside answers, so authoritative content with strong attribution wins
• Real-time merchant catalog data — via store-sync infrastructure that makes merchant catalogs instantly discoverable in Perplexity results
• Comparison and trade-off content — Perplexity is built for research, and shoppers using it tend toward comparative, considered purchases
• Transparent buyer protection — PayPal serves as the merchant of record, with buyer protection applying to transactions

What's distinctive

Perplexity's commerce experience is the most researcher-coded of the four — its core users come for answer engines, and shopping behavior reflects that. The Comet browser as a distribution surface is also meaningful — it positions Perplexity as a daily destination rather than just a search tab. And the merchant-of-record structure underneath the transaction layer reaches a broad payment-network footprint via PayPal, opening Perplexity to small merchants that don't have the engineering resources for direct AI integrations.

Implications for content strategy

Perplexity rewards content that makes a brand citable. That means structured product feeds with clean merchant-of-record data, transparent comparison and trade-off language, and strong third-party review and editorial coverage. As with ChatGPT, brand site investment matters more for Perplexity than for retailer-walled assistants.

The four assistants compared — at a glance

Where they converge — and what content strategy actually requires

The four assistants differ in catalog source, distribution, and commercial model. They converge on what they reward.

A product page that performs across all four is one that scores well on the same six dimensions, regardless of which assistant is reading it:

These are the same six dimensions that define what Rufus evaluates inside Amazon. They are also what Sparky checks against structured data, what ChatGPT pulls from the web, and what Perplexity cites. The implementation details differ. The underlying rubric does not.

The implication: content strategy is assistant-agnostic. The brands earning citations from Rufus, eligibility from Sparky, surface area from ChatGPT, and citations from Perplexity are not optimizing for four separate surfaces. They are optimizing for the underlying rubric — and the four assistants are all reading that rubric in their own way.

That's why platform-by-platform optimization breaks down. Writing for Rufus, then writing again for Sparky, then writing again for ChatGPT, then writing again for Perplexity — that's four overlapping campaigns producing inconsistent content. Writing once for the underlying rubric, with retailer-specific implementation tuned per surface, is an operating model. The AI product descriptions piece covers the practical mechanics.

The agentic horizon — what comes after AI-assisted shopping

The category framing in this piece treats AI-assisted shopping (Rufus, Sparky, ChatGPT, Perplexity) as the dominant mode in 2026. That framing has a shelf life.

Three forces are pulling all four assistants toward fully autonomous operation:

• Amazon Buy for Me can shop and purchase from external retailers without human review. Rufus itself now takes agentic actions on shoppers' behalf — auto cart-add, reorder, price alerts, auto-buy at target prices.
• Sparky was built agentic from day one. It plans, reasons, and acts across Walmart's stack rather than answering single questions.
• ChatGPT and Perplexity both have functioning agentic commerce stacks (Agentic Commerce Protocol with Stripe; PayPal Instant Buy). Both have shipped to all U.S. users free.

The clean separation between "AI-assisted human" (assistant recommends, human transacts) and "autonomous agent" (agent transacts without human review) is dissolving in practice. By the end of 2026, the distinction may not be meaningful at the consumer level — it will be a setting on a continuum, not a category boundary.

For content strategy, the implication is that descriptions need to be ready for both modes simultaneously. The conversational-citation rubric (Q&A depth, persona signals, claim citability) and the programmatic-eligibility rubric (structured-attribute completeness, contradiction-free claims, parity across surfaces) need to be satisfied on the same page. The AI product descriptions piece goes deep on what that looks like in practice.

Where this leaves enterprise consumer brands

Four AI shopping assistants, four different mechanics, one underlying rubric. The brands earning surface area across all of them in 2026 are doing one thing structurally right: they are not running four campaigns. They are running one operating model that produces content scoring well against the convergent rubric, with retailer-specific implementation tuned to each assistant's evaluation logic.

That operating model is what Genrise is built to do. The platform monitors every SKU across every retailer, scores PDPs on the AI Shelf Readiness Index across five dimensions (Content Foundation, SEO Performance, AI Shelf Visibility, Retailer Algorithm Fit, and Brand's Right to Win), and continuously generates content briefs and copy aligned to all three personas — including the AI-assisted human persona that Rufus, Sparky, ChatGPT, and Perplexity all serve. A/B-tested campaigns across consumer healthcare brands consistently show 0.7% to 6% conversion uplift per SKU within a two-month window, with positive uplift on every test SKU. Across the catalog, continuously improving content quality compounds into 2–5% incremental annual revenue growth.

Optimizing for one assistant is a campaign. Structuring content so it performs across all four continuously, every SKU, every retailer, with humans approving the work — that's the operating model the AI-reader era rewards.