OpenAI has opened a new part of ChatGPT to businesses. Online shops and services in the US can claim their company, get verified, and publish a business agent: a labelled chat, kept apart from your main conversation, that answers questions about what the company sells. It appears when ChatGPT mentions the business in an answer, and it is in beta for now.
The store writes the script and OpenAI owns the stage
A business gives the agent two things: instructions on how to answer, and its product details. It cannot bring an agent of its own. The conversation itself stays inside ChatGPT, so the business does not get transcripts. It gets themes of the conversations and counts of mentions and clicks, and when the shopper is ready to pay, the agent sends them to the business's own website.
This is not the first time a big platform has done it. Google has had a Chat button in Search since January, working like a sales associate for retailers such as Lowe's and Reebok. What is different now is the place. Many people already ask ChatGPT what to buy, so the next question, "what is the difference between these two?", can be answered by the shop itself without leaving that conversation.
The free chat and the paid ad are the same agent
Since September, OpenAI has been testing an ad that opens a branded chat instead of sending the shopper to a website, and Wayfair is one of the testers. The free business agent is the unpaid version of that ad: the agent behind it is the same. So what a shop can buy is a way into the conversation, and what it gets for free is the salesperson. OpenAI has said the free agent does not make a business show up more often in answers, and has not ruled out charging for the Business Hub later.
The verification step explains why the free door is open at all. Before Google made its shopping results paid-only, the free listings had a quality problem: they often led to items that were out of stock or were not the item shown. By 2020 Google could check a product feed against the shop's site much better, and it reopened the door. OpenAI is building the same trust first, by confirming who the business is and by using only the details the business supplies.
What a shop owner does with it
Take an online shop that sells rain jackets. A shopper asks ChatGPT for a jacket for her commute, the shop comes up, and she opens its chat and asks how two jackets differ. The agent can only answer from what the shop wrote. If one jacket is described as waterproof when it is only water-resistant, every shopper who asks gets that wrong answer in the shop's name. If sizing is missing, the agent cannot say whether the jacket runs small, and she goes elsewhere to find out.
So the useful first step is to list the questions your customer service inbox gets most (is it waterproof, does it run small, how long does delivery take, how do returns work) and make sure each answer is written down in the details the agent reads. For a shop with a few hundred products, that is an afternoon's work on the twenty or so products people ask about most.
Shoppers can use the same logic from the other side. The verification tells you the chat belongs to the real business, which settles whether you are talking to the right company. It does not settle whether the advice is neutral. A sensible way to use it is to ask your main ChatGPT conversation to compare shops and products, and then open each shop's chat for the things only that shop knows: stock, sizes, delivery and returns.
Every shop will be able to have a salesperson like this, because it costs almost nothing to set up. The difference between shops will be how carefully each one wrote down what it sells.