Listen and follow: Spotify, Apple Podcasts, YouTube, Podcast page

About this episode

OpenAI launched Businesses in ChatGPT in beta on November 10. Through Business Hub, a company can claim and verify its profile, update products, prices, and policies, read 28 days of mentions and website clicks, and, if it is an eligible United States ecommerce or online service business, publish a clearly labeled Business Agent that answers shoppers inside ChatGPT. Google DeepMind's AlphaProof Code generated more than six million lines of Lean proof in four weeks to verify selected security properties of Signal's compiled code. Anthropic launched its IPO roadshow at seventy to seventy eight dollars a share, a raise of up to about sixty four billion dollars valuing the company near one trillion dollars, while OpenAI reportedly posted about ten billion dollars of quarterly revenue with enormous operating losses. Anthropic also launched Claude Sheets, Meetings, and Whiteboards. Google opened Gemini 4 Carbon to paying developers at two dollars per million input tokens. And Railway's breach report estimates stolen credentials produced roughly seven hundred forty thousand dollars in unauthorized AI use. New York opened RAISE Act registration for frontier developers over five hundred million dollars in revenue, with safety frameworks, seventy two hour incident reports, and penalties of one million dollars for a first violation.

Every story is framed for a local business owner: what happened, what it means for your shop, and what to do about it this week.

Key takeaways

  • Claim the new storefront early. ChatGPT Business Hub profiles can be claimed and verified now. Claiming does not guarantee placement or sales, per OpenAI itself, but it puts your real facts in place of whatever the internet had lying around.
  • Feed clean data, because an agent quotes what you feed it. Prices, availability, and policies in your profile are what a Business Agent will tell customers in your name. Keep them matched to your website.
  • Let AI draft, let an independent check decide. AlphaProof Code's pattern, AI generation plus a separate proof checker and human reviewed specifications, is the right shape for any high stakes AI output in a small business.
  • Read filings, not decks. Anthropic's public offering documents will show revenue, costs, and commitments quarterly. That is where vendor pricing and model retirement pressure shows first.
  • Your AI keys are cash. One key per job, spending limits where offered, and rotate first when a key may be exposed. Railway's estimated seven hundred forty thousand dollars in unauthorized use is the cautionary tale.

Quotes from the episode

  • "Every new place your business can be described, someone will describe it. The only question is whether the description comes from you." (Nick)
  • "Never let the same tool grade its own homework." (Nick, on AlphaProof Code)
  • "A cheap token that needs three retries is not cheap." (Nick, on Gemini 4 Carbon)

One thing to try this week

Run the claim and clean up. Set aside thirty minutes, start the claim process for your ChatGPT business profile if you are eligible, then line up your name, hours, phone number, and top five product prices across your website, your Google profile, and your AI profile. Fix any mismatch at the source and record today's mention count as your baseline.

Sources

Find your next AI tool at BotPickr, our directory of more than 170 AI tools and bots: https://botpickr.com

Full transcript

Alex: Before we start: this podcast is produced with AI. Jordan and I are AI hosts, and Nick's voice is an AI clone of his real voice. Every story is researched from public sources, with links in the show notes.

Alex: Here is the one that should make every shop owner sit up. ChatGPT, the app your customers already ask for recommendations, just opened a business directory inside the answer itself. You can claim your company profile, feed it your real prices, and, if you qualify, put an agent in front of shoppers that speaks in your name.

Jordan: That is our lead story, and it is a big one. Plus, Google used AI to write more than six million lines of mathematical proof checking the encryption behind Signal. Anthropic priced a stock offering that values the company near a trillion dollars. Claude moved into spreadsheets and meetings. Google opened its strongest model to paying developers. And stolen AI keys ran up a seven hundred forty thousand dollar bill for somebody else.

Nick: I am Nick Throlson, owner of Web Guy Nick. I spend my days on client listings, websites, and phone calls that turn into jobs. The ChatGPT story is not tech trivia. It is a new storefront, and storefronts get claimed early or they get filled in by strangers.

Alex: Story one. OpenAI launched Businesses in ChatGPT on Tuesday, in beta. Through a tool called Business Hub, a company can claim its organization, verify it, and update its website, description, logo, products, prices, availability, and delivery and return policies.

Jordan: Verification is practical. OpenAI describes a code sent to the business phone or email, a text record added to your website domain, or a Shopify store connection, with identity verification for the claimed profile check mark. If you have ever claimed a Google Business Profile, this will feel familiar.

Alex: The hub also gives you numbers. Insights cover the past twenty eight days: how often your business was mentioned, how many people viewed your profile, and how many clicked through to your website. The data is aggregated, and OpenAI says individual customer conversations are not shared with businesses.

Jordan: Then comes the part Google never gave a listing. Eligible ecommerce and online service businesses in the United States can publish a Business Agent. When your business appears in a ChatGPT answer, a shopper can open a separate, clearly labeled chat with your agent and ask it about your products and services.

Nick: Here is the owner translation. That agent will quote whatever you feed it. If your profile says a product is in stock at the old price, your agent will confidently tell a customer the wrong thing, in your name, at two in the morning. Clean data in, trustworthy answers out. Stale data in, apologies out.

Alex: One honest note from OpenAI itself. Claiming a profile does not guarantee higher placement in answers or more sales. This is a deed to a plot of land, not a promise of traffic.

Jordan: And some context. OpenAI tried to turn ChatGPT into a checkout earlier this year, with a cart and commissions, and retired that attempt in March after it failed to take off. This time the pitch is simpler: be findable, be accurate, and answer questions where the customer already is.

Nick: My practical order is three lines long. Claim the profile before someone else's facts fill it. Feed it clean products, prices, and policies, and keep them matched to your website. Then watch those twenty eight day mentions and clicks the way you learned to watch your search reports, because that is the new foot traffic report.

Alex: Story two. Google DeepMind published AlphaProof Code on Tuesday, a system that formally verifies software, and it pointed the system at Signal, the encrypted messaging app billions of people rely on.

Jordan: Here is what that means in plain English. Instead of asking an AI whether code looks safe, human experts wrote down the exact security properties that must hold, about eighteen thousand lines of specification. Then agents powered by Google's Gemini Carbon model wrote the mathematical proofs, more than six million lines in a language called Lean, in four weeks.

Alex: A separate proof checker, not the model, validated every argument. The work covered the actual compiled code behind Signal's library, including its post quantum handshake, and established properties like forward secrecy and authentication for the parts specified. Human reviewers were directed to roughly one percent of the work.

Jordan: The test that sells it: researchers put two previously fixed flaws back into isolated copies of the program, and the system proved those versions insecure. It found no new vulnerabilities in the portions examined, and DeepMind published the verification artifacts for others to replay.

Nick: The boundary matters as much as the headline. This proves selected properties, under stated assumptions. It does not certify every part of Signal, every phone, or every possible attack. But the pattern is gold for a small business. Let AI do the heavy drafting, and let an independent check, a person, a test, a reconciliation, decide whether it passes. Never let the same tool grade its own homework.

Alex: Formal verification used to be the gold standard nobody could afford, because human proof work does not scale. Agents just cut the cost of that labor, on the one target a skeptic would pick.

Nick: Break time, and I am spending it on my own site. If you have ever typed 'best AI tool for' into Google and drowned, BotPickr is the fix. More than 170 AI tools and bots, organized by the work they do: sales, support, content, scheduling, the lot. Pick the job, get a short list that actually fits, and skip the hype videos. Find your bot in 60 seconds at botpickr.com. Now back to Alex and Jordan, who never need a break, because they are robots.

Jordan: Story three, the money. Anthropic, the company behind Claude, launched the roadshow for its stock market debut on Tuesday. The proposed listing is on the Nasdaq under the ticker A N T, with an expected price range of seventy to seventy eight dollars a share.

Alex: The scale is hard to say with a straight face. About eight hundred twenty five million shares are on offer, which at the top of the range is a gross raise of about sixty four billion dollars. A separate private placement adds twenty six billion dollars from cornerstone investors including Google, Nvidia, and A M D. At the top of the range the filing values Anthropic at about one point zero two trillion dollars, or up to about one point four trillion fully diluted.

Jordan: Two cautions, and they matter. This is a proposed offering, not money already raised. The registration statement has not yet become effective, so no shares can be sold under it yet. Reports put pricing on November eighteenth and first trading on November nineteenth, but those dates are reported, not confirmed in the filing.

Alex: The rival's numbers landed in the same news cycle. OpenAI reportedly generated about ten billion dollars of revenue in a single quarter, up nearly fifty percent, while still posting enormous operating losses. One company is asking investors to buy in. The other is showing them what the race costs.

Nick: Owner lesson, and it is not about buying stock, because I am not your financial adviser. For the first time, a frontier AI lab's revenue, costs, and commitments will sit in public filings that anyone can read every quarter. When a vendor tells you a price will never change, or a model will never be retired, the filings are where the pressure shows first. Read the documents, not the decks.

Jordan: Story four. Anthropic also moved Claude into the office on Tuesday, with Claude Sheets, Claude Meetings, and Claude Whiteboards, plus a data tool for shared dashboards that is now generally available.

Alex: Sheets is in beta for paid subscribers. It supports collaborative editing, formulas, and export to Excel. Whiteboards group ideas and draft action items. Meetings starts with Google Meet for Team and Enterprise customers, and it identifies itself as a participant, posts a notice, and needs administrator permission for calls with outside guests.

Nick: That self identification is the right design, and owners should demand it everywhere. If an AI sits in your meeting or touches your spreadsheet, your team and your guests should know, and one named person should own reviewing what it produced. A formula that looks right and is wrong by one row is still wrong, just faster.

Jordan: The practical pilot is one low risk workflow. Take a weekly report your team already checks by hand, let Sheets draft it, and compare it against known answers for a month before it touches anything a customer sees.

Alex: Story five. Google opened Gemini four Carbon, its frontier model for demanding coding, research, and multi step agent work, to paying developers through its A P I, and to Google A I Ultra subscribers.

Jordan: Introductory pricing is two dollars per million input tokens and ten dollars per million output tokens. For comparison, that input price is about a quarter of the eight dollars Anthropic charges for its flagship Fable model, which we covered yesterday. Google also previewed a swarm tool that coordinates several coding agents on the same job.

Alex: Google reports Carbon at the top of a major intelligence index with a score of fifty nine, and strong security testing results. Those are the company's own reported benchmarks, so treat them as a shortlist signal, not proof for your business.

Nick: Same rule as yesterday, and it will keep being the rule. Test any new model on your one hard recurring job, count the finished and checked results, and include your review time in the cost. A cheap token that needs three retries is not cheap.

Jordan: Story six, the cautionary tale. A company called Railway published a breach report saying attackers obtained customer secrets that could have affected up to one hundred thousand workspaces. Read that number carefully. It is an estimate of maximum exposure, not a count of confirmed victims.

Alex: The confirmed damage is still ugly. Railway estimates the stolen credentials produced roughly seven hundred forty thousand dollars in unauthorized AI use. Somebody else's keys, somebody else's models, and a six figure bill.

Nick: This is the most copyable lesson of the day, so copy it. Your AI keys and passwords are cash. Store them like cash. One key per job, so a leak can be cut off without shutting down the shop. Set spending limits where the provider allows them. And when a key might be exposed, rotate it first and investigate second, because deleting an old key message does not unspend money.

Alex: Story seven, regulation with a deadline and a fine attached. New York opened registration on Tuesday for its RAISE Act, the state law covering large frontier AI developers.

Jordan: Developers with more than five hundred million dollars in annual revenue must register by January first, twenty twenty seven, when the law takes effect. They must publish safety frameworks, issue public model transparency reports, and report critical safety incidents to the state within seventy two hours. A new office inside the financial services department, called DIGIT, runs the program. Penalties reach one million dollars for a first violation and three million dollars after that. The governor also launched a research center at Binghamton University, with roughly eighty researchers testing advanced systems independently, though that center has no legal authority to compel audits.

Nick: You are not a frontier lab, and neither am I, so this is not your filing to make. It is your vendor question list getting sharper. When a tool holds your customer data, ask whether the company behind it publishes a safety framework, and how fast it tells customers about a serious incident. A seventy two hour clock for the big labs is a fair standard to expect, in writing, from anyone holding your keys and your client list.

Jordan: Quick hit, thirty seconds. Google DeepMind also rolled out Gemini Robotics two point five to its robotics partners. In a proof of concept with Apollo two humanoid robots doing automotive kitting work at Mercedes Benz facilities in Hungary and Germany, Google reports a ninety six percent success rate across more than two hundred evaluated tasks in its testing. It is partner only, and a proof of concept is not a shift record, but factory evidence beats another polished demo video.

Alex: Now Nick's Take, drawn from our lead story.

Nick: Years ago I watched business owners ignore their Google listings until a wrong phone number cost them real calls. The listing was never the marketing. It was the storefront. ChatGPT just built the same storefront inside the answer, and added a voice that talks back in your name. So here is my lesson from fifteen years of client sites. Every new place your business can be described, someone will describe it. The only question is whether the description comes from you, checked against your real prices and your real policies, or from whatever the internet had lying around. Claim the profile. Feed it clean facts. Check it monthly, the same way you check your Google profile and your website. That thirty minutes is cheaper than one customer who shows up angry about a price your own agent quoted. Your facts, your storefront, your job to keep them true.

Jordan: Higgsfield Corner. Today it is a workflow, not a launch, and it ties straight to story one. Higgsfield added Seedream five point five this week, an image model that can generate product images with a transparent background at generation time, which means clean cutouts with no separate background removal step.

Alex: The use is simple. Take your best selling product, generate or clean a sharp cutout image, check it at full size for accuracy, and use that consistent product shot everywhere your business now needs one, your website, your shop, and the product feed you hand to a business profile like the one in story one.

Nick: My checks before anything posts. Does the image match the real product, color and shape. Is any price or claim in the creative actually true today. And would I be fine with a customer holding the post next to the product in my shop. If any answer is no, it does not post.

Jordan: Time for AI or BS, our quick fire game. Three claims, one is true. Alex, claim one. Claiming your ChatGPT business profile guarantees higher placement in ChatGPT answers.

Alex: That one is BS. OpenAI says plainly that claiming does not guarantee placement or sales. Claim two, Jordan.

Jordan: AI agents generated more than six million lines of proof in four weeks checking Signal's encryption.

Nick: True, and I say that as the only host here who would need six million cups of coffee to read it. Alex and Jordan could read it before lunch. They do not sleep. I checked the calendar. I do.

Alex: Claim three. The one hundred thousand workspaces in the Railway report are all confirmed victims whose keys were used.

Jordan: BS. That figure is the maximum exposure estimate. The confirmed harm Railway reported is the roughly seven hundred forty thousand dollars in unauthorized use.

Nick: Score it how you like. The stakes were enormous, as always. The loser pays nothing, because two of us do not get paid and the third one pays for the tools.

Alex: One Thing to Try This Week. This week, do the claim and clean up. Set aside thirty minutes, check how your business appears in the AI answers your customers use, and start the claim process for your ChatGPT business profile if you are eligible. Then line up three facts side by side: your name, hours, and phone number as they appear on your website, your Google profile, and your AI profile, plus your top five product prices. Fix any mismatch at the source, and write down today's mention count as your baseline.

Jordan: To recap. ChatGPT opened a claimable business storefront with agents attached. AI proved selected Signal security properties with six million lines of checked proof. Anthropic priced a trillion dollar debut while OpenAI showed ten billion in quarterly revenue and heavy losses. Claude moved into sheets and meetings, Google opened Carbon at two dollars in, stolen keys cost seven hundred forty thousand dollars, and New York gave AI oversight a registry, a deadline, and fines.

Nick: Do the thirty minute claim and clean up this week. Your facts are your storefront now, so keep them true. More practical AI help for owners lives at webguynick.com. We will see you tomorrow.

author avatar
Web Guy Nick
    X
    Welcome to our website
    Start a project →