Winning the ChatGPT Shelf

In ecommerce, most teams have spent the last decade optimizing for search: SEO, Google Shopping feeds, on-site search, paid keywords, and “people also bought” carousels.

Agentic Commerce changes the game.

ChatGPT, Claude, Gemini and other AI assistants are quickly becoming shopping entry points. OpenAI’s own merchant page now describes ChatGPT as a place where “millions of people use ChatGPT to figure out what to buy” and, with Instant Checkout, “buy directly from you inside those conversations.”

If you want your products to be the ones these agents recommend, you need to think less like an SEO specialist and more like someone fighting for shelf space inside an AI’s brain.

This article is a playbook for doing exactly that.

From Keywords to Questions

Traditional ecommerce discovery is built on keywords and clicks:

User types “best trail running shoes under 150,” scans a list of links, clicks a few, and gradually narrows to one product.

In ChatGPT-style flows, users increasingly skip that dance:

“I run 40 miles a week on mixed trails, budget 150 dollars, I pronate slightly, and I need them by Friday. What should I buy?”

Instead of a page of blue links, they get one to three highly curated options, often with live price and stock, and in some cases a “Buy now” button right in chat thanks to Instant Checkout.

This matters because:

• Each answer is high intent. The user is ready to purchase.
• The agent is doing the work of comparison on their behalf.
• Only a handful of products get surfaced. There is no second page of results.

Your job becomes: be one of those products.

What ChatGPT Actually Sees

A common misconception is that ChatGPT “sees” your website the way a human does. It does not. In the context of Agentic Commerce, it primarily sees:

1. Your structured product feed, formatted using OpenAI’s Product Feed Spec.
2. Checkout state, managed via the Agentic Commerce Protocol (ACP) and your PSP (e.g., Stripe).
3. Additional signals like historical performance, reviews, and merchant policies.

OpenAI’s Product Feed Spec is explicit:

“Provide a structured product feed so ChatGPT accurately indexes and displays your products with up-to-date price and availability.”

The feed can be updated every 15 minutes, which means freshness is no longer a nice-to-have; it’s part of your ranking story.

If your data is missing, malformed, or stale, ChatGPT has three options:

1. Skip your products entirely.
2. Fallback to linking out to your site instead of Instant Checkout.
3. Guess based on similar items (hallucination risk).

You want to be in the subset of products ChatGPT can trust enough to recommend and, ideally, sell directly.

Ranking Signals in Agentic Commerce

OpenAI has said that its product recommendations in ChatGPT are ranked on “price, availability, quality, seller prominence, and checkout ease,” not paid placement.

You can think of the ranking function as a blend of:

• Eligibility – is this product present in the feed with all required fields?
• Relevance – does it match the user’s constraints and preferences?
• Reliability – how often does the data line up with reality (price, stock, shipping)?
• Experience – can the user complete checkout smoothly?
• Reputation – ratings, reviews, and brand trust signals.

Concretely, this means:

• A perfectly structured product that is out of stock will lose.
• A product with a vague description and missing attributes is less likely to be recommended than one with rich, specific data.
• If your checkout fails frequently or shipping estimates are wrong, you may get less exposure over time.

Treat ChatGPT as a new search engine, but one where the snippets are product cards and the click-through is often a completed order.

Designing an AI-Ready Product Catalog

If you’re used to optimizing for Google Shopping, good news: a lot of that work carries over. But OpenAI’s feed collapses what used to be multiple feeds (product, inventory, regional pricing, reviews) into one richer dataset.

Here’s what to focus on.

1. Nail the basics for every eligible SKU

For each product you want AI agents to sell:

• Stable id and canonical link
• Clear title and description that mention key use cases
• price and currency in the correct format
• availability and inventory_quantity kept up to date
• At least one image_link with decent resolution
• Brand, category, and key attributes (size, color, material, etc.)

A surprising number of feeds fall down on the basics: missing GTINs or MPNs, broken links, mismatched currencies, or stale stock. The Product Feed Spec calls these out explicitly as required or recommended fields.

If you wouldn’t ship a product page with “TBD” or “N/A” to a human, don’t ship that to ChatGPT either.

2. Encode what actually matters to buyers

Agents are good at reading prose, but structured hints help a lot. Ask:

• What questions do support or sales teams answer repeatedly?
• What filters do users rely on most in your own search?
• What attributes appear in your best reviews?

Examples:

• For running shoes: pronation support, terrain type, drop, weight.
• For luggage: liters, carry-on compliance, material, lock type.
• For skin care: skin type, active ingredients, fragrance, allergens.

Where possible, turn these into fields or consistent phrases in your feed, not just marketing copy.

3. Normalize across the catalog

AI agents can reason, but they still benefit from clean, consistent categories:

• Map synonyms (“navy,” “midnight”) to unified values (“blue”).
• Align size systems across regions (US vs EU vs UK).
• Ensure your product_category and product_type follow a predictable taxonomy.

A feed with 20 different ways of saying “blue” and 4 different size notation systems is harder for an agent to work with than a normalized one.

4. Keep the feed fresh

OpenAI’s spec supports updates as frequently as every 15 minutes.

You don’t necessarily need to update that often for every field, but you should:

• Update inventory whenever stock changes or thresholds are hit.
• Reflect price changes and promotions promptly.
• Remove discontinued SKUs or mark them as unavailable.

If ChatGPT recommends a product as “in stock” and the Instant Checkout flow fails, that erodes both user trust and the agent’s trust in your data.

A Checklist for Winning the ChatGPT Shelf

Here’s a practical checklist you can use with your team.

Data readiness

• [ ] We have a single source of truth for our product data (or a plan to reconcile feeds).
• [ ] Each eligible product has all required fields from the Product Feed Spec.
• [ ] We’ve identified and cleaned obviously bad data (broken URLs, invalid currencies, etc.).
• [ ] Attributes are normalized (colors, sizes, categories, materials).

Experience readiness

• [ ] Our pricing and inventory data can be updated at least multiple times per day.
• [ ] We can calculate shipping options and taxes programmatically for a given address.
• [ ] Our existing checkout flow can handle orders sourced from external agents (no brittle assumptions about the front-end).

Strategy readiness

• [ ] We know which categories and SKUs we want AI agents to prioritize.
• [ ] We’ve aligned internally on acceptable discounting, shipping SLAs, and return policies for this channel.
• [ ] We have a plan to monitor which SKUs are being shown and which ones are failing eligibility or checkout.

You don’t have to get everything perfect on day one. But you do want to avoid being the merchant whose products never show up or whose checkout fails in front of early adopters.

Measuring and Iterating

One of the biggest challenges with Agentic Commerce today is that the analytics are still immature. You don’t get the same rich dashboard you’re used to from ad platforms or your own site.

Still, you can and should:

• Track orders where source = chatgpt or channel = agent in your order system.
• Log errors and retries from your ACP endpoints (checkout sessions, payments).
• Watch for patterns in which SKUs are selling via ChatGPT and which never appear.

Over time, you’ll learn:

• Which product attributes correlate with being recommended.
• Where your feed is consistently failing validation.
• How agentic buyers differ from your other channels (AOV, return rate, category mix).

Treat this like you treated SEO and paid search in the early days: instrument, learn, iterate.

Final Thoughts

Agentic Commerce isn’t a thought experiment anymore. ChatGPT already lets U.S. users buy directly from Etsy sellers, with support for over a million Shopify merchants rolling out.

Whether you build the feed and eligibility layer in-house or with a specialized tool, the merchants who do this early will:

• Have more of their catalog visible to AI shopping assistants.
• Offer a genuinely better experience with instant checkout in chat.
• Own the “default recommendation” position in their categories before competitors can react.

Winning the ChatGPT shelf is not about clever prompts or one-off experiments. It’s about doing the unglamorous work of making your catalog, data, and checkout truly AI-ready.

The sooner you start, the easier it will be to ride the next wave instead of watching it pass you by.