ChatGPT builds a shopping answer from three inputs: its own reading of public retail pages, the structured product data merchants submit through OpenAI's Agentic Commerce Protocol, and the model's synthesis of both, filtered by OpenAI's product policies. The product card, the rich result with an image, price, and rating that sits inside the answer, goes to the product that is accurately described in structured data, corroborated by third-party sources ChatGPT trusts, and ranked well on the specific searches ChatGPT runs behind your question. As of July 2026, OpenAI states these results are "organic and unsponsored, ranked purely on relevance to the user," so no vendor can sell you a guaranteed spot in them.
Last verified: July 2026.
Key takeaways
- ChatGPT assembles shopping answers from three inputs: its reading of public retail pages, structured product feeds submitted through OpenAI's Agentic Commerce Protocol, and the model's own synthesis under OpenAI's policies.
- Product results are organic and unsponsored per OpenAI; you cannot buy your way into them, and any guaranteed-placement claim contradicts OpenAI's stated position.
- OpenAI ranks merchants on availability, price, quality, and whether they are the maker or primary seller, so feed accuracy and stock are ranking inputs, not just operations.
- The card is built from feed fields plus off-site corroboration, so schema accuracy and third-party reviews both matter, and the review summaries shown are model-written from public sites.
- Optimization is a loop: baseline product visibility, fix the feed, strengthen PDP answers, win the fan-out searches, earn citations, monitor rendering, and attribute to revenue, then repeat.
- Rich product cards are a ChatGPT-led surface today; other engines render product information differently, so measure ChatGPT as the primary card surface.
ChatGPT accounts for 96.1% of United States large language model (LLM) referral traffic to online stores, according to Alhena's 2026 cohort research, which is why ChatGPT shopping optimization is now its own discipline rather than a footnote to search engine optimization (SEO). This guide explains how the product card is assembled, what each part of it is built from, and a seven-step playbook to compete for it. It is deliberately not a feed-submission tutorial: for the field-by-field mechanics of getting your catalog into ChatGPT, see the companion ChatGPT product feed guide.
Disclosure: Alhena AI publishes this guide and sells an AI visibility product mentioned in it. Statements about how ChatGPT selects and displays products come from OpenAI's own public pages and developer documentation, linked at first use and verified in July 2026. OpenAI does not publish its full ranking function; where the mechanics are not disclosed, this guide says so rather than guessing.
How does ChatGPT choose which products to show?
ChatGPT chooses products from three sources at once. It reads publicly available retail pages directly and cites them, it ingests structured catalog data that merchants push through OpenAI's Agentic Commerce Protocol, and it applies its own model reasoning and safety policies to decide what is relevant and safe to surface. OpenAI describes the selection as drawing on "structured metadata from first-party and third-party providers," the model's own responses, and OpenAI's product policies. In practice this means two doors into the card: your own product feed, and the wider web's description of your product.
When several merchants sell the same item, OpenAI says ChatGPT ranks them on "availability, price, quality, and whether they are the maker or primary seller of that item," and also considers whether Instant Checkout is enabled. Being the brand that makes the product, keeping it in stock, and pricing it competitively are therefore ranking inputs, not just merchandising choices.
These results are organic. OpenAI states plainly that product results are "not ads, nor influenced by any OpenAI partnerships." You cannot pay to appear in the organic product results, and any tool promising guaranteed ChatGPT placement is describing something OpenAI says does not exist. The lever you do control is the quality and accuracy of the signals ChatGPT reads.
ChatGPT also summarizes reviews inside the shopping experience. OpenAI notes that ChatGPT "may also display product review summaries" that are model-generated from reviews on public websites. That makes your third-party review footprint, not just your on-site star rating, part of what the model repeats back to a shopper.
What is a ChatGPT product card built from?
A ChatGPT product card is assembled from a small set of attributes, each drawn from a specific source. The table below maps every visible element of the card to where ChatGPT gets it and the lever you use to influence it. The source column is what tells you whether a feed fix or an off-site effort moves the needle.
| Card element | What ChatGPT shows | Where ChatGPT gets it | Your optimization lever |
|---|---|---|---|
| Title and image | Product name and primary photo | Feed title and image_url fields, plus the product detail page (PDP) | Clear, specific titles; a clean primary image in the feed and on the PDP |
| Price | Displayed price, with a lowest-price flag across merchants | Feed price field and third-party price providers | Accurate, current pricing; frequent feed refresh so the shown price matches your site |
| Availability | In stock, out of stock, pre-order, or backorder | Feed availability field | Real-time stock accuracy through catalog sync |
| Star rating and reviews | Rating plus a model-written review summary | Feed review fields and reviews on public websites | Structured ratings in the feed; a healthy third-party review presence |
| Brand and seller | Who makes and sells the item | Feed brand, seller_name, and seller_url fields | Being identifiable as the maker or primary seller |
| Product answers and features | Key features and buyer-question answers | Feed q_and_a field and PDP content | FAQ blocks on the PDP that answer real buyer questions |
| Corroborating citations | The sources ChatGPT read to build the answer | Public retail pages, review sites, and listicles | Earning mentions on the third-party pages ChatGPT trusts |
The field names above come from OpenAI's product feed specification, which defines required attributes such as item_id, title, price, availability, brand, and image_url, and recommended ones including gtin, review fields, and a q_and_a block for buyer questions. Two facts about the card follow from that spec. First, ChatGPT reads structured data, so schema accuracy is a direct ranking input rather than a nice-to-have; the same logic that governs product schema for Google applies here, and Alhena's schema markup guide for generative engines covers the on-page side. Second, the card rewards being the clearly identified maker or primary seller, because OpenAI names that as a ranking factor. For the step-by-step of building and submitting the feed itself, use the ChatGPT product feed guide; this article picks up where that leaves off.
The seven-step ChatGPT shopping optimization playbook
Optimization for ChatGPT is a loop, not a one-time submission. The seven steps below move from measurement to feed hygiene to off-site corroboration and back to measurement, because the only honest signal that any change worked is whether more of your products show up and convert.
Step 1: Baseline which of your products ChatGPT already recommends
Start by measuring, because you cannot optimize a card you have never seen. The manual version is to run your top buyer prompts in ChatGPT ("best waterproof hiking boots for wide feet," "gentle vitamin C serum for sensitive skin") and record which of your products appear, in what position, and whether a full card or only a text mention shows. The scalable version is SKU-level tracking, which watches product-card presence per prompt and per topic instead of counting brand mentions. Alhena's AI visibility platform flags which specific products surface, marks topics where none of yours appear as opportunities, and detects invisible bestsellers, products that sell well on your site but never appear in AI answers. Brand-level tracking cannot catch that gap, which is the subject of the companion piece on SKU-level AI visibility. If you already run an on-site assistant, mine its conversations first: the questions shoppers ask your Shopping Assistant or Support Concierge (web chat, email, Instagram DMs, WhatsApp) are exactly the prompts worth baselining, and Alhena feeds them into prompt tracking automatically.
Step 2: Fix your product feed and structured data first
Feed accuracy is the highest-leverage fix because price and availability are ranking factors and OpenAI shows the price it received from a provider, not the price on your live PDP if the two disagree. Get the required feed attributes right, keep price and stock synchronized in near real time, and populate the recommended fields, especially gtin and reviews, so ChatGPT can match your item to the same product sold elsewhere. A live catalog sync from Shopify, WooCommerce, Magento, or Salesforce Commerce Cloud is what keeps the shown price honest between refreshes. The feed mechanics live in the ChatGPT product feed guide; do not skip this step, because every later step is undermined by a stale or inaccurate feed.
Step 3: Make your product pages answer the buyer's actual question
ChatGPT reads your PDP directly and its feed spec includes a q_and_a field, so the buyer questions you answer on the page become raw material for the card. Add FAQ content that resolves the real objections in your category, such as fit, ingredients, compatibility, returns, and sizing, in plain language a model can extract. This is the same content discipline that AI engines reward across the board, and Alhena's FAQ and content engine generates and audits these blocks against the questions shoppers actually ask, rather than the ones a brand wishes they asked.
Step 4: Win the fan-out searches behind the answer
Before ChatGPT answers a shopping question it runs several web searches of its own, and your product can only be read if it ranks on those underlying searches. These are called fan-out queries, and your AI visibility is downstream of your organic position on them, which is the bridge between classical SEO and answer engine optimization (AEO) that most "SEO is dead" takes miss. Map the fan-out searches your buyer prompts trigger, find the ones where you are absent from the top results, and create or strengthen pages to rank on them. The companion guide on fan-out queries explains how to find and prioritize these searches.
Step 5: Earn the third-party citations ChatGPT trusts
Because ChatGPT reads public retail pages, review sites, and listicles, and cites them, the pages that mention your product off-site are part of your card. A strong presence on the review platforms and category roundups your buyers already trust gives ChatGPT more corroboration to pull from, and it feeds the model-written review summaries OpenAI displays. This is off-site work that no feed setting can substitute for: the card rewards being genuinely well-regarded on the open web, not just well-described in your own feed.
Step 6: Monitor how your card actually renders
Appearing is not the same as appearing correctly. A shopper who sees a wrong price, a stale out-of-stock flag, or a missing image on your card is a shopper you lose at the moment of highest intent. Rendering analysis checks whether a full product card or only a text mention appeared, and whether the price shown matches reality. Alhena distinguishes shopping-carousel cards from text mentions and surfaces where your rendered price, image, or card position is off, so a feed error becomes a fixable alert instead of a silent leak. One honest limit: rich product cards are a ChatGPT-led surface today, and other engines such as Google AI Overviews, Perplexity, Gemini, and Claude render product information differently, so monitor ChatGPT as the primary card surface while treating the others as text-answer visibility.
Step 7: Attribute and iterate on revenue, not mentions
A mention is a vanity metric until it produces a checkout. Close the loop by attributing AI-engine traffic to actual purchase behavior, so you optimize for the products and prompts that drive revenue rather than the ones that merely get named. Alhena classifies referral traffic from ChatGPT and other engines and joins it to checkout events, then benchmarks it against your sitewide baseline, which turns "we appear in ChatGPT" into "ChatGPT-referred sessions convert at this rate versus our site average." Treat this as attribution, not proof of incremental lift: it tells you which AI-sourced sessions convert, not that ChatGPT caused a sale that would not have happened otherwise. That distinction is covered in the companion guide on AI search revenue attribution. With revenue in view, return to Step 1 and re-baseline.
What you cannot control in ChatGPT shopping, and which claims to distrust
OpenAI does not publish its full ranking function, so any playbook that promises a formula for the top slot is overreaching. The steps above optimize the inputs OpenAI has disclosed, and no more. Results also vary by user, session, and history, because ChatGPT personalizes to shopper preferences and past interactions, so two people asking the same question may see different cards and your own test is not a universal reading.
You also cannot pay for the organic product results. OpenAI's commerce feed spec does contain an is_ads_eligible flag, which signals that paid formats may develop separately over time, but as of July 2026 the product discovery results are the organic, unsponsored kind, and a paid pathway would be distinct from them rather than a way to buy up the organic ranking. Treat "guaranteed placement" and "pay to rank in ChatGPT" pitches with the same skepticism OpenAI's own wording invites.
Finally, price and stock accuracy is a trust liability as much as a ranking factor. OpenAI cautions that prices, stock, and discounts change frequently and may not match the retailer's page, which is precisely why feed freshness and rendering monitoring, Steps 2 and 6, protect conversion at the moment a shopper is deciding.
The bottom line: win the card by being the best-corroborated answer
ChatGPT shopping optimization is not a growth hack layered on top of a product; it is the discipline of being, and provably being, the best-described and best-corroborated answer to a shopper's real question. The brands that win the card keep their feed accurate to the cent, answer buyer questions on the page, earn genuine third-party trust, rank on the searches ChatGPT runs behind the scenes, and then measure the result in checkouts rather than mentions. Everything OpenAI has disclosed about ranking rewards that behavior, and nothing it has disclosed rewards a shortcut. The work is the strategy.
Alhena AI, founded in 2022 by ex-LinkedIn and Meta engineers, is an Agentic commerce AI platform whose product suite spans AI Shopping Agents, Support Concierge, Voice AI, and AI Visibility for AEO and generative engine optimization (GEO) tracking. Its AI visibility product tracks how brands and specific products appear across ChatGPT, Google AI Overviews, Gemini, Perplexity, and Claude, with a free tier and paid plans from $199 per month. Because Alhena also runs on-site shopping and support agents, it can join AI visibility to purchase outcomes in one system, which is what makes revenue attribution in Step 7 possible rather than aspirational.
About the publisher: Alhena AI, founded in 2022 by ex-LinkedIn and Meta engineers, is an Agentic commerce AI platform on a mission to make online shopping more fun, efficient and social. Alhena's product suite spans AI Shopping Agents, Support Concierge, Voice AI, and AI Visibility (AEO and GEO tracking).
