The price that ChatGPT shows has to live in the HTML your server sends, not the JavaScript that paints it afterward. The Semrush Blog guide to ChatGPT shopping surfaces a 2024 Vercel test that found OpenAI crawlers do not execute JavaScript, and a 2026 experiment where ChatGPT summarized only the raw HTML when a user pasted a URL.
That detail matters because the surface changed. OpenAI launched shopping inside ChatGPT in 2025 with product cards pulled from the web, then added Instant Checkout through its Agentic Commerce Protocol to pass payment details to merchants. In March 2026 OpenAI said it was stepping back from Instant Checkout for lack of flexibility to focus on discovery, and it expanded ACP so merchants share catalog and promotion data instead of closing the sale inside the chat.
The plumbing now has two paths. Approved merchants submit a product feed to OpenAI over ACP via SFTP, file upload or API, or through providers like Stripe and Salesforce, with OpenAI recommending a full daily file and intraday updates via API for price and promotions, while Shopify and Etsy catalogs are already integrated. Everyone else is discovered from the open web, which is why the nine required feed fields like item_id, title, description, url, brand and price have to match what the product page itself renders as text.
If you do not rank in Google Shopping, you rarely rank in ChatGPT.
A Semrush experiment cited in the guide found the top ChatGPT recommendation sat in the top three Google Shopping results 75% of the time, and a March 2026 Peec AI study found about 83% of ChatGPT carousel items also appeared in the top 40 organic Google products, so the guide pushes the same fundamentals: accurate titles, GTIN and MPN identifiers, extra images, clean Merchant Center diagnostics and real review text on the page. That also means schema matters, from basic Product offer markup to recommended fields like material, color and shipping details and ProductGroup for variants, plus a clear H1 for the product name and tables for specs instead of images of specs. On the crawler side you need to allow OAI-SearchBot in robots.txt and any Cloudflare bot rules, and you can still block GPTBot to opt out of training, with the practical setup being to allow both OAI-SearchBot and ChatGPT-User so ChatGPT can surface and fetch the page while deciding on GPTBot separately, and to use OpenAI’s published IP ranges for OAI-SearchBot and OAI-AdsBot when configuring access. An agent readiness scanner is the fast way to catch a blocked crawler before you chase content tweaks.
OpenAI says shopping results are not ads and it labels ads separately, and it tells merchants to keep feed descriptions factual while the richer, fan-out aware copy lives on the site, like calling out stainless steel for odor and cleaning when users ask for the best cat fountain. The guide notes shopping is live only in the U.S. for now and that optional feed fields like condition, dimensions, age_group and star_rating are not confirmed by OpenAI to boost visibility, so the immediate win is making pages machine readable and consistent across feed, schema and on-page text.
