# ChatGPT Business Hub Product Feeds: CSV Setup Guide

> Which feed Business Hub needs, how to deliver it (CSV, hosted URL, Shopify, SFTP), the required columns, and how to fix the rows the importer rejects.

Canonical: https://revnu.com/blog/chatgpt-business-hub-product-feeds-csv-setup-guide
Published November 19, 2026 by George Jefferson, Co-founder and CEO, Revnu

# ChatGPT Business Hub Product Feeds: CSV Setup Guide

A ChatGPT Business Hub product feed CSV is a file with one row per purchasable item, carrying the same required columns on every row. You upload it, host it at a URL, connect it through Shopify or send it over SFTP once your business profile is approved. Most rejected feeds fail for the same few reasons: a missing required column, a price that is not written as a number plus a currency code, an availability value outside the allowed list, or variants that share an ID. This guide covers the file you need, how to deliver it, and a row-by-row path for fixing it when the importer says no.

As checked on 18 November 2026: Business Hub is in beta, and the accepted file types, field lists and delivery options below come from OpenAI's help pages plus the setup screens and published schema notes available on that date. Re-check them before each major feed change.

## Feed, profile and agent: what a product feed does and what has to be approved first

A product feed shares your prices and availability with ChatGPT from a file you control. It is separate from your business profile and separate from a business agent. OpenAI's help page says retailers can connect a product feed, including a Shopify feed, to share product prices and availability.

Three things have to be true before the feed matters:

- You have claimed the business and OpenAI has approved the claim. You can claim a business if you own it or are authorised to represent it.
- Persona identity verification is approved. If you cannot create an agent or add products, OpenAI's guidance is to check that you are on the claimed profile and that Persona verification was approved.
- Your product URLs sit on the domain tied to the profile. If an import fails on URLs, compare the profile's website domain with the domain in the url column.

Set expectations early. A clean import means your catalog data was accepted. It does not mean ChatGPT will show your products, rank them or send you revenue, and nothing in the setup flow promises any of that.

## Choose how to deliver the feed: CSV upload, hosted file URL, Shopify connection or SFTP

Business Hub offers four ways to get a feed in, and the right one depends on how often your catalog changes. Setup asks for a feed name, at least one country, an optional currency and a delivery method.

| Method | Best for | Watch out for |
|---|---|---|
| Manual CSV upload | First test, small catalogs, one-off fixes | Goes stale the day prices change; you must re-upload |
| Hosted file URL | Catalogs that change; Business Hub fetches the file again on the schedule you choose | The URL must serve the actual feed file, not an HTML page or a login redirect |
| Shopify connection | Shopify stores; shares product details, pricing and availability from the Shopify catalog | Controlled by your Shopify catalog data, so fix problems at the source |
| SFTP | Merchants who already push files to a server | Place files in the root directory; nested folders can trigger a layout error |

On the file type question, XML or CSV: the current OpenAI documentation specifies CSV uploads, and it does not list Parquet, gzipped JSONL, or TSV as accepted feed formats. Treat CSV as the safe baseline. Save it as UTF-8 and quote any field that contains commas or line breaks. The documentation does not state a feed size limit or how long upload history is kept, so test with a small sample before sending a full catalog.

One caveat on SFTP: a successful transfer only proves the file arrived. It says nothing about whether the rows processed. Always check upload history and the Issues tab afterwards.

If you run a hosted URL, pick it for the long term. A catalog that updates on a schedule beats a CSV someone has to remember to upload.

### Can the feed be fetched on a schedule or pushed over SFTP?

Yes. The documented delivery methods are manual upload, a hosted HTTPS URL, and SFTP, while Shopify is a product-feed connection rather than a fourth delivery method. A one-off CSV upload suits a first test. A hosted file URL suits a feed your store already generates, because the file at that address can be refreshed without re-uploading. A Shopify connection avoids exporting files at all. SFTP suits larger catalogues or systems that push files on their own timetable. Pick the method that matches how your catalogue already changes, so the feed stays current without manual work.

### Should I use CSV or XML for the feed file?

Use CSV. This importer takes CSV, so if your platform exports XML, convert it to a flat CSV with one row per variant before uploading. Choose CSV for any new feed because it is easy to inspect in a spreadsheet, easy to diff between exports and easy to validate row by row. Whichever format your source system uses, the rules that matter are the same: UTF-8 encoding, exact header names, unique IDs and clean numeric values.

## Required fields: the exact CSV columns, types and value formats

Every row needs all of the required columns. The current OpenAI-format feed lists eleven: the nine basic fields plus `target_countries` and `store_country`. Some earlier walk-throughs describe nine, but the current schema also marks the two country fields as required, so include them.

| Column | What it holds | Format notes |
|---|---|---|
| item_id | Unique ID for the purchasable item | Stable, one per variant, never reused |
| title | Product name | Plain text |
| description | Product description | Plain text |
| url | Product page | Full URL on your profile's verified domain |
| brand | Brand name | Plain text |
| seller_name | Who sells it | Plain text |
| image_url | Main image | Full URL to an image file |
| availability | Stock state | One of `in_stock`, `out_of_stock`, `pre_order`, `backorder`, `unknown` |
| price | Price with currency | `79.99 USD` |
| target_countries | Where you sell it | Array of uppercase two-letter codes, e.g. `["US"]` |
| store_country | The seller's store country | Uppercase code, e.g. `US` |

A sample header and row for a made-up store:

`item_id,title,description,url,brand,seller_name,image_url,availability,price,target_countries,store_country`

`SKU-1001,Trail Bottle 750ml,Insulated steel bottle,https://example-store.com/trail-bottle,Example Co,Example Store,https://example-store.com/img/trail-bottle.jpg,in_stock,24.00 USD,"[""US""]",US`

Note the doubled quotes around the array inside a CSV cell. That is standard CSV escaping, and getting it wrong is a common cause of broken `target_countries` values. The standard OpenAI-format upload currently targets the US, so these country values do not change the default market.

The documentation also lists optional fields. Add them only after the required set imports cleanly.

### What format does a product feed need to be in?

A product feed is a flat file with one row per sellable item (each variant counts as its own row) and one header row naming the columns. Use UTF-8 encoding, a consistent delimiter, quoted text where values contain commas, and plain values with no HTML or currency symbols in numeric fields. Every row needs a stable unique ID, a title, a price, an availability value and a product URL. Keep column names exactly as the specification below lists them, because the importer matches on header text.

### Which required fields cause validation errors when they are missing or wrong?

A missing or malformed required field rejects the whole row, not just the field. The usual causes are an empty ID, a duplicate ID, a price that includes a currency symbol or thousands separator, an availability value outside the allowed list, and a product URL without https. Fix these in the source file, not in the importer. Run the checklist below before each upload so you see every failing row at once instead of discovering them one re-upload at a time.

## Variants, availability and prices: the rules that reject the most rows

Three rules account for most row-level rejections: price format, availability values and variant identity.

**Price.** Write a number in major currency units, a space, then an uppercase ISO 4217 currency code: `79.99 USD`. Use a decimal point, no currency symbol, no thousands separator, and no more fractional digits than the currency allows. `$79.99`, `79,99 EUR` and `79.99usd` all fail.

**Availability.** The value must be exactly one of `in_stock`, `out_of_stock`, `pre_order`, `backorder` or `unknown`. An empty or unrecognised value rejects the row. Use `pre_order` for items offered before release and `backorder` for items awaiting stock, and update the feed as stock changes. A feed that says `in_stock` for something you cannot ship is worse than a rejected row.

**Variants.** Each purchasable variant gets its own distinct, stable `item_id`. Variants of one product share a stable `group_id`, set `listing_has_variations=true`, and carry a `variant_dict` of the selected options, such as `{"color":"Black","size":"10"}`. The `group_id` must differ from every variant's `item_id`. Write one row per variant. Never write one row for a product with the sizes crammed into the title.

Two habits prevent most of this. Generate IDs from your catalog's SKU, not from row position, and run price and availability through a fixed formatter rather than typing them. A draft schema with different field names also circulates online. For file upload, work from the current OpenAI product-feed specification, not a draft.

## Why the importer is rejecting your feed: errors and fixes

When the importer rejects a feed, the cause is almost always in the first rows of the preview, so read the error list before changing anything. The table maps what you are likely to see to the fix.

| Symptom | Likely cause | Fix |
|---|---|---|
| Missing required field on every row | Header names do not match the schema, or a required column was dropped | Rename headers to the exact column names; confirm all eleven are present |
| Price rejected | Currency symbol, comma decimal, lowercase or missing currency code | Rewrite as `amount CURRENCY`, e.g. `79.99 USD` |
| Availability rejected | Free text such as "In stock" or an empty cell | Map to one of the five allowed values |
| Variant rows collide | Duplicate `item_id`, or `group_id` equal to an `item_id` | Make each `item_id` unique; give the group its own ID |
| `target_countries` unreadable | Array not quoted or escaped in the CSV | Quote the cell and double the inner quotes |
| URL or domain error | Product URL domain differs from the profile's website domain | Match the domains, or fix the verified domain on the profile |
| Hosted URL fails to import | URL returns a page, not the file, or needs a login | Open it in a private window; it must download the raw file |
| SFTP layout error | File placed in a nested folder | Move it to the root directory |
| Invalid format or encoding | Not UTF-8, stray quotes, wrong delimiter | Re-save as UTF-8 and re-export with proper quoting |

If the failures cluster around the business rather than the rows, check that the claim is approved and Persona verification is complete. Then fix the source file and upload the corrected feed again. For a large catalog, send a small sample first and a full snapshot only once the sample passes.

### What do I do when the importer says the file format is invalid?

Check the file itself before the data. Re-save it as UTF-8 CSV, confirm the first row is a header with the exact column names, and make sure every row has the same number of columns. Stray line breaks inside unquoted descriptions, a spreadsheet that exported with semicolons, or a byte order mark are the most common culprits. Open the raw file in a text editor to see what the importer sees. If the structure is clean, move on to row-level errors such as prices and availability values.

## A validation checklist to run before every upload

Run these checks in order before every upload, because each one catches a class of rejection cheaply.

1. **Headers.** All eleven required column names are present, spelled exactly, with no extra whitespace.
2. **Row completeness.** No required cell is empty on any row. Filter each column for blanks.
3. **IDs.** `item_id` is unique across the file. For variants, each has its own ID and every group's `group_id` differs from its members' IDs.
4. **Prices.** Every price matches `number space CODE`. A single regular-expression filter finds the exceptions.
5. **Availability.** The column contains only the five allowed values. Pivot the column to see distinct entries.
6. **URLs.** `url` and `image_url` are full links, the product URLs sit on your profile's domain, and a sample opens correctly.
7. **Countries.** `target_countries` is a quoted array of uppercase codes and `store_country` is an uppercase code.
8. **Encoding.** UTF-8, with commas and line breaks inside fields quoted.
9. **Sample first.** Upload 20 rows, read the preview, then send the full file.

After you confirm an upload, Business Hub says processing may take a few minutes to a few hours. The Products area has Feeds, Upload history and Issues tabs. Check Issues rather than assuming silence means success.

## What to do after the feed imports: keep it current and measure it separately

An imported feed is a snapshot of your catalog, and it decays. Hosted URLs are re-fetched on the schedule you chose, so keep the file at that URL accurate. With manual upload, put a re-upload on someone's calendar. Update availability when stock changes, and keep the prices in the feed consistent with your site and your delivery and return policies.

Measure it separately from everything else. Business Hub's own data, any prompts you test yourself, and your website analytics are three different populations. Do not join them at user level or infer a click-through rate from mentions. A feed import is a catalog change, not a proven cause of traffic or sales, so judge it on what your own site sessions and orders show.

This is where Revnu fits. Revnu is an agent that runs growth experiments for your business and reports back in Slack. You can hand it the next measured step after the feed is live, such as the page, outreach or search work you want to test, and it does that work within approvals you set. It does not connect to Business Hub or convert or manage your feed, so keep the feed itself in the file or Shopify connection you set up above. For related reading, see [how paid ads automation splits between you and an agent](https://revnu.com/blog/paid-ads-automation-for-startups-what-to-delegate).

## Ship a 20-row sample today, then pick the next growth test

A clean Business Hub feed comes down to the right delivery method, all eleven required columns, `amount CURRENCY` prices, five allowed availability values and unique variant IDs. Fix those and most rejections disappear. Start with a 20-row sample, confirm it passes the preview, then send the full catalog and keep it fresh. Once the feed is live, put Revnu on the next experiment worth running: tell it your goal at revnu.com and approve the first test it proposes.

## More ChatGPT Business Hub guides

- [ChatGPT Business Hub: What It Is and Who Can Use It](https://revnu.com/blog/chatgpt-business-hub-what-it-is-and-who-can-use-it)
- [How to Claim and Verify Your Business in ChatGPT](https://revnu.com/blog/how-to-claim-and-verify-your-business-in-chatgpt)
- [ChatGPT Business Hub Metrics and Weekly Reporting](https://revnu.com/blog/chatgpt-business-hub-metrics-and-weekly-reporting)
- [Turn ChatGPT Business Hub Data Into Revenue Decisions](https://revnu.com/blog/turn-chatgpt-business-hub-data-into-revenue-decisions)
- [ChatGPT Business Agents: Eligibility and Setup](https://revnu.com/blog/chatgpt-business-agents-eligibility-and-setup)

## Related Business Hub guides

- [ChatGPT Product Feed Fields Explained One by One](https://revnu.com/blog/chatgpt-product-feed-fields-explained-one-by-one)
- [Hosted URL and SFTP Product Feeds for ChatGPT](https://revnu.com/blog/hosted-url-and-sftp-product-feeds-for-chatgpt)
- [WooCommerce and BigCommerce Product Feeds for ChatGPT](https://revnu.com/blog/woocommerce-and-bigcommerce-product-feeds-for-chatgpt)
- [ChatGPT Business Hub for Shopify Stores: Setup Guide](https://revnu.com/blog/chatgpt-business-hub-for-shopify-stores-setup-guide)

## Frequently asked questions

### What file format does a ChatGPT Business Hub product feed need?

CSV is the safe baseline: UTF-8, one row per purchasable item, with every required column on each row. You can upload the file manually, host it at a URL, connect Shopify or use SFTP.

### Why is my product feed being rejected by the importer?

Common causes include formatting errors, missing required fields, or invalid values in the data. Read the preview and Issues tab, fix the source file and re-upload.

### Should I deliver the feed by CSV upload, URL or SFTP?

Use a one-off CSV upload for your first test and small fixes. Use a hosted file URL when your catalog changes, since Business Hub re-fetches it on your schedule. Shopify stores can connect the store directly. SFTP suits merchants already pushing files; put files in the root directory.

### Does a successful import mean ChatGPT will recommend my products?

No. OpenAI states that connecting a feed does not publish your business agent or guarantee product recommendations. An imported feed means your prices and availability were accepted. Judge results by your own website sessions and orders, measured separately from anything inside Business Hub.

### Can Revnu connect to Business Hub or manage my feed?

No. Revnu does not connect to Business Hub or convert and maintain feeds. It is an agent that runs growth experiments for your business and reports in Slack. Use it for the next approved growth test after your feed is live, while the feed stays in your file or Shopify connection.
