ChatGPT Product Discovery vs Instant Checkout: What Merchants Need to Do Differently Now

Adobe Analytics, which tracks more than one trillion visits to US retail sites, reported that AI-driven traffic grew 393% year over year in Q1 2026, and that by March 2026 those visitors converted 42% better than non-AI traffic, a full reversal from March 2025 when the same channel converted 38% worse.

B2B Centr tracks the commerce and go-to-market infrastructure shifts that decide which businesses get found, get chosen, and get paid inside AI-mediated buying journeys.

That conversion reversal landed at almost the same moment OpenAI stepped back from in-chat purchasing and refocused ChatGPT on product discovery, leaving merchants with two very different jobs that are still routinely treated as one.

This article breaks down exactly how ChatGPT product discovery and Instant Checkout differ, what changed across 2026, and the specific operational moves merchants need to make now.

Key Takeaways

• OpenAI ended in-chat Instant Checkout in March 2026 and refocused ChatGPT on product discovery.
• Discovery is the durable channel; checkout has moved back to merchant-controlled storefronts and apps.
• Adobe measured AI referral traffic converting 42% better than non-AI traffic by March 2026.
• Product feeds, not crawled web pages, now decide whether ChatGPT surfaces your SKUs.
• Prioritise feed accuracy, landing page parity, and correct AI-referral attribution before protocol work.

The State of AI Commerce Traffic in 2026

The channel stopped being experimental somewhere between the 2025 holiday season and the first quarter of 2026.

Adobe recorded a 693% year over year surge in AI-driven retail traffic during November and December 2025, then 393% growth across Q1 2026, with March alone up 269%. Revenue per visit from AI referrals ran roughly 37% above non-AI traffic by March 2026.

Shopify's own commerce data pointed the same direction. In May 2026 the platform reported that AI-referred sessions converted at nearly 50% higher rates than organic search on product detail pages, that AI outperformed organic search in 23 of 25 merchant categories by an average of 56%, and that AI-referred orders carried 14% higher average order values.

AI referral sessions on Shopify grew more than 8x year over year between January 2025 and March 2026, with AI-attributed orders up nearly 13x.

Volume remains the honest caveat. OpenAI's own research put shopping-related prompts at roughly 50 million per day inside ChatGPT, around 2% of total query volume, against a base of roughly 900 million weekly active users as of February 2026.

A peer-reviewed study published in Marketing Science, covering 973 ecommerce sites with more than $20 billion in combined revenue, found ChatGPT genuinely competitive for complex, considered purchases while remaining a small slice of overall ecommerce referral volume.

The channel is high quality and growing fast, not yet large in absolute terms.

Why Merchants Confused Discovery With Checkout

When OpenAI launched Instant Checkout on 29 September 2025, the framing collapsed two separate capabilities into one announcement.

ChatGPT would recommend products, and ChatGPT would let users buy them without leaving the chat.

Etsy sellers went live first, with more than a million Shopify merchants promised next, powered by the Agentic Commerce Protocol co-developed with Stripe and released under Apache 2.0.

Most merchant teams read that as a single integration decision: get into ChatGPT commerce, or do not. That framing was wrong then and is actively costly now.

Discovery and transaction are separate systems with separate inputs, separate economics, separate failure modes, and, as 2026 proved, separate survival rates.

What Actually Happened: The March 2026 Pivot

On 16 February 2026, OpenAI expanded "Buy it in ChatGPT" to all US users including the free tier. Roughly five weeks later the company reversed course.

In March 2026, OpenAI confirmed it was ending native Instant Checkout, telling press that the initial version did not offer the flexibility it wanted, and that it would let merchants use their own checkout experiences while OpenAI focused its efforts on product discovery.

Three numbers explain the retreat:

1. Adoption never scaled. Out of Shopify's millions of merchants, roughly a dozen had actually gone live with in-chat checkout. Shopify's president publicly placed the bottleneck on the AI platform side rather than merchant willingness.
2. Conversion was worse in-chat. Walmart measured checkout inside ChatGPT converting roughly 3x worse than a click-through to walmart.com, even though ChatGPT delivered around 2x the new-customer rate Walmart sees from search engines.
3. The operational stack was incomplete. As of February 2026, systems for collecting and remitting state sales tax were not in place, a clear signal that transaction volume never justified the build.

Etsy has since described the durable pattern plainly: shoppers who discover products through ChatGPT typically return to Etsy's own site to complete the order.

Shopify replaced the pilot with Agentic Storefronts, which makes eligible products discoverable in ChatGPT by default, links out to the merchant store, requires no dedicated app, and adds no transaction fee beyond standard processing.

ChatGPT Product Discovery vs Instant Checkout: The Core Differences

The strategic read is straightforward. Discovery is where the demand now enters. Checkout is where the margin, the customer relationship, and the compliance burden stay.

Merchants who built toward in-chat transaction in 2025 optimised the wrong half of the funnel.

The Economics: Fees, Margin, and Customer Ownership

Under the original Instant Checkout model, OpenAI charged merchants a reported 4% transaction fee on completed purchases, layered on top of standard payment processing of roughly 2.9% plus $0.30.

On a $100 order that put combined platform and processing cost near $7.20, against typical retail gross margins that leave little room for a new intermediary tier.

Discovery-led routing changes that arithmetic. Under Shopify's Agentic Storefronts model there is no incremental platform fee, and the merchant keeps the checkout surface, the upsell logic, the loyalty enrolment, the payment method mix, and the post-purchase sequence.

For businesses where lifetime value depends on repeat purchase or subscription conversion, ceding first-order customer capture to an assistant was always the expensive option, even at high in-chat conversion rates that never materialised.

There is a second-order cost too. In-chat transactions produce thinner behavioural data. If you cannot see the browse path, the comparison set, or the abandoned variant, your merchandising and ecommerce personalisation models degrade over time.

The Product Feed Is Now Your Storefront

ChatGPT does not primarily crawl your site to build product recommendations. Merchants push a structured feed to an allow-listed OpenAI endpoint over encrypted HTTPS, and that feed becomes the source of truth.

Updates can be sent as often as every 15 minutes, which makes real-time price and inventory accuracy achievable rather than aspirational.

Key specifications merchants should be working against:

• Formats: JSONL is the reference format with the richest attribute coverage. CSV and Parquet are supported, with Parquet aimed at catalogues in the millions of SKUs.
• Spec version: The stable specification released on 30 January 2026 added product extensions, discount fields, and payment handler metadata. The spec versions roughly quarterly.
• Control flags: enable_search and enable_checkout operate independently, letting you gate discovery and transaction eligibility at SKU level.
• Identifiers: GTIN, UPC, or MPN materially improve catalogue matching and reduce mismatch errors.
• Logistics fields: Weight, dimensions, delivery regions, shipping cost, and delivery estimates feed directly into recommendation eligibility.
• Variants: item_group_id, colour, size, gender, and size system prevent variant collapse in results.
• Rich media: Additional images, 3D models in GLB or GLTF, and optional product video, which is unique to OpenAI's spec.
• Descriptions: The field accepts plain text, HTML, and markdown simultaneously, so structured formatting helps the model parse attributes during conversational matching.

Common rejection causes are unglamorous and fixable: missing required fields, non-unique product IDs, marketing copy in place of factual attributes, and price mismatches between feed and landing page.

How ChatGPT Decides Which Products to Surface

Selection is a matching problem, not a ranking auction. The assistant parses a conversational constraint set (budget, use case, size, timeline, aesthetic, compatibility) and matches it against structured attributes.

That has four practical consequences.

• First, attribute completeness beats keyword density. A product missing material, dimensions, or compatibility data simply cannot be matched against prompts that specify them.
• Second, factual descriptions outperform persuasive ones. Copy written to win a click on a search results page performs worse than copy written to answer a specification question.
• Third, corroboration matters. Third-party reviews, comparison content, and editorial mentions feed the model's confidence in recommending you, which is why answer engine optimisation and feed hygiene are complementary rather than competing workstreams.
• Fourth, freshness is a ranking input in effect. If your feed says in stock and your site says sold out, the resulting bounce is a quality signal against you.

Merchants are already seeing the surface expand.

Independent monitoring in August 2026 tracked product card visibility rising around 20% across client sets, with cards per qualifying prompt jumping from a typical one or two to six or eight, alongside local inventory appearing in product and offer cards and testing of a dedicated shopping sidebar.

What Merchants Need to Do Differently Now

1. Reallocate the integration budget. Engineering time earmarked for in-chat transaction plumbing should move to feed quality, catalogue enrichment, and landing page parity. The transaction work has a lower expected return than it did twelve months ago.

2. Fix feed-to-page parity. Every discrepancy between what the assistant states and what the shopper finds on arrival converts a pre-qualified visitor into a bounce. Price, availability, shipping cost, and delivery estimate are the four fields that break trust fastest.

3. Build AI-specific landing experiences. A visitor arriving from ChatGPT has already compared alternatives. They do not need a category page or a top-of-funnel explainer. They need specification confirmation, availability, and a short path to purchase.

4. Instrument attribution properly. A significant share of AI referrals arrive without clean referrer data and get bucketed as direct traffic. Until you tag and segment this channel, you are underinvesting in a source that Adobe measured at 37% higher revenue per visit.

5. Keep checkout agent-ready anyway. Discovery-led routing does not remove the need for machine-readable, low-friction purchase flows. Card networks are wiring their own rails into assistants, so making your checkout AI agent ready remains a live requirement rather than a deferred one.

6. Plan for multi-protocol reality. ACP is OpenAI and Stripe's standard. Google announced the Universal Commerce Protocol in January 2026 with Shopify, Target, Etsy, Walmart, and Wayfair, adding Merchant Center onboarding and cart support in March 2026. Klarna has since adopted it. A useful starting point is this comparison of agent payment protocols.

7. Diversify beyond ChatGPT. ChatGPT's share of US daily AI app users fell from 57% to 42% between August 2025 and February 2026 while Gemini roughly doubled to 25%. In measurable B2B AI referrals across March and April 2026, ChatGPT held 62.6%, Claude 18.5%, Gemini 10.6%, and Perplexity 7.3%. Single-channel strategies are fragile.

Measurement: Getting Credit for AI-Driven Revenue

Three measurement failures are near universal right now.

Misattribution. AI referrals frequently lose referrer headers and land in direct or unassigned buckets, so the channel looks smaller than it is.

Wrong benchmark. Comparing AI referral volume to organic search volume makes the channel look irrelevant. Comparing conversion rate, revenue per visit, and average order value makes it look like a priority. Both are true; only one drives correct decisions.

No visibility baseline. Traffic is a lagging indicator. The leading indicator is how often you appear in the answer at all. Tracking prompt-level presence, card inclusion, and competitor share of recommendations mirrors the discipline covered in this guide to AI search visibility.

Risk, Trust, and the Protocol Layer

The infrastructure underneath agentic transactions is still consolidating. Five standards now compete for the same job: ACP, UCP, AP2, Visa Trusted Agent Protocol, and Mastercard Agent Pay.

Card networks are building rails that would let an agent transact at any participating merchant instead of a short enrolled list, which is a materially different model from merchant-by-merchant onboarding.

The Visa and Mastercard approaches differ meaningfully in how they handle agent identity and authorisation.

Fraud and authorisation remain the hardest unsolved problems, alongside agent identity verification and dispute handling for transactions no human directly initiated. Merchants evaluating exposure should read the current thinking on whether agentic commerce is safe before enabling autonomous purchase paths.

The B2B Angle: Gated Catalogues Are a Discoverability Liability

Most B2B sellers keep pricing and catalogue detail behind a sales-rep gate or a login wall. In a discovery-led model, that is functionally identical to being absent.

If buyers use assistants to build shortlists, and your specifications, compatibility data, and indicative pricing are unreadable, you are excluded before the evaluation starts.

The fix does not require publishing your full rate card. Publish structured specifications, compatibility matrices, minimum order quantities, lead times, and price bands.

Gate the quote, not the qualification data. B2B purchase cycles are exactly the complex, considered decisions where the Marketing Science research found assistants performing best.

Conclusion

B2B Centr exists to give operators a clear read on infrastructure shifts before they become table stakes, and this is one of them.

ChatGPT product discovery and Instant Checkout were never the same capability, and 2026 settled the question of which one merchants should build for.

OpenAI wound down native in-chat purchasing in March 2026 and redirected its effort to discovery, while Shopify made products discoverable by default and routed the transaction back to the merchant.

The channel arriving from that discovery converts better, spends more, and grows faster than almost anything else in the mix.

The merchants who win it will be the ones whose product data is accurate, complete, and refreshed, whose landing pages match what the assistant promised, and whose analytics can actually see the traffic.

Read Next

• What Are Verifiable Credentials in AI Agent Payments? W3C Standards Explained
• Visa Trusted Agent Protocol vs Mastercard Agent Pay: Which Network Layer Wins in 2026?
• Is Agentic Commerce Safe? Fraud and Consumer Protection in AI-Driven Payments

FAQs

1. What is the difference between ChatGPT product discovery and Instant Checkout?

The difference between ChatGPT product discovery and Instant Checkout is where the transaction completes. Product discovery surfaces your products inside conversational answers and routes the shopper to your own storefront, app, or marketplace listing. Instant Checkout allowed the purchase to complete inside the chat window using the Agentic Commerce Protocol. OpenAI wound down native Instant Checkout in March 2026 and refocused on discovery, so discovery is the active channel for most merchants today.

2. Is ChatGPT Instant Checkout still available for merchants in 2026?

ChatGPT Instant Checkout is no longer available in its original native form for most merchants in 2026. OpenAI confirmed in March 2026 that it was allowing merchants to use their own checkout experiences while it focused on product discovery, and moved transactional experiences toward retailer-built ChatGPT apps. The Agentic Commerce Protocol continues to exist and evolve, but in-chat purchase from standard product listings was discontinued.

3. How do merchants get their products to appear in ChatGPT?

Merchants get their products to appear in ChatGPT by submitting a structured product feed rather than relying on site crawling. Apply through OpenAI's merchant portal, receive credentials for an allow-listed HTTPS endpoint, and push a feed in JSONL, CSV, or Parquet format against the current specification. Feeds can be refreshed as often as every 15 minutes. Shopify merchants are enrolled by default through Agentic Storefronts without a separate integration.

4. Does ChatGPT traffic convert better than organic search for ecommerce?

ChatGPT traffic converts better than organic search for ecommerce in most current datasets, though the margin varies by source. Adobe measured AI referral traffic converting 42% better than non-AI traffic in March 2026 with 37% higher revenue per visit. Shopify reported nearly 50% higher conversion than organic search on product detail pages. A study of 94 ecommerce brands found a narrower gap of 1.81% against 1.39%. Absolute volume remains far smaller than organic search.

5. What should merchants prioritise for ChatGPT commerce right now?

Merchants should prioritise feed accuracy, landing page parity, and attribution for ChatGPT commerce right now. Complete every specification field the assistant needs to match a conversational constraint, ensure price and availability match between feed and site, build post-comparison landing experiences for pre-qualified visitors, and tag AI referrals so the channel is not misattributed as direct traffic. Protocol and checkout integration work should follow, not precede, those fundamentals.

Disclaimer

This content is provided for informational purposes only and does not constitute legal, financial, or compliance advice. Protocol versions, governance arrangements, and partner counts cited here reflect publicly announced milestones as of August 2026 and are moving quickly. Adoption figures come from vendor and foundation announcements with differing methodologies and should be treated as directional signals rather than guaranteed outcomes.