How to Connect Your Shopify Products to ChatGPT with Nuso Feeds

Why AI shopping surfaces matter now

Something significant is happening to product discovery. For fifteen years, the playbook for a DTC brand was stable: get found on Google, get seen on Meta, convert on your own site. But a growing share of purchase journeys now starts inside an AI assistant. Someone asks ChatGPT for "a linen shirt under £80 that ships to the UK" and gets back a shortlist of actual products — with images, prices, and links to buy.

The brands that appear in those answers are the brands whose product data ChatGPT has. It's that simple. This isn't a channel where clever creative or a big ad budget wins — at least not yet. It's a channel where being present, with accurate, well-structured product data, is most of the battle. And as with every new discovery channel, the brands that show up early build an advantage that's hard to claw back later.

The mechanics are straightforward: ChatGPT's merchant programme lets you connect a product feed — a file describing your catalogue in a standard format — which ChatGPT imports on a schedule and uses to surface your products in shopping conversations. The catch, historically, has been the feed itself. Exporting your catalogue, reshaping it into the required format, keeping prices and stock levels current as your store changes — that's a recurring chore that most lean teams quietly deprioritise. Which means stale prices, out-of-stock products still being recommended, and new launches invisible for weeks.

That's the problem Feeds solves.

What Nuso Feeds does

Nuso already syncs your full Shopify catalogue every 30 minutes — it's the same product data that powers your analytics, catalogue views, and range planning. Feeds puts that data to work as a distribution asset: it generates a live product feed directly from your synced catalogue, served from a secret URL that looks like this:

https://nuso.co.uk/feeds/<token>/products.csv

Three things make this different from a manual export:

• It's always fresh. The feed regenerates on every fetch from your latest synced data. Because Nuso's product sync runs every 30 minutes, whatever pulls the feed — ChatGPT, Google, Meta — always gets prices, stock levels, and images that are at most about half an hour behind your store.
• It's spec-compliant out of the box. The ChatGPT preset follows the OpenAI product feed specification: one row per variant, with title, description, product URL, image, price formatted like 79.99 GBP, sale price derived from your Shopify compare-at price, availability (in_stock, out_of_stock, or backorder), brand, GTIN where your barcode is a valid GS1 code, variant grouping with option names like colour and size, and extra images. No column mapping, no format wrangling.
• It's secured by the URL itself. The token in the URL is the secret — there are no credentials to manage, and nothing to configure on the receiving end.

You'll find Feeds in a new page under Marketing → Feeds, and there's a quick-access card in Settings → Integrations. Each feed is available in three formats — JSONL, CSV, and TSV — and you can attach rules to control exactly which products it contains.

Step-by-step: connect your catalogue to ChatGPT

Step 1: Create your feed in Nuso

1. In Nuso, go to Marketing → Feeds and click New feed.
2. Choose the ChatGPT preset. This locks the output to the OpenAI product feed spec, so the file is formatted exactly as ChatGPT expects.
3. Give the feed a name — something like "ChatGPT — full catalogue" so you can tell it apart later.
4. Optionally, add rules to shape what's included. Rules can include or exclude products by tag, product type, vendor, or stock status. Two we see used often: exclude anything tagged sample, and exclude out-of-stock products so ChatGPT never recommends something a customer can't buy.
5. Click Create, then copy the CSV URL from the feed's detail view.

Step 2: Paste the URL into ChatGPT's merchant setup

In ChatGPT's merchant setup, choose the option to connect your feed via URL. In the File URL field, paste the URL exactly as you copied it from Nuso.

One detail genuinely matters here: use the CSV URL — the one ending in .csv. ChatGPT's URL importer currently rejects the JSONL version of the feed ("Connection failed"), even though JSONL is the format OpenAI's file-upload spec describes — the JSONL and TSV URLs remain available for API-side consumers and other channels. And copy the URL whole: the most common connection failure we've seen is a URL retyped by hand or truncated. Paste, don't type.

Leave the HTTP Basic auth fields blank. Your Nuso feed doesn't use basic auth — the secret token inside the URL is what protects it. Filling in credentials here will cause the fetch to fail.

Step 3: Preview and verify

Once ChatGPT fetches the file, it shows you a preview of the products it detected. Give this a sanity check: do the product count, titles, prices, and images look right? If you added exclusion rules in Step 1, confirm the excluded products aren't there. This is the moment to catch anything odd — before your catalogue goes live in shopping results.

Step 4: Set the import schedule to daily

Set ChatGPT's import schedule to daily. Here's why that's enough: your Nuso feed isn't a static file that goes stale between imports. It regenerates from your synced catalogue every time it's fetched, and the sync runs every 30 minutes — so at the moment ChatGPT pulls it each day, it reflects your store as of roughly half an hour ago at most. Price changes, stock movements, new images: all picked up automatically, with no one on your team touching an export ever again.

What gets included in your feed

By default, a feed contains your active, published products that have a price and an image — the products a customer could actually see and buy on your storefront. Each variant is its own row in the file, and variants are grouped under their parent product with their option names (colour, size, and so on), so ChatGPT understands that "Linen Shirt — Sage / M" and "Linen Shirt — Sage / L" are the same product in different sizes, not two separate products.

GTINs are included only when the barcode on the variant is a valid GS1 code — an invalid or made-up barcode is worse than none, so Nuso checks rather than passing through whatever is in the field. Sale prices come from your Shopify compare-at price: if a variant's compare-at is higher than its price, the feed reports the compare-at as the regular price and the current price as the sale price, which is how shopping surfaces expect discounts to be described.

Troubleshooting

Beyond ChatGPT: Google Shopping and Meta

ChatGPT is the newest surface, but the same feed engine covers the established ones too. Feeds ships with presets for Google Shopping and Meta (Facebook & Instagram catalogues), each formatted to that platform's specification. You can run all three from the same synced catalogue — one feed per destination, each with its own URL and its own rules.

That last part is worth pausing on. Because rules are per-feed, you can send different slices of your catalogue to different platforms: your full range to Google, only full-price products to Meta, everything in stock to ChatGPT. One source of truth, three tailored feeds, zero exports.

Product discovery is fragmenting, and the operational cost of showing up everywhere used to be real — a feed per platform, each one a maintenance burden. Feeds collapses that cost to a one-time, ten-minute setup per platform. If AI shopping becomes the channel we think it will, you'll want your catalogue to have been there from the start.