# ChatGPT Business Profiles Meet a $1.3 Trillion Anthropic IPO

By Richard Mackoy (@richardmackoy) · Published 2026-11-13

Canonical: https://voce.com/@richardmackoy/chatgpt-business-profiles-meet-trillion-anthropic-rd8r32

---

OpenAI opened business claims inside ChatGPT on Nov. 10, letting verified businesses control what the assistant says about them. On the same day, Anthropic filed an updated prospectus to raise up to $64.4 billion, with shares expected to trade Nov. 19 under the symbol ANT. Real estate agents now face a discovery layer they do not control and cannot buy their way around.

The two events are unrelated on paper. They land on the same calendar week, and they point the same direction. For a decade, an agent's visibility was a function of a Google Business Profile, a Zillow ranking, and paid leads. Now a growing share of buyers and sellers start by asking an assistant to narrow the field. That assistant pulls from whatever data it can verify, and it does not show ten blue links. It shows a shortlist.

OpenAI's own framing puts the shift plainly: as people turn to AI to compare options and make decisions, the company is building a way for businesses to participate directly ([OpenAI](https://openai.com/index/introducing-businesses-chatgpt)). The practical question for a listing agent in a mid-size market is narrower. When a buyer asks which agent handles mountain property in their county, who does the model name?

#### Key Takeaways

-   OpenAI launched Businesses in ChatGPT in beta on Nov. 10, 2026, with a claim and verification flow open to eligible businesses wherever ChatGPT is available.
-   Verified profiles surface mentions, website clicks, and discovery topics, giving agents a first read on how AI answers represent them.
-   Business Agents are limited to eligible ecommerce and online service businesses in the US, so most brokerages cannot publish one yet.
-   Anthropic filed to raise up to $64.4 billion at a roughly $1.3 trillion valuation, with shares expected to trade Nov. 19 under the ticker ANT.
-   Claiming a profile does not change visibility on its own, and ChatGPT says it may still use other sources when answering about a business.

## OpenAI shipped a claim-and-verify layer, not a directory

Business Hub is a claim-and-verify system, not a listing directory. A business finds its profile, proves control through phone, business email, a DNS record, or Shopify, then completes identity verification with Persona before it can publish edits, see insights, or build an agent. The verified badge appears only after that second check clears.

For most agents, the eligibility line is the story. Local stores and restaurants can currently claim only an existing US listing, and business agents are restricted to eligible ecommerce and online service businesses in the US with a website ([OpenAI Help Center](https://help.openai.com/en/articles/20001635-setting-up-your-business-on-chatgpt)). An independent agent with a brokerage site sits closer to the ecommerce side of that line than a restaurant does. A team lead running five offices under one brand has a different problem: which entity owns the profile.

That is the part worth slowing down on. Agents who have spent years consolidating reviews onto one Google Business Profile per office now have a second surface where entity confusion costs them. If the profile is claimed under a franchise name, a team name, or a personal name that differs from the license record, the assistant has fewer reasons to treat them as one business.

A concrete version of this shows up on the broker side often enough to be unremarkable. Picture a five-office team operating under a regional brand: two offices list on the Google Business Profile as the brand name, two as the team name that leads the top-producing office, and one under the managing broker's personal name because that is what the license was originally filed as. All five are legitimate. The assistant has no way to know that. It sees three distinct businesses with overlapping service areas, and when a buyer asks for an agent covering a specific county, it will cite whichever entity has the clearest supporting record. The other four are not penalized. They simply are not candidates.

The insights tab gives a partial read on whether any of this is working. It reports mentions and website clicks over a rolling 28-day window, plus discovery topics drawn from conversations where the business was cited. Mentions are counted differently depending on the business type, and OpenAI states plainly that counts may be incomplete and that ChatGPT may also use other sources when answering questions about a business.

![ChatGPT business profile and insights interface](https://convex.voce.com/api/storage/6361581f-40f6-40ec-93ab-9cdaf677c2a4)

1.2 billionweekly ChatGPT users OpenAI cites as the audience for Businesses in ChatGPT[OpenAI](https://openai.com/index/introducing-businesses-chatgpt)

## AI shortlists now carry more weight than a ranking

A search results page shows options and lets the buyer sort them out. An assistant produces a shortlist, often two or three names, and the buyer works from that. The volume of traffic is not the issue yet. The concentration is.

Buyer behavior has moved faster than most agent marketing has. **The National Association of Realtors reported that roughly one in five prospective buyers has used an AI tool for homebuying research, rising to 28% of millennials and 32% of Gen Z buyers (NAR).** Use clusters early in the process: estimating affordability, learning how the transaction works, and researching neighborhoods.

That timing matters more than the headline percentage. The agent recommendation often lands in the same window as the affordability math, which means the model's shortlist can shape the call before the buyer ever opens a portal. Homebot's read on generative engine optimization for agents puts the same shift in plain terms: search engines deliver options, assistants deliver answers ([Homebot](https://homebot.ai/blog/geo-real-estate-agents-ai-search)).

Agents who have heard the GEO pitch should treat the sourcing behind it with care. Several of the bolder claims circulating this year, including conversion comparisons between AI-referred and portal leads, come from firms that sell AI visibility services. One roundup of the data separates the trade association figures from the vendor-published ones and notes the latter should be read as directional ([Rise Realty](https://www.riserealtyca.com/blog/how-many-homebuyers-now-use-ai-to-find-a-real-estate-agent)). The defensible version is narrower and still uncomfortable: a meaningful share of buyers now researches through assistants, and few agents have done anything to control how they appear there.

The Florida experiment shows where this ends if agents are skipped entirely. A startup called Homa says at least 10 homes have closed end to end through its AI system, with buyers self-representing and keeping a commission that runs around 2.5% to 3%, and a Miami broker quoted in the coverage argues the generated advice misses off-market context that decides deals ([AOL](https://www.aol.com/articles/florida-homebuyers-using-ai-homes-212914816.html)). That is a niche, not a market. It does show that the question buyers ask has changed from which agent should I call to what can I do without one.

## Anthropic's filing commits the money behind the assistant layer

Anthropic filed an updated prospectus on Nov. 10 to sell 825 million shares at $70 to $78 each, aiming to raise up to $64.4 billion with shares expected to start trading on the Nasdaq on Nov. 19 under the symbol ANT ([Reuters](https://www.reuters.com/business/anthropic-ipo-raise-much-644-billion-latest-draft-prospectus-shows-2026-11-10)). At the top of that range the company would be valued at about $1.4 trillion on a fully diluted basis, up from $965 billion in its last private round.

The numbers behind that valuation explain why the assistant layer is not going away. Revenue for the first nine months of 2026 reached $33.24 billion, up roughly 1,200% from $2.55 billion a year earlier, according to a summary of the filing ([LinkedIn News](https://www.linkedin.com/news/story/anthropic-seeking-13t-valuation-in-mega-ipo-8922905)). The same filing carries about $900.9 billion in compute and infrastructure commitments, with $70.7 billion due in the year after Sept. 30.

That is the part agents should read twice. A company taking on that much fixed capacity has to keep pushing its models into more decisions, more surfaces, and more commercial workflows. Discovery is one of those surfaces, and it is the one closest to a commission check.

The valuation also marks how far the money has moved since the start of the year. Anthropic's May 2026 Series H valued the company at $965 billion, roughly two-thirds of the figure in the current filing ([SmartAsset](https://smartasset.com/investing/anthropic-ipo)). Some investors told CNBC that even [$1.5 trillion](https://www.linkedin.com/news/story/anthropic-seeking-13t-valuation-in-mega-ipo-8922905) looked like a stretch, a signal that the market is pricing AI infrastructure with less patience than it showed in the spring.

For a brokerage, none of this changes the day-to-day. What it changes is the assumption that AI search is a side experiment run by a few startups. The models doing the recommending are now attached to public-market revenue targets, which usually means more distribution deals and more pressure to make answers commercially useful.

## The Homa experiment shows the alternative, and its limits

The Florida case is the clearest precedent for what happens when buyers skip the shortlist entirely. A startup called Homa says at least 10 homes have closed end to end through its AI system, with buyers self-representing and keeping a commission that runs around 2.5% to 3% ([AOL](https://www.aol.com/articles/florida-homebuyers-using-ai-homes-212914816.html)). Ivan Chorney, a Miami-area broker with Compass, told the New York Post that generated advice tends to reproduce formulas the portals already offer while missing the off-market context that decides deals.

That reaction is the useful part for agents weighing the claim flow. The work that resists automation is the same work that makes an entity citable: a named school boundary change, a specific building's flood history, an HOA fee range that is not published anywhere. A model assembling a shortlist needs a source to pull from, and a brokerage page that states those details in plain language is the source it will reach for. Generic neighborhood copy gives it nothing to cite, which is why it defaults to whichever profile looks most confirmed.

For brokerages weighing where to spend marketing dollars, the ordering has shifted. Paid leads still work, and portal traffic has not collapsed. But the paid lead arrives with a name attached and a cost per contact, while an assistant recommendation arrives with the buyer already partway through a decision. One is a transaction. The other is a position.

The Anthropic filing and the OpenAI launch will not be remembered as the week AI search won. They mark the point where the assistant layer stopped being free surface area and started carrying revenue targets and verification rules. Agents who treat their profile the way they treated their first Google listing a decade ago will be fine. The ones who wait for the category to settle will find that the shortlist formed without them.