OpenAI launches Businesses in ChatGPT beta with DNS checks, Persona IDs and product feeds

What OpenAI shipped

On 10 November 2026 OpenAI put Businesses in ChatGPT into beta, and according to a technical walkthrough on dev.to, the launch is really two products under one name.

The first is Business Hub, the management layer. An authorized person claims a listing, edits details such as the description, logo, category and policies, connects a feed of products with prices and stock levels, and reads aggregate performance data covering mentions, website clicks and the topics that drive discovery. Insights span only the trailing 28 days and appear only above a threshold OpenAI has not specified. The company says it does not share user conversations or agent chats with the business.

The second is Business Agents, an optional layer. Eligible US ecommerce and online-service businesses with a website can build, test and publish an agent that lives on their ChatGPT profile; conversations with it are labeled as distinct from ChatGPT's own answers. In its current form the agent cannot take payments, bookings or order lookups — it explains products and policies and links back to the site.

Claiming requires neither an ads account nor ChatGPT Enterprise, and the product is separate from ChatGPT Business, the paid team workspace. OpenAI's FAQ, as cited by dev.to, is blunt on two points: claiming a business does not make ChatGPT recommend it more often, and ads do not sway ChatGPT's independent answers.

How the claim path works

There is no public form to submit to. The documented route is to sign in to ChatGPT, open Business Hub from the sidebar and search for the business by name. The dev.to author notes that unofficial guides circulate a direct Business Hub address, but it behaves like a signed-in ChatGPT page rather than an open application.

After selecting the matching listing and confirming the managing account, verification runs through whichever method OpenAI presents:

• A code by SMS or automated voice call to the business phone number on file
• A code to the email address associated with the business
• A DNS TXT record, copying the record type, host and value exactly, waiting for propagation, then triggering domain verification
• A Shopify connection, authorizing the app and signing in to the store being claimed

OpenAI gives no review timeline, only that it varies and that a decision or a request for more information arrives by email.

Persona gates everything after approval

A successful claim unlocks almost nothing on its own. Per the dev.to walkthrough, profile edits, feeds, insights and agents all stay locked until identity verification through Persona completes — a check that may require a government-issued ID, business registration details and proof the submitter may act for the company. The blue badge arrives with Persona approval, not with the claim, and listings created fresh through Persona publish automatically.

Already-claimed listings get a dispute route that also runs through Persona. Local shops and restaurants are limited to claiming existing US listings, while eligible online businesses that find nothing may see an option to create one. Brands are managed individually, so claiming a parent does not cover its locations. The dev.to author found no fee listed on OpenAI's help pages for claiming or publishing a regular agent.

Align entity data before claiming

Here the dev.to post shifts from documentation to advice, and labels it as such. OpenAI requires submitted information to be accurate while reserving the right to answer from other sources, and Persona failures are tied to names, addresses and registration numbers matching official records. The recommendation is to make the business name, address or service area, phone, website URL, hours and category identical across the site, its schema.org LocalBusiness JSON-LD, Google Business Profile and Bing Places, and to point sameAs at profiles that corroborate the entity.

Two measurement rules from OpenAI sharpen this. For online businesses, a mention counts only when ChatGPT references the website, not when it merely names the brand. For local businesses, a mention counts whenever ChatGPT surfaces the location, click or no click. The canonical URL therefore becomes a measurement decision: if ChatGPT cites a regional microsite or a booking engine, the numbers marketing tracks may never move.

Product feeds

Retailers can connect a Shopify feed or upload a CSV built to OpenAI's specification, which requires nine fields per purchasable item or variant: item_id, title, description, url, brand, seller_name, image_url, availability (in_stock, out_of_stock, pre_order, backorder or unknown) and price expressed as amount plus currency, for example 79.99 USD. Feeds must be UTF-8, use public HTTP(S) URLs and real brand and seller names, currently target the US, and item_id values must stay stable and never be reused. Connecting a feed publishes no agent and guarantees no recommendations.

Why it matters

This is merchant-side plumbing for ChatGPT as a discovery channel, and the work it creates is bigger than a form submission: a multi-stage pipeline of claim, verification and identity checks that web and SEO teams will be asked to finish on short notice. The mention-counting rules turn mundane choices — which URL is canonical, whether listing data agrees everywhere — into directly measurable outcomes. The guardrails matter just as much: no conversation sharing with businesses, no ad influence on answers. If those hold as the beta expands, Businesses in ChatGPT becomes a standard channel; if they erode, so does the trust the whole layer depends on.