When a product doesn’t show up in ChatGPT, don’t start by rewriting the description. Find the stage that failed first. There are five, and they fail in different ways:
- Admission: is the product allowed into a source ChatGPT reads?
- Retrieval: does that source return it for this kind of search?
- Facts: does the record say what the shopper asked for?
- Market and context: is the shopper somewhere ChatGPT shopping runs, and does their request rule you out?
- Offers: is the product shown, but with another seller?
Log what you checked and the date at every stage. We found no merchant-facing ChatGPT report in OpenAI’s documentation, so most checks run through Shopify or your own feed. And one missing result in your own ChatGPT chat proves little: OpenAI says results depend on the shopper’s query and context, including Memory, and that “not all available products will necessarily be shown”. Checked October 27, 2026; re-check before acting.
First, check what “missing” means
Before the tree, pin down the symptom. Three different complaints get called “missing from ChatGPT”:
| What you saw | What it could mean | Where to start |
|---|---|---|
| Your product never appears for any search you try | Admission, or retrieval for every query | Stage 1 |
| It appears for its name but not for a need (“quiet humidifier for a big room”) | Retrieval or facts for that need | Stage 2, then 3 |
| The product appears, but with another store | Offers | Stage 5 |
Write down the exact request you typed, the date, whether you were signed in, and whether your account has Memory or Custom instructions. That’s your first log entry. OpenAI’s own example shows why: if a shopper has said they dislike clowns, ChatGPT “may leave out clown costumes”. Your chat history is context too.
The troubleshooting tree
Work down the stages. At each one, answer the question. A “no” stops you there: fix it, or record why you can’t, before moving on. Rows marked “documented” come from a current OpenAI or Shopify page. Everything else is our proposed method.
Stage 1: Admission
| Question | How to check | If the answer is no |
|---|---|---|
| Which route are you on? | Shopify or Etsy: catalog already integrated, says OpenAI. Anyone else: a direct feed, after approval | Not on Shopify or Etsy and not approved? You’re on the waitlist, or relying on the open web. Nothing to tune yet |
| Shopify: is the store eligible for Shopify Catalog? | Shopify says Sales channels > Agentic shows a product count or “Ineligible” next to Shopify Catalog (documented; screen not verified by us) | Click “Ineligible” for the reason. Shopify publishes each message and its fix. See Shopify Catalog requirements |
| Shopify: plan? | Starter or higher, or the Agentic-plan exception for stores without a paid-plan online store (documented) | Change plan, or confirm you qualify for the exception |
| Shopify: does the store meet the ChatGPT requirements? | Sells to US customers; Terms of service, Privacy policy and Return and refund policy completed; Supplemental Terms accepted (documented) | Complete the missing item |
| Shopify: has ChatGPT’s access been switched off? | The ChatGPT channel setting. Turning access off can take up to 7 days to apply (documented) | Turn it back on if that was a mistake. Shopify documents a delay only for turning access off; how long turning it back on takes isn’t documented |
| Shopify: does this product pass the product rules? | Title, at least one image, price above zero, published with a URL, not Unlisted, not hidden from search engines (documented) | Fix the product. A free product is excluded |
| Is it in an excluded or prohibited category? | Shopify’s excluded list, and OpenAI’s prohibited products. Shopify classifies from your text, and excludes “Uncategorized” (documented) | If the category is excluded, stop: no copy change fixes that. If it’s misclassified, make the title and description say plainly what it is |
| Is it B2B-only, password-protected or behind a login? | Shopify excludes these (documented) | Expected behaviour |
| Direct feed: did the row ingest? | Required fields present; availability one of the allowed values; no non-spec field names; is_eligible_search not false (documented) |
Fix the row. A blank or unrecognized availability rejects it |
Stage 2: Retrieval
| Question | How to check | If the answer is no |
|---|---|---|
| Does your source return the product for a plain search by its name? | Shopify: the search preview in the Agentic section shows raw Shopify Catalog results (documented) | If even the name doesn’t find it, go back to Stage 1 |
| Does it come back for the need, phrased the way customers ask? | The same preview, with a real customer request | Note where it ranks. Shopify says assistants “often re-rank” these results, so treat it as directional |
| Did anything change recently? | Your change log: price, stock, title, Unlisted status, channel settings | OpenAI says price and shipping updates may lag. A removed feed row can persist up to 14 days |
The retrieval stage is the one you can see least. A Shopify Catalog result isn’t a ChatGPT result. Our note on Catalog research shows how to word a finding so it doesn’t claim more than it can.
Stage 3: Facts
| Question | How to check | If the answer is no |
|---|---|---|
| Does the record state each requirement in the shopper’s request? | Split the request into requirements and look for each in the title and description, as text with a unit | Add the fact if you have it, with its condition |
| Does a stated fact rule you out? | The same read | That’s a true no, not a bug. Leave it |
| Is the key fact near the start? | Its position in the description. OpenAI says ChatGPT “may generate simplified product titles and descriptions” | Move it up |
| Do your own copies disagree? | Your description against Shopify’s generated fields, which Shopify marks Inferred | Settle the true value at the source |
Stage 4: Market and context
| Question | How to check | If the answer is no |
|---|---|---|
| Is the shopper in a market where ChatGPT shopping runs? | OpenAI: “currently live for ChatGPT users in the U.S.” | Canada isn’t documented as live in the pages we read |
| Did the shopper’s context exclude you, such as a budget or a stated dislike? | Re-read the request. A stated budget makes ChatGPT “focus more on price” | You can’t change a shopper’s context. You can make your price and facts easy to compare |
Stage 5: Offers
| Question | How to check | If the answer is no |
|---|---|---|
| Is the same product sold by other stores? | Search your product’s exact name and model | If so, ChatGPT may list several merchants for it |
| Are your price and stock current? | Your admin, today | Fix them; changes may take time to appear |
| Are you the maker or primary seller? | Your own answer | OpenAI names maker or primary-seller status as a factor. A reseller may sit lower on the list |
Shopify’s Global Catalog clusters results by Universal Product ID and includes offers from several merchants, so identical products are often grouped. If you resell a product someone else makes, Stage 5 is your realistic ceiling.
The evidence log
Keep one row per check. A dated log turns “ChatGPT ignores us” into something you can act on, or hand to Shopify support.
| Date | Stage and question | What you saw, quoted | Next action |
|---|---|---|---|
| [date] | 1: Catalog eligible? | [count, or the Ineligible message] | [fix, or none] |
| [date] | 2: Returned for the need? | [preview position, or not returned] | |
| [date] | 3: Fact stated? | [quote and field] | |
| [date] | 4: Market and context | [where the shopper is; budget or other constraint] | |
| [date] | 5: Other sellers? | [who, at what price] |
Worked case: the humidifier that “never shows up”
This is a labelled hypothetical, built on one real record. Imagine a Shopify store selling a top-fill humidifier. A customer emailed that ChatGPT never suggests it when they ask for a “humidifier for dry winter air large room quiet top fill”.
For the record, we use position 4 of the Shopify Catalog results we saved for that exact search on October 26, 2026. Its description rates the unit for single rooms up to 400 square feet and says it “runs quietly enough to leave on overnight”, with no decibel figure.
Stage 1. The store checks its Agentic section: eligible, product counted, policies complete, sells to the US. Pass. Log it.
Stage 2. The product came back in Shopify’s catalog for this exact search, at position 4. So it isn’t missing from the source. Whatever ChatGPT does next, it re-ranks a list this product is already on. Pass, with that caveat.
Stage 3. Here’s the finding. “Large room” is in the request, and the record says 400 square feet. For a shopper with a 450 sq ft room, that’s a stated no. For “quiet”, the record has only words. Among the ten results, a decibel figure appeared in the description of 4 (positions 2, 3, 5 and 7), and in the description or Shopify-generated metadata of 6 (adding 6 and 9). It appeared in no title. So other listings gave a number a reader could compare, and this one gave an adjective.
Stage 4. The customer is in the US. Pass.
Stage 5. Not reached. The product isn’t losing a seller contest; it’s losing on facts.
Diagnosis. Two different problems, with two different fixes. The 400 sq ft limit is true, so it should stay: it sends large-room shoppers elsewhere, which beats a return. The noise claim is fixable only if the store has a measurement. If it does, put the decibel figure and the setting it was measured on in the first sentence of the description. If it doesn’t, the honest line is that noise wasn’t measured. Don’t invent a number.
Rewriting the whole description would have changed neither fact.
Common misdiagnoses
| Symptom | Tempting fix | What the documents say |
|---|---|---|
| “I blocked AI crawlers, so ChatGPT can’t see me” | Unblock everything | Shopify says robots.txt affects only open-web discovery and doesn’t stop Shopify Catalog sending your data |
| “I need an llms.txt file” | Install an app | Shopify already serves /agents.md, /llms.txt and /llms-full.txt for every store, and says they don’t replace Shopify Catalog |
| “I need to apply to OpenAI” | Fill in the merchant form | OpenAI says Shopify and Etsy catalogs need no application |
| “I need more reviews” | Chase a review count | OpenAI names reviews as one input and publishes no threshold. Shopify’s listing quality counts only reviews verified by trusted sources |
| “A competitor shows up with my product” | Rewrite the description | That’s the seller list: availability, price, quality and maker status |
| “I removed ChatGPT’s access yesterday and my products still show” | Assume the setting is broken | Shopify says turning off Catalog access can take up to 7 days, and products can still be found through web crawling |
For the positive version of this sequence, see ChatGPT product recommendations. For direct-feed rows, see the ChatGPT product feed guide. For a full store review, see the AI visibility audit.
How we checked
On October 27, 2026 we read OpenAI’s help article on shopping, its merchant page, its product feed specification and file upload overview, and Shopify’s pages on selling on ChatGPT, Shopify Catalog and product discovery, managing agentic storefronts, Catalog requirements and the Global Catalog. Every “documented” row paraphrases those pages as they read that day. We didn’t log into any Shopify admin or OpenAI partner account. The worked case uses one anonymized record from our saved Shopify Catalog search of October 26, 2026, with no buyer or country context.
Your next step
Copy the evidence log. Pick the one product a customer told you they couldn’t find, and fill in Stage 1 today, with the date and what your admin actually says. Don’t touch the copy until Stages 1 and 2 have a row each.