Claim Your Business on ChatGPT: Business Hub Setup (2026)

How to claim, verify and run your business profile inside ChatGPT, what the new insights really measure, and where Business Agents and sponsored placements fit.

On November 10, 2026, OpenAI opened Businesses in ChatGPT in beta, a way for a company to claim its profile inside an assistant that 1.2 billion people use every week - OpenAI. Until that day, ChatGPT described your business from whatever it could find about you on the open web, and you had no way to correct it, no way to add your current prices, and no idea how often it mentioned you. Now there is a dashboard called Business Hub where a verified owner can edit the profile, upload a product feed, and see three numbers that marketers have wanted from AI assistants since they started answering shopping questions: how often ChatGPT mentions you, how often people open your profile, and how often they click through to your website. For now, OpenAI charges nothing for any of it - Digiday.

There are two catches, and this guide spends real time on both. The first is that claiming is not ranking. OpenAI states plainly that claiming does not guarantee higher placement or sales, and that ChatGPT keeps making independent recommendations and comparisons - OpenAI. The second is that the beta is uneven: the conversational Business Agents that come with it are limited to eligible ecommerce and online service businesses in the United States, verification runs through two separate gates, and early users have hit waitlists, blank dashboards and verification errors.

This guide starts with what launched and why it matters structurally, then walks through every step of the claim, the product feed specification field by field, the insights dashboard and its blind spots, Business Agents and the policy that governs them, the link to ChatGPT Ads, how the same move looks on Google, Apple, Microsoft, Meta and Amazon, and what can go wrong. It ends with a 15-day plan, because Black Friday falls on November 27 and AI shopping assistants are a live channel this season.

This guide reflects Businesses in ChatGPT as of November 12, 2026, two days into a beta. Screens, eligibility and policies are changing week to week, so check the linked OpenAI pages before you act on any single detail.

Contents

1. What OpenAI launched on November 10
2. Why claiming matters: how ChatGPT knew your business before, and what changes
3. Who can claim today, and who has to wait
4. Before you start: what to have ready
5. Step by step: claiming and verifying your business
6. After approval: a profile ChatGPT can describe accurately
7. Product feeds: the spec, the fields and keeping them fresh
8. Reading the insights dashboard without fooling yourself
9. Business Agents: building, testing and staying inside the rules
10. Sponsoring your agent with ChatGPT Ads
11. The wider map: claiming your business across every AI engine
12. Where it breaks: hijacked claims, thin data and other failure modes
13. A 15-day plan before Black Friday
14. What comes next, and how to decide

1. What OpenAI launched on November 10

Businesses in ChatGPT is three products launched as one. The first is the claimed profile: through Business Hub, a company can claim its organization in ChatGPT and, once verified, update key details and add self-service product feeds with pricing and availability - OpenAI. A claimed profile carries a blue check so people can see it is managed by the business itself, and claiming is initially available to eligible businesses that have a website - OpenAI.

The second is insights. Business Hub shows aggregated numbers about how people engage with a business in ChatGPT, including the number of mentions and website clicks it receives, plus "discovery topics", the recurring themes of the conversations where its website is cited. OpenAI says it does not share users' conversations or Business Agent chats with businesses, and that insights only appear once enough conversations or interactions have happened to meet privacy thresholds - OpenAI.

The third is the Business Agent, a conversational assistant that represents the business inside ChatGPT. When a business appears in an answer, people can click through to its profile and, if an agent exists, start a separate, clearly labeled conversation with it to ask follow-up questions about its offerings. Businesses can also sponsor that agent and make it the destination of their ChatGPT Ads. Agents are available at launch to eligible ecommerce and online service businesses in the United States, with more markets and categories promised over time.

OpenAI published a 49-second film, "Your business, in ChatGPT", as the opening of its announcement. It is the quickest way to see the new surfaces in motion before reading about each one.

The screenshot below is the part a customer sees. A question about toy stores returns three shops in prose, and clicking the first opens its profile in a side panel: the blue check, a category, a Start chat button for the Business Agent, a View website button, delivery and returns terms, an "About" description, and product cards with prices pulled from the feed.

Notice what the panel does not contain: reviews, a star rating or a ranking position. Everything in it is either what the business told ChatGPT (its description, policies and products) or a way to reach the business (the agent and the website). That design choice says a lot about what OpenAI thinks a claimed profile is for, and it leads straight to the question of why it matters at all.

The launch did not come out of nowhere. The ecommerce analyst Juozas Kaziukėnas spotted the analytics feature inside ChatGPT as early as August and described it as part of a larger Business Hub available only to businesses that had verified ownership, for example by connecting a Shopify store - Search Engine Roundtable. And the conversational part had a dress rehearsal: in September, OpenAI began testing Sponsored Agents with select US advertisers, letting people who tapped an ad start a labeled conversation with a business-sponsored agent - OpenAI. Businesses in ChatGPT takes that agent out of the ad unit and attaches it to the business itself.

2. Why claiming matters: how ChatGPT knew your business before, and what changes

To judge what a claimed profile is worth, start with a structural question: where does ChatGPT's picture of a business come from? A language model answering "which bakery near me does gluten-free birthday cakes" or "what is a good invoicing tool for a two-person agency" cannot know the answer from training alone, because prices, hours and product ranges change daily. So it reaches outward. OpenAI's search crawler, OAI-SearchBot, is the robot that surfaces websites in ChatGPT's search answers, and sites that block it are not shown in those answers except as navigational links - OpenAI. On top of the crawl sit partners. OpenAI's help page on ChatGPT search says it sometimes partners with other search providers, names Microsoft and Shopify among them, and notes that restaurant availability may come from OpenTable, Resy or Yelp - OpenAI Help Center. Yelp, which has signed an agreement with OpenAI, describes surfacing its reviews, ratings, photos and business details directly inside ChatGPT - Yelp. Add review sites, directories, articles and forums, and for Shopify merchants a structured product catalog that OpenAI says was already integrated before this launch, and you have the full set of inputs.

That list has a notable absence. A BrightLocal study of 200,085 searches found that while Google Business Profile matters for Google's AI Mode and AI Overviews, it is "hardly used by ChatGPT", whose relationships with Yelp and Bing mean it surfaces slightly different businesses - BrightLocal. The same study found ChatGPT names an average of 4.1 businesses per response, and that repeat searches overlap by only 20 to 33 percent. A business that perfected its Google profile years ago may therefore be invisible, or described from a stale Yelp page, inside ChatGPT.

The weakness of that arrangement is that the business itself was the one party with no direct line. A shop that changed its returns window, a software company that dropped its free plan, a restaurant that moved across town: each had to wait for the crawl to notice, and for other sites to update their descriptions. And none of them could see whether ChatGPT mentioned them at all, let alone what people asked when it did. It is the same split our guide to getting your site cited by ChatGPT and Claude calls the great decoupling: your information is read and used more than ever, while far less comes back to you, in clicks or in data.

A claimed profile changes three things, and it helps to separate them because they have very different value.

A verified first-party channel. Your description, policies, contact details and (for retailers) a structured feed of products, prices and stock become something ChatGPT can read directly from you instead of inferring from others. OpenAI's own wording is careful: ChatGPT "may use this information to answer customer questions" - OpenAI. "May" is not "will", but it is a channel that did not exist on November 9.

A measurement loop. Mentions, profile views, website clicks and discovery topics give a business its first official window into ChatGPT. For years the only way to estimate AI visibility was to pay a third-party tool to run thousands of sample prompts; now the platform itself reports the count.

A conversational door. For eligible businesses, the agent turns a mention into a conversation the business shapes, with its own information and guidelines.

The obvious analogy is Google Business Profile, and it is worth testing rather than accepting. In Google's local results, the profile is a direct ranking input. Google says it ranks local results by relevance, distance and prominence, that businesses with complete and accurate information are more likely to show up, and that there is no way to pay for a better local ranking - Google. A local pack is, at heart, a database query over profiles.

ChatGPT's answer is not a database query. It is a synthesis written by a model that reads many sources and then decides what to say, which is why OpenAI insists that claiming does not guarantee higher placement. The structural consequence is that a claimed profile in ChatGPT is mainly worth three things: accuracy (the model stops saying wrong things about you), conversion (the person who already found you gets a better path to your site or your agent) and measurement. Whether you get mentioned in the first place still depends mostly on your website and on what other sites say about you, the same reputation game our guide on getting your products recommended by ChatGPT describes in detail. Even OpenAI's definition of discovery topics points this way: they are themes of conversations "where their website is cited", which means your website, not your profile, is what the model is quoting.

Pressure-test the conclusion from the other side, though. If a claimed profile does not move placement, is it worth an afternoon? Three reasons say yes. Accuracy has real money attached when an assistant quotes a wrong price or a dead returns policy to someone ready to buy, and the baseline is not reassuring: SOCi's 2026 Local Visibility Index, which tracked about 350,000 locations of 2,751 multi-location brands, found business information in ChatGPT accurate 68.3 percent of the time against Google Maps data, and found only 1.2 percent of brand locations recommended by ChatGPT, against 35.9 percent appearing in Google's local 3-pack - SOCi. The measurement is free, where comparable third-party tracking costs a monthly subscription. And the profile is the only route to the agent, which is the most interesting part of the launch for any business that sells online. The honest version of the pitch is not "claim to rank", it is "claim so that when ChatGPT does mention you, it gets you right and you can see it happening".

3. Who can claim today, and who has to wait

Eligibility is the first place the beta gets uneven, so it pays to read it closely. OpenAI's announcement says any eligible business located where ChatGPT is available can claim through Business Hub, that existing OpenAI Enterprise customers and advertisers can work with their OpenAI representative instead, and that Shopify merchants can connect their store from Business Hub or the Shopify admin to streamline verification - OpenAI. The landing page adds the conditions: eligibility depends on factors including industry, location and OpenAI's verification and integrity checks, and you must own the business or be authorized to represent it - OpenAI.

In practice, the type of business matters as much as the country. A hands-on test by the reputation platform Birdeye found that local stores and restaurants can only claim a listing that already exists in the United States, while eligible online businesses may be offered the option to create a new listing - Birdeye. Outside the US, reports already describe claims landing on a waitlist; one agency owner outside the US was placed on one, and publishers have asked whether the program is meant for them at all, since the flow is built around businesses that sell things - PPC Land.

The agent column deserves a second look, because eligibility there is stricter than for the profile. The Business Agents policy says OpenAI considers both the business category and the agent's intended use before allowing publication, may decline access or defer a category until it can support it safely, and that business verification alone does not guarantee agent eligibility - OpenAI. A verified US store can therefore still be told its category waits. If your business touches health, finance, legal advice or anything where an answer could cause harm, assume your agent will take longer than your profile.

The structural reason for the caution is that a profile is a set of facts, while an agent is a speaker. A wrong opening hour embarrasses a business; a confident wrong answer about a medication, a loan or a lease can hurt a customer, and the agent speaks inside an interface people trust for neutral answers. Expect OpenAI to widen profile eligibility quickly and agent eligibility slowly.

4. Before you start: what to have ready

The claim itself takes minutes. What slows people down is the material around it, so gather this first.

A ChatGPT account the company controls. The account you claim with becomes the profile's manager, and Birdeye's walkthrough advises using a company-controlled account rather than a personal one, because later team invitations may depend on it - Birdeye. If an employee claims with a personal login and later leaves, you inherit a recovery problem. A shared role mailbox (for example an operations or marketing address on your own domain) is the safer owner.

A live website on a domain you control. Claiming is initially limited to businesses with a website, and the most robust verification route is a DNS record on that domain. Check now that you, or whoever runs your domain, can add a TXT record at the registrar or DNS host.

A business phone that can receive a call or text. Phone verification sends a code by text or call to the main business number. Use a direct line rather than a switchboard extension, because an automated menu will swallow the code.

Identity documents for the second gate. After the claim is approved, OpenAI requires a separate identity check, run by the verification company Persona, before you can publish profile changes, connect a product feed, view insights or create an agent. Third-party guides report that it may ask for a government ID, business registration details and proof that you are authorized to act for the business - Spiritt. Have the registration document and an ID for the person claiming ready.

Your facts, written once and checked. Opening hours, service area, delivery terms, returns window, the one-paragraph description you want an assistant to repeat. The profile is only as good as what you put in, and inconsistency between your profile and your website is exactly the kind of signal a model has to resolve on its own.

Your product data, for retailers. A clean export of every product with a stable ID, title, description, URL, brand, seller name, image, availability and price covers the nine required fields of OpenAI's feed format. Section 7 goes through them one by one.

One more check takes thirty seconds and is easy to forget: make sure your site does not block OpenAI's search crawler. Allowing OAI-SearchBot in robots.txt is OpenAI's own recommendation for appearing in ChatGPT's search answers - OpenAI. Blocking the training crawler (GPTBot) while allowing the search one is a legitimate choice; our robots.txt guide shows the exact lines. Early dashboards also show a "Website access" panel that reads "Access confirmed", which OpenAI has not explained but which suggests it checks whether it can reach your site - PPC Land.

5. Step by step: claiming and verifying your business

The flow below combines OpenAI's own pages with two hands-on walkthroughs published in the first 48 hours: Barry Schwartz of Search Engine Roundtable claimed his company RustyBrick on launch day, and Birdeye documented every screen of its own claim. Screens may differ slightly for your business type, because Business Hub shows only the verification options it considers available to you.

Step 1: open Business Hub. Sign in to ChatGPT with the company-controlled account and open Business Hub from the sidebar, or go straight to chatgpt.com/business-hub. OpenAI's pre-claim screen sets out what you get: a view of how often your profile is seen and clicked, the ability to add photos, products and key details so ChatGPT can answer more accurately, and a blue checkmark that tells customers the profile is managed by the actual business.

Step 2: find your listing. Search for your business by name and check the address before you select it. Continuing means you accept the Businesses in ChatGPT terms, and eligible online businesses that do not find themselves may see an option to create a new listing - Birdeye. If you run several brands or locations, note that each brand and each location is claimed separately; claiming the brand does not claim its locations, and no bulk upload for profiles is documented yet.

Step 3: confirm the managing account. Select "Get started" and confirm which ChatGPT account will manage the profile. This is the moment the warning in Section 4 matters: whoever confirms here becomes the admin who can later invite teammates.

Step 4: verify that you are the business. Business Hub offers up to four routes, depending on the business: a business phone (a code by text or automated call), a business email (a code sent to the address associated with the business), the website domain (a DNS record) or a Shopify connection, where you select Connect Shopify, review the permissions and log in to the store you are claiming - Spiritt. Schwartz was offered phone, text and DNS, and chose DNS - Search Engine Roundtable.

Step 5: wait for the review. Business Hub shows three stages (claim received, review in progress, decision reached), OpenAI emails the decision or a request for more information, and you can withdraw the claim while it is under review. Birdeye's own claim was approved in a little over two hours, though OpenAI has published no service level for reviews, so treat that as one data point rather than a promise.

Step 6: confirm and tidy up. The approved profile shows its category, hours, address and phone, with optional next steps offered for ChatGPT Ads, a ChatGPT Business plan and a Business Agent. Birdeye noticed that a box opting you into OpenAI marketing email is pre-checked at this stage; untick it if you do not want the mail.

Step 7: pass the identity check. Claiming is the first gate; the second is identity verification through Persona, which OpenAI requires before you can publish changes to your business information, connect a product feed, view insights or create an agent. According to Spiritt's reading of OpenAI's help center, claiming alone does not add the verified badge, which appears once the identity check is approved, and submitting your documents to Persona is not the same as being approved.

The four verification routes suit different businesses, and choosing well saves a day:

Two practical notes complete the picture. Agencies do not need your password: the admin who claimed the business can add teammates in Business Hub and grant an agency access to manage the business or its agent, with each person on their own ChatGPT account. And patience is part of the process, because updates can take time to appear in a listing or in a ChatGPT answer even after they are approved.

6. After approval: a profile ChatGPT can describe accurately

Once both gates are behind you, Business Hub opens into three areas that early users describe as Overview, Agent builder and Products - PPC Land. The profile itself is short, and that is the point: every field is something an assistant can repeat to a customer without interpretation.

Look again at the side panel in Section 1. The fields that reach the customer are a category, a website, delivery terms ("Free delivery over $150"), a returns window ("Returns within 45 days"), an "About" paragraph and product cards. Each of those answers a question a shopper would otherwise ask the model and the model would otherwise have to infer from your site or from a review that may be two years old. So the job here is not copywriting in the marketing sense. It is writing the version of your business you would want read aloud by a neutral clerk.

Three principles follow from how answer engines resolve conflicting information.

Write facts, not adjectives. "Wooden toys and art kits for ages 3 to 8, every item under $30, ships in 2 business days" gives a model something to match against a question like "creative gifts under $30 for a five-year-old". "The most magical toy experience in California" gives it nothing, and the Business Agents policy already warns that concrete claims should not mislead people, a standard worth applying to the profile too - OpenAI.

Match your website word for word on anything numeric. If your profile says 45-day returns and your returns page says 30, you have handed the model a contradiction to resolve, and it may resolve it against you or repeat both. Update the website first, then the profile, and keep a short internal list of the facts that must agree everywhere: hours, prices, shipping thresholds, returns window, service area, contact channels.

Name your category the way customers search. The profile's category and description are what make you a candidate for the question in the first place. A bakery that only writes "artisan patisserie" may miss "gluten-free birthday cake"; a software company that only writes "workflow orchestration platform" may miss "tool to automate invoices". Use the customer's words, not the industry's.

The profile also gives you a reason to audit what the rest of the web says. Search ChatGPT for your own category and city (or category and use case, for an online business), read how it describes you and your competitors, and note every wrong or stale fact. The fix for most of them is upstream: your own pages, your listings on the directories and review sites that dominate your category, and the articles that already rank. Our guide to being recommended by ChatGPT covers that off-site work in depth; the claimed profile is where you state the truth, and the rest of the web is where the model checks it.

7. Product feeds: the spec, the fields and keeping them fresh

For retailers, the product feed is the most consequential part of the claim. A feed is a structured file listing every product you sell, one row per item or variant, which ChatGPT can read directly instead of parsing your product pages. Business Hub accepts either a CSV upload that follows OpenAI's product feed specification or a link to a hosted feed file, and connecting one does not publish an agent or guarantee that ChatGPT recommends your products - Spiritt.

The quiet news here is who gets to do this. OpenAI's developer documentation for agentic commerce still says onboarding product feeds in ChatGPT is available to approved partners who apply through a form, and recommends uploading a full snapshot at least once a day - OpenAI. Business Hub turns that partner program into a self-service upload for any verified retailer. Structured access to ChatGPT's shopping results used to be a negotiated integration; as of this week it is a form field.

OpenAI's specification asks for nine required fields per row, plus many optional ones - OpenAI:

The specification's minimal example is a single JSON line, which is the easiest way to see what one valid row looks like:

{"item_id":"MUG-350-BLUE","title":"Blue ceramic mug, 350 mL","description":"Dishwasher-safe glazed ceramic mug with a handle.","url":"https://example.com/products/mug-blue","brand":"Northline","seller_name":"Northline Home","image_url":"https://example.com/images/mug-blue.jpg","price":"18.00 USD","availability":"in_stock"}

The details that decide whether a feed is accepted are mostly about formats. Availability must be one of in_stock, out_of_stock, pre_order, backorder or unknown, and unknown must be written explicitly rather than left blank. Prices are written as an amount in major units, a space and an uppercase three-letter currency code, so 79.99 USD means seventy-nine dollars and ninety-nine cents, with no thousands separators. Optional fields you do not know are simply omitted, and the spec explicitly forbids placeholder strings such as null, unknown or n/a where they are not listed values. Product and image URLs must be publicly accessible, identifiers stay strings so that leading zeros survive, and in a CSV any cell containing commas, quotes or line breaks must be quoted.

Beyond the required nine, a handful of optional fields do most of the work of helping ChatGPT answer real shopping questions. Variants get their own rows that share a group_id, set listing_has_variations to true and carry a variant_dict such as {"color":"Black","size":"10"}. A gtin must be exactly 8, 12, 13 or 14 digits with a valid check digit. A sale_price only counts if it is above zero and strictly below the regular price. A blank shipping_price means unknown, not free shipping, so write 0.00 USD if shipping really is free. Returns are described by accepts_returns, return_deadline_in_days and a return_policy link, and product reviews by review_count and a star_rating on a zero-to-five scale with two decimals. Those last fields are exactly what a shopper asks about ("can I return it?", "is it any good?"), which is why they are worth the effort even though they are optional.

If you already run a Google Merchant Center feed, most of the work is done. Google's required attributes (id, title, description, link, image_link, availability, price) map almost one to one, and Google writes prices the same way, as in its example 15.00 USD - Google. The differences are small but fatal if missed: OpenAI calls the link url and the image image_url, spells pre-order as pre_order where Google uses preorder, and requires brand and seller_name on every row.

Freshness is where most feeds fail, and the reason is structural rather than technical. A feed is a copy of your catalog, and every copy starts going stale the moment it is made. A one-off CSV upload is fine for a ten-product shop whose prices change twice a year; for anyone running promotions it guarantees that ChatGPT will quote last week's price this week, which is precisely the inaccuracy the claim was meant to remove. The robust pattern is a hosted feed URL generated automatically from the system that holds the truth, refreshed at least daily as OpenAI's commerce documentation recommends, and immediately after any price or stock change you care about.

There are several ways to get there, and the right one depends on where your catalog lives. Shopify merchants get it for free: OpenAI says eligible Shopify products are already integrated through Shopify Catalog, and its Shopify path has three steps, connecting the store, reviewing the business details while product details, pricing and availability flow in automatically through Shopify's catalog, then previewing and publishing an agent - OpenAI. Merchants on other platforms can schedule a CSV or JSON export, or use the feed management tools they already pay for to output a second feed in OpenAI's field names. And a business whose software is written and maintained by an AI coder (whether that is Claude Code, Codex, or a platform such as Founden that runs one for each company it builds) can have that coder generate the feed straight from the product database and serve it at a stable address, so the feed is regenerated rather than re-exported.

One warning for Shopify merchants before you assume the integration is painless: OpenAI's ChatGPT app in the Shopify App Store, the connector that syncs a catalog to Business Hub and runs ChatGPT ad campaigns, is free but carried a 1.5-star average across 69 reviews when we checked on November 12, with recent reviews describing verification problems such as a business name that could not be changed after setup - Shopify App Store. Check your synced business details carefully before you publish anything.

8. Reading the insights dashboard without fooling yourself

The insights tab is the part of the launch that gets the most attention, because it is the first time ChatGPT has reported to a business how often it talks about it. It is also the easiest part to misread.

OpenAI's example dashboard, for its fictional toy shop, shows the layout every claimed business gets: a Brand insights tab and an Agent insights tab, a seven-day and 28-day toggle, three tiles (mentions, profile views and website clicks), a mentions trend line, and the top five discovery topics with their share of mentions.

Mentions count how often your business appeared in ChatGPT's responses. The analyst who spotted the feature before launch described it as measuring two things, the number of times a business appeared in ChatGPT responses and the number of clicks on links to its website, each tracked daily - Search Engine Roundtable. OpenAI has not published a precise definition, so it is not yet clear whether two appearances in one conversation count once or twice.

Website clicks are recorded clicks on links to your site, and the tooltip says it outright: clicks do not confirm website visits or conversions - Search Engine Roundtable. Some clicks never load your page; some visitors bounce. Treat the number as an upper bound on visits, not a count of them.

Discovery topics are the most useful tile and the least defined. The announcement calls them common themes across conversations where your website is cited, while the landing page describes them as the themes where your business appears, two definitions that can produce different lists - PPC Land. Either way, they tell you which questions you are winning, which is the raw material for your content plan.

The first real dashboards posted on launch day put the demo in perspective. BuildUp Bookings reported 690 mentions and 24 clicks over 28 days, and a Spanish-language dashboard from a business called Zianvo showed 28 mentions and one click; both read "Data through Nov 9 (UTC)", a day before launch, which suggests the counts start before you claim and run about a day behind. OpenAI's demo implies that roughly 35 clicks follow every 100 mentions; the real dashboards show about three and a half.

Two dashboards are not a benchmark, and the ratio will likely differ by category, since someone asking where to buy a specific product has more reason to click than someone asking for a comparison of approaches. But the gap is a useful corrective. A mention is the assistant saying your name inside an answer the user may never act on; the click is the decision. Plan around single-digit click rates until your own data says otherwise.

Thresholds are the other trap. Insights only appear once enough conversations or interactions occur, and OpenAI does not say how many. Schwartz's RustyBrick dashboard showed dashes and blank charts at first - Search Engine Roundtable. A blank tile means "below the reporting threshold", not "zero", so a small business may see nothing for weeks even while ChatGPT is mentioning it.

The disciplined way to use these numbers is to triangulate them with data you already own. OpenAI says ChatGPT automatically adds the parameter utm_source=chatgpt.com to referral URLs from its search results - OpenAI Help Center, so filter your analytics on that source, compare its sessions with Business Hub's click count over the same window, and read the gap as clicks that never became visits. OpenAI's wording covers its search results, so treat the tagged sessions as a floor rather than a complete count. Then compare those sessions with orders, sign-ups or calls. The dashboard tells you where the funnel starts; only your own analytics can tell you where it ends.

What the dashboard does not show is as important as what it does. It shows no prompts (only themes), no competitor comparison, no position within an answer and no data from any other AI engine, which is why the analyst Kevin Indig called it "a sort of Search Console for ChatGPT" while another practitioner called it closer to a brand listing dashboard - PPC Land. Section 11 comes back to what that means for the paid AI visibility tools.

9. Business Agents: building, testing and staying inside the rules

A Business Agent is the business's own assistant inside ChatGPT. OpenAI describes it as an agent that answers questions with your product and policy information, using guidelines you control, so customers can explore their options and take the next step with your business - OpenAI. A customer reaches it from your profile's Start chat button, and the conversation that follows is clearly labeled, separate from the customer's original ChatGPT conversation and distinct from ChatGPT's independent answers - OpenAI.

OpenAI's example shows the intended use. A shopper asks the toy shop's agent for a gift for a six-year-old who loves drawing, under $30, and the agent recommends two products from the shop's own catalog, with prices and photos, in the shop's voice.

It helps to be precise about what this is and is not. It is a salesperson who knows your catalog and policies and works inside ChatGPT's interface. It is not a checkout: for now, buyers are sent to the brand's own website, while OpenAI works on agent payments with Visa - The Next Web. It is not a support desk with access to orders, and it is not a channel that feeds you transcripts, because OpenAI does not share Business Agent chats with businesses.

Building and testing one

The build happens in Business Hub's agent builder: after claiming, eligible organizations can build, test and launch an agent. OpenAI's Shopify path describes the last step as testing your Business Agent and publishing it when you are ready, using guidelines you control and information you set. In practice that means three kinds of input, and the quality of each decides the quality of the agent.

Information. The catalog (from the feed or Shopify), the profile, and the policies a customer asks about: shipping, returns, warranties, sizing, compatibility, plan limits, cancellation terms. Anything not written down is something the agent will either decline to answer or improvise, and improvising is what the policy forbids.

Guidelines. The rules of the conversation: which products it should recommend for which needs, how it should handle a question outside your range (point to a competitor category? say you do not carry it?), when it should send the customer to a human, which topics it must not touch. Write them as you would brief a new hire on the shop floor, with your actual edge cases.

Tests. Because you will never see real conversations, the preview is the only quality control you get before customers do. Run it through the questions your customer service inbox actually receives, the comparison questions your best salespeople answer, and a few hostile ones: a request for a discount you do not offer, a medical claim about a product, a question about a competitor, a customer who says no. If the agent invents a promotion in the preview, it will invent one in production.

Our guide to building a support agent for your own site covers the same discipline for an assistant you host yourself. The difference here is that the business controls less: the model, the interface and the logs belong to OpenAI.

The rules an agent has to follow

The Business Agents policy is short and worth reading in full, because it applies when you create an agent and every time you update it - OpenAI. Its practical content sorts into what an agent may do and what it must never be set up to do:

The policy also reaches into safety: an agent must never be instructed to ignore or downplay a user's immediate safety concern, including a possible medical emergency, and it must not automate high-stakes decisions without human review. Enforcement is graduated. If OpenAI identifies a concern it may suspend or remove the agent, limit features, or restrict the business's ability to create or manage agents, particularly when problems are serious or repeated, and updates are expected to address review decisions rather than work around them.

Read that table as a description of a very particular kind of salesperson: one who cannot close, cannot collect data freely and cannot push, but who can be present at the exact moment a customer is comparing options. That is a different job from the conversion-optimized chat widgets many stores run on their own sites, and it rewards different writing. Clarity about fit ("this kit is for ages 5 and up; for a three-year-old, our wooden blocks are a better choice") is worth more than persuasion, because persuasion is what the rules restrict and clarity is what the customer came for.

Who should build one now, and who should wait

The case is strongest for businesses where buyers ask comparison questions before they buy: stores with many similar products, sizes or variants; software and online services with several plans; anything where "which one is right for me" is the question that loses sales today. For those businesses the agent answers the exact question the profile cannot.

The case is weaker for a shop with a handful of products, a business whose policies are not written down anywhere, or one in a sensitive category, where the policy's restrictions on advice and personal data will shape the agent more than your guidelines will. Waiting is also reasonable if you cannot commit to updating the agent's information when prices or policies change: a stale agent is worse than no agent, because it speaks with your name and a blue check.

10. Sponsoring your agent with ChatGPT Ads

The paid side of the launch is one sentence in OpenAI's announcement: businesses can sponsor their Business Agent and use it as the destination for ChatGPT Ads, with sponsored placements clearly labeled and separate from ChatGPT responses. OpenAI's head of monetization, Benji Shomair, put the relationship between the two plainly: "The sponsored agent is the paid experience, but the agent that sits behind it should be the same." He also told Digiday that Business Hub and agents are free for now, without ruling out a paid model later, and that paying should not affect how frequently a business appears in ChatGPT's answers - Digiday.

So the organic agent and the sponsored agent are the same assistant reached through two doors. The organic door is your profile, which opens only when ChatGPT decides to mention you. The paid door is an ad, which opens when you win an auction. That makes the agent the landing page of a ChatGPT campaign, and it is why Section 9's advice about testing matters twice: an untested agent wastes organic visitors and paid ones.

How ChatGPT Ads work in November 2026

Ads have run in ChatGPT since testing began in the United States on February 9, 2026. They appear for users on the Free and Go plans, never for Plus, Pro, Business, Enterprise or Edu accounts, never for accounts identified as belonging to people under 18, and never near sensitive topics such as personal health, mental health or politics - OpenAI Help Center. The standard unit sits below a ChatGPT response and carries the advertiser's name and logo, a headline, a description, a landing page and an image - OpenAI Help Center. OpenAI's stated principle is answer independence: ads do not influence the answers ChatGPT gives, and they are always separate and clearly labeled - OpenAI.

The platform has grown fast. ChatGPT Ads reached a $1 billion annualized revenue run rate by the end of August - OpenAI, and in October OpenAI expanded it to 89 countries, saying more than 100,000 advertisers had run campaigns since launch - OpenAI.

The buying mechanics are familiar to anyone who has run search ads. Advertisers choose views (CPM), clicks (CPC) or conversions, and conversion campaigns are billed per valid click or per thousand impressions, never per conversion; the minimum daily budget is $25 in the United States - OpenAI Help Center. Ads are chosen in a relevance-weighted, second-price auction, and OpenAI suggests starting CPC campaigns with a maximum bid of $3 to $5 per click. Prices have come down since launch: Digiday reported in April that CPMs had fallen from $60 to as low as $25 within nine weeks, and that the minimum spend for early managed campaigns had dropped from $250,000 to $50,000 - Digiday.

What OpenAI has not published is how a sponsored agent is priced, or exactly what is billed: the click into the agent, the impression, or something else. The plumbing exists, though. OpenAI's Ads API specification already lists a campaign mode called business_agent alongside product_feed, with a field for the agent's ID - OpenAI, even though the written documentation does not describe it yet. Expect sponsored agents to arrive as an ordinary campaign type in Ads Manager, bought with the same CPM, CPC and conversion options as any other ad.

What the early performance data says

Before you budget, look at the one independent number on paid ChatGPT traffic. A working paper by Maximilian Kaiser and Christian Schulze, based on data from ecommerce websites, found that paid ChatGPT visits generated $0.89 in revenue per session for US advertisers against $2.45 for paid Google Search visits, with conversion rates of 0.37 percent against 1.17 percent - Digiday.

That sits awkwardly next to Adobe's finding (Section 13) that AI-referred traffic to retailers now converts better than other traffic, and the two are not in conflict once you separate who is clicking. Adobe measures all visits arriving from AI tools, most of them people following the assistant's own recommendation at the end of their research. The paper measures people who clicked an ad shown beside an answer, a colder audience. The structural reading is that ChatGPT's organic recommendation is valuable precisely because it cannot be bought, and the paid units around it inherit only part of that value.

That is the strongest argument for a sponsored agent over a plain ad. An ad sends a curious user to a website and hopes; a sponsored agent lets that user ask the questions that stood between curiosity and purchase, inside the interface they were already using. Whether that closes the gap with search ads is not yet measured anywhere public.

The practical rules for a first test follow from all of this. Only sponsor an agent you have tested against real customer questions. Set up conversion tracking first, because OpenAI's own documentation offers a measurement pixel and a server-side Conversions API, and the Business Hub click count is not a visit count - OpenAI. Start on CPC with a capped daily budget, run it long enough to cover a full weekly cycle, and judge it on revenue per paid session against your search campaigns, not on mentions or clicks. And keep the organic profile and feed accurate regardless, because the paid door and the organic door lead to the same agent.

11. The wider map: claiming your business across every AI engine

ChatGPT is not the first assistant to invite businesses in, and it helps to see the whole map, because a customer who asks ChatGPT on Monday may ask Gemini, Siri or Copilot on Tuesday. One naming warning first: three companies now sell something called a Business Agent. Google launched one in January, a branded agent that eligible US retailers can activate and customize in Merchant Center - Google. Meta launched its own in June for WhatsApp, Messenger and Instagram, said more than one million businesses were already using it, and made getting started free with paid subscriptions to follow - Meta. OpenAI's is the third.

Google remains the deepest profile system. Creating a Business Profile is free - Google, verification methods are chosen automatically by Google from phone, text, email, live video, recorded video and postcard, and reviews take up to five business days - Google. For its AI features, Google's guidance to site owners is to keep Merchant Center and Business Profile information up to date, and it says no special machine-readable files or AI text files are needed to appear in them - Google Search Central. Google is also piloting an AI messaging agent on Business Profiles in Search and Maps, offered at no cost, though it only works for customers on networks and phones that support RCS - Search Engine Roundtable.

Apple replaced Business Connect this year with Apple Business, a free service in the US and more than 200 countries and regions from April 14 that lets businesses reach customers across Apple Maps, Mail, Wallet, Siri and more - Apple. Apple has not said whether Apple Intelligence uses that data in generated answers.

Microsoft offers Bing Places for Business free of charge - Microsoft Bing, lets owners import listings from Google Business Profile with instant verification - PPC Land, and runs Copilot Checkout, where the merchant stays the merchant of record and Shopify merchants are enrolled automatically after an opt-out window - Microsoft Advertising. Its AI Performance report in Bing Webmaster Tools shows how a site's content appears in Copilot and Bing's AI summaries, the closest thing to Business Hub's mention count on another engine - Microsoft Bing.

Perplexity made shopping free for all US users, with PayPal powering checkout and retailers staying merchant of record - Perplexity, but it has no program for a local business to claim a profile. Amazon went the other way: its Buy for Me feature lists products from outside websites inside Amazon's app, and businesses that object have to opt out by email - Modern Retail.

"-" marks what the platform has not published or what we could not verify.

The practical lesson from this table is that no single profile reaches every assistant. ChatGPT reads from Yelp, Bing and Shopify far more than from Google, and Google reads from its own profiles; an analysis of 6.8 million AI citations by Yext found Gemini favoring websites (52.1 percent of citations) while OpenAI leaned on listings (48.7 percent) - Yext. The sensible order for most businesses is to keep the Google profile and the website accurate (they feed the largest number of surfaces), claim the ChatGPT profile now, check Yelp and Bing Places because ChatGPT and Copilot lean on them, and only then reach for the rest. Our guide to getting your app into Gemini and Google Search covers the Google side in depth.

What the claim means for AI visibility tools

The launch also lands on a young industry. A wave of "generative engine optimization" tools sells estimates of how visible a brand is in ChatGPT and other assistants, usually by running fixed lists of prompts every day and counting the brands that appear. Several are well funded: Profound raised $180 million at a $1.8 billion valuation in September - SiliconANGLE, Peec AI raised a $21 million Series A last year - Peec AI, and Sitecore acquired Scrunch in June - Sitecore.

Their published prices give a sense of what Business Hub's free numbers compete with:

Prices are as published on each vendor's own pricing page on November 12, 2026, and change often.

The structural difference is the data. The tools simulate: they ask sample questions and count answers, which is an estimate of a distribution nobody outside OpenAI can see. Business Hub counts: it reports what actually happened across real conversations, but only for ChatGPT, only for your own business, and only as themes rather than prompts. Trending Topics put the threat bluntly, noting that the startups relied on simulated prompts and estimates while OpenAI has the real usage data - Trending Topics. Digiday's preview pointed to what stays hidden: prompts, user data and any ranking against competitors - Digiday.

So the honest conclusion is a split. A small business that only cares about ChatGPT and has no budget for monitoring now has a free, real count, which removes most of the reason to pay for a single-engine tracker. A brand that needs to know how it compares with competitors, what exact questions it loses, or how it does on Gemini, Perplexity and Copilot still needs a tool, and can now calibrate that tool's estimates against OpenAI's real numbers.

12. Where it breaks: hijacked claims, thin data and other failure modes

Every system that lets businesses claim a profile attracts people who claim profiles that are not theirs, and Business Hub is two days old. Schwartz said on launch day that he was "honestly nervous about people hijacking businesses" - Search Engine Roundtable, and OpenAI has not described how a business disputes a profile someone else claimed wrongly. The scale of the problem on a mature platform is sobering: Google says it removed more than 13 million fake Business Profiles in 2025 and blocked 79 million inaccurate or unverified edits - Google. The defense is simple and urgent: claim your own profile before someone else does, because the owner of the claim controls the description, the feed and the agent.

The other failure modes are quieter but more common.

Verification friction. The identity check, the DNS record, and the Shopify connection all fail in ordinary ways: a record that expires before it propagates, a phone code that lands in a switchboard, a Shopify sync that locks in the wrong business name. The low rating of OpenAI's Shopify app is the clearest early signal, and claims from outside the US are already landing on waitlists.

Thin data. Thresholds hide small numbers, counts lag by about a day, and definitions are still loose. A young or niche business may see an empty dashboard for weeks and draw the wrong conclusion that ChatGPT never mentions it.

Mistaking a claim for a ranking. AI assistants are far more selective than search results. SOCi found ChatGPT recommending only 1.2 percent of brand locations against 35.9 percent appearing in Google's 3-pack, and BrightLocal found repeat searches overlapping by only 20 to 33 percent. A claim does not change either number by itself. The levers for being mentioned (reputation on third-party sites, a website that answers questions clearly, accurate listings on Yelp and Bing) sit outside Business Hub.

An agent that speaks wrongly in your name. The Business Agents policy puts responsibility squarely on the business: businesses are responsible for the information and instructions they provide - OpenAI. With no transcripts to review, a stale price or an over-eager guideline can repeat itself for weeks before anyone notices.

Platform churn. OpenAI's commerce features have changed course before. Instant Checkout launched in September 2025, and in March 2026 OpenAI said its initial version "did not offer the level of flexibility that we aspire to provide", letting merchants use their own checkouts while it focused on product discovery - OpenAI. PPC Land reports that Walmart had found in-chat checkout converting three times lower than sending shoppers to its own site - PPC Land. Our guide to selling your product inside ChatGPT traces that history. The lesson is not to avoid Business Hub; it is to keep the work that survives any redesign (accurate data, a good website, a clean feed) separate from the features that may not.

Trust pressure from ads. OpenAI insists that paying for ads does not change how often ChatGPT recommends a business, and sponsored placements are labeled and separate from answers. But ads are now a billion-dollar run-rate business for OpenAI (Section 10), and every advertising platform eventually faces the tension between what advertisers want and what users trust. Watch the labeling as closely as you watch your numbers.

13. A 15-day plan before Black Friday

The calendar makes this launch more urgent than a typical beta. Adobe expects traffic from AI tools to US retail sites to rise 130 percent year over year this holiday season, with the biggest jumps on Thanksgiving (159 percent), Black Friday (95 percent) and Cyber Monday (88 percent) - Adobe. The traffic is also worth more than it used to be: in September, AI traffic to US retail sites converted 90 percent better than non-AI traffic and generated 83 percent more revenue per visit, a reversal from a year earlier when it converted worse - Adobe.

Keep the scale in proportion, though. Across 118,671 sites, Ahrefs measured ChatGPT at 0.36 percent of all visits in October, against 24.74 percent for Google - Ahrefs. And shoppers still want to keep their hands on the wallet: in Deloitte's holiday survey, 39 percent said generative AI is their primary tool for at least one holiday shopping task, but only 16 percent would let it complete a purchase - Deloitte. AI is a research channel that is growing fast and converting well, not yet a checkout. That is exactly the stage a claimed profile, an accurate feed and an agent are built for.

With Black Friday on November 27, here is a sequence that fits in the time left, assuming you start on November 12:

The order matters more than the dates. Accuracy comes first, because every later step multiplies whatever the profile and feed say. The agent comes second, because it is only as good as the information under it. Paid promotion comes last, because without a few weeks of organic numbers you cannot tell what the ads added.

14. What comes next, and how to decide

OpenAI calls this launch an early step and has said where it goes next: more markets and business categories for both the profiles and the agents, tools for organizations with more complex needs, and more ways for businesses to connect with customers directly or through agents, in ChatGPT and beyond - OpenAI. Read alongside Google's agent checkout in AI Mode, Microsoft's Copilot Checkout and Meta's agents on its messaging apps, the direction is clear even if the details are not. The assistant is becoming the place where comparison shopping happens, and every platform is building a seat at that table for the businesses being compared.

Reason from first principles about what that does to the economics of being found. In a list of ten links, being sixth still earned some clicks. In an answer that names about four businesses, the ones not named earn nothing, which makes the inputs to that answer far more valuable than they were. There are only three of them a business can influence: what it says about itself (now a claimed profile and a feed), what others say about it (reviews, articles, listings, the reputation work this guide keeps pointing to), and how it behaves when a customer arrives (the agent, the website, the price). Business Hub gives a business direct control over the first, a measuring stick for the result, and a voice at the third. It gives no control over the second, and the second still decides most of who gets named.

The next step is easy to predict because the pieces already exist. A shopper's own assistant will increasingly do the comparing, and a business's agent is the natural counterpart for it to question: what fits, what it costs, when it ships, whether it can be returned. When that happens, the businesses with structured, accurate, current information will be the easy ones to recommend, and the ones whose facts live in a PDF and a two-year-old review will not. Our guide to selling to AI agents looks at that machine-buyer future in detail.

For the decision in front of you this week, the framework is short:

• Claim the profile if you are eligible. It costs an afternoon, protects you from someone else claiming you, fixes how ChatGPT describes you, and gives you free, real numbers.
• Connect a feed if you sell products, but only one that refreshes automatically from the system that holds your prices. A stale feed is worse than none.
• Build an agent if you are a US online store or service whose customers ask comparison questions, and you can keep its information current. Otherwise wait and watch how the early agents behave.
• Sponsor the agent only after you have a few weeks of organic data and a conversion measure you trust, with a budget you are willing to treat as a test.
• Keep paying for a visibility tool only if you need competitors, exact prompts or other engines. For ChatGPT alone, Business Hub now gives you the real count.

The assistants are not replacing the work of being a business worth recommending. They are making it more visible, to customers and now, for the first time, to the business itself.

In this topic: Getting customers

• Block AI Crawlers in robots.txt, Keep AI Search (2026)
• Google Connected Apps Platform: Gemini + Search (2026)
• How to Create Agent Skills for Your Product (2026)
• Get Your Products Recommended by ChatGPT 2026
• Sell Your Product Inside ChatGPT: 2026 Guide
• Generate a Month of Ad Creative for $5 (2026)
• Sell to AI Agents: 2026 Setup Guide
• Get Your Site Cited by ChatGPT and Claude 2026

How do you get found and get customers? Every guide on it

Data behind this topic: Startup and AI directories: what each did with our listings.