ChatGPT Wallet Lets AI Agents Check Out for You

The GEO Show
November 16, 2026
Watch this episode on YouTubeListen on Apple Podcasts, Spotify and more

OpenAI launched ChatGPT Wallet so its agents can complete purchases with a saved card, approved by Face ID or a fingerprint. Paris Childress explains why checkout is moving to the agent and the card network, then covers HUMAN data showing Meta's Muse sends the most AI agent requests, the Internet Licensing Collective and Yahoo Scout, Google testing results without the AI Mode tab, chatbots pointing shoppers to risky retail sites, Reddit's 13% traffic drop and Canada's OpenAI copyright case.

Key takeaways

  • ChatGPT Wallet moves checkout to the agent and the card network, so check that your bot protection lets verified agents through.
  • Meta's Muse, not ChatGPT, sent the most AI agent requests in HUMAN's data.
  • Split agent traffic by user agent and signature before you allow, block or optimize.
  • Citations are becoming a priced input for publishers through the ILC.
  • ChatGPT Wallet: OpenAI's agents can check out with a tokenized card and biometric approval, and bot protection settings can quietly break agent checkout.
  • Meta's Muse leads agent traffic: HUMAN Security says Muse sent 59% of the AI agent requests it saw, nearly four times ChatGPT agent's share.
  • The Internet Licensing Collective: the relaunched RSL Collective signs Yahoo Scout as its first paying licensee, turning grounding and citations into priced inputs.
  • Google tests results without the AI Mode tab: what a merged experience would mean for measuring AI Mode.
  • Chatbots and risky retail sites: up to 15% of answers pointed shoppers to risky or fake stores, and smaller brands are most exposed.
  • Reddit visits fell 13%: why Reddit is a citation channel, not a traffic source.
  • Canada: the publishers' copyright suit against OpenAI stays in Ontario, under Canadian law.
  • Lightning round: AI images return to AI Overviews, Brave encrypts result links, Dejan maps brands in Gemini 4 Hydrogen, a new AI Max click parameter, and United Airlines terms.

Full transcript

Paris: Welcome back to The GEO Show, brought to you by GEOforge, full self-driving for AI visibility. I'm your host, Paris Childress, recording on Monday, November 16th. ChatGPT can now pay for things on your behalf, but the agent knocking on your site most often this fall wasn't from OpenAI at all. Let's get into it First story, OpenAI launches ChatGPT Wallet so its agents can check out. On Friday, OpenAI launched ChatGPT Wallet.

Paris: It lets OpenAI's dots agents complete purchases with a saved card, and every purchase needs a face ID or fingerprint check. It's starting with a limited group of US ChatGPT Pro subscribers. So here's how it works. The user adds an eligible US-issued Visa, Mastercard, or Amex card once. The agent then pays with a network payment token, so neither the agent nor the merchant sees the real card number.

Paris: Users review the items, total, and card, then approve each purchase with a device security check. It's limited number of US Pro users aged eighteen and over. Spending limits and pre-approved controls are planned with no given date. PPC Land, citing OpenAI's post, lists fraud partners Forter, Riskified, Sardine, and Signifyd, and payment providers Adyen, Checkout.com, and Stripe OpenAI's commerce partner playbook tells merchants to let its cloud browser through. Verified agent requests are signed in with Web Bot Auth, and the playbook gives allow rule examples for Akamai, Cloudflare, Human, and Vercel.

Paris: It warns that sites which block agents may lose users, and it doesn't mention the agentic commerce protocol. All right, this is a steady march for agents towards checking out. Checkout is moving to the agent and the card network. The merchant's job is becoming reachable and verifiable by agents, and bot protection settings can quietly break that Is this OpenAI's second try at native checkout different because the agent pays on the merchant's existing site instead of asking merchants to integrate? Keep in mind here that this is pro only, US only for now, and every purchase needs biometric approval That said, I think that the writing is on the wall here.

Paris: Agents are gonna continue to take more actions on users' behalf, including now checking out and making payments with an authorized card. Next up, Meta's Muse now sends most agent requests to the sites it protects. Human Security's October report says that Meta's Muse agent sent fifty-nine percent of the AI agent requests it saw on the sites that it protects. That's nearly four times ChatGPT Agent's share. Total agent requests grew two point six x in just one month.

Paris: Here are the relative shares. Meta Muse at fifty-nine percent, ChatGPT Agent at fifteen point nine percent, Codex browser at eight point eight percent, Codex Chrome extension at five point two percent, Perplexity Comment, Perplexity Comet at three point six percent, Claude Chrome extension at three point five percent, and Gemini agent at zero point eight percent. So I suppose if you add up ChatGPT Agent, Codex browser, and Codex Chrome extension, that total is roughly thirty percent. So it's a significant number too if you add those three together. But the headline is that Meta Muse is really dominating here on the share of agent requests.

Paris: Agent requests were two point six times higher than just September's level and eight times higher than August's level. Eighty-five percent of requests hit product and search pages, and just one point nine percent hit checkout, though the checkout requests did grow one hundred and seventeen percent month over month. If we break this down by sector, e-commerce is sixty percent of this, travel is nineteen percent, media is also nineteen percent. And Search Engine Journal adds DataDome's report that Muse moved its web bot auth signing domain to muse.ai between 7th and 9th of November. About half of Muse's requests were unsigned as of November 10th Well, this is quite eye-opening, and it really highlights the meteoric success of Meta Muse as the probably the leading agent right now.

Paris: So optimizing for agents isn't only about OpenAI. According to this vendor's data, Meta sends the most agents to commerce pages right now. My advice here is to split your agent traffic by user agent and signature in your server logs before deciding which agents to allow, block, or optimize for. Next story. Publisher collective ILC launches, and Yahoo Scout is its first paying license.

Paris: The RSL Collective has relaunched as the Internet Licensing Collective or ILC. Its first commercial licensee is Yahoo Scout, which will pay into a royalty pool for using publisher content in AI answers. The amount was not disclosed So what is this ILC? It's actually a nonprofit publisher-led collective with already more than three thousand members. The license is a blanket license covering training, retrieval or RAG, retrieval-augmented generation, grounding, agentic use, and AI Overviews.

Paris: Joining is free and non-exclusive. Operating costs are projected at or below ten percent. So here's how payment works. Each licensee pays into its own royalty pool. Payouts follow rules that weigh how content is used and whether it's cited.

Paris: Co-founder Doug Leeds declined to give Yahoo's fee or the exact formula. Mather Economics estimates collective licensing should be worth somewhere between fifteen to thirty-five percent of AI company revenue. Digiday says that based on OpenAI and Anthropic revenue, and Mather says that the ILC didn't pay for the study. ILC says that it's in talks with some major AI companies, but didn't name them specifically So this is a sign that grounding and citation are becoming licensable priced inputs. Over time, that could affect which sources answer engines can cite or choose to cite So that makes me wonder, if citations drive royalties, will publishers start optimizing to be cited the way that brands do?

Paris: On to the next story. Google tests result pages without the AI Mode tab or the Ask Anything box. SEO consultant Glenn Gabe posted screenshots of Google results with no AI Mode tab and no Ask Anything box under AI Overviews. Search Engine Roundtable couldn't produce it, and Google hasn't commented. Gabe's screenshots on X show the missing tab and an AI Overview without the follow-up box.

Paris: Search Engine Roundtable's Barry Schwartz couldn't reproduce it. This comes about ten days after a test that removed the AI Mode button from the Google homepage. Some observers suspect a monetization test. Chris Long suggested that AI Mode may no longer be a separate experience. Both are just speculation, and Google has not made a statement at all on this.

Paris: So if AI Mode merges into the main results instead of sitting behind a tab, it gets harder to measure as a separate service. That said, I would advise to keep reporting AI Overviews and AI Modes separately for now, but expect that the line between them will continue to blur. And my opinion is that AI Mode is the ultimate Google experience. It is inevitable, and it's really just a question of when, but not if Next up, chatbots pointed shoppers to risky or fake retail sites in up to 15% of answers. Brand protection firm PODQI, P-O-D-Q-I, asked ChatGPT, Gemini, and Perplexity where to buy 151 brands online.

Paris: Risky sites such as impersonators Or counterfeit sellers showed up in 15% of Perplexity's answers, 8% of ChatGPT's, and 4% of Gemini's. The test used six thousand seven hundred and ninety-five answers across one hundred and fifty-one brands in two September test periods using prompts like, "Where can I find brand online?" And what counts as risky? Impersonators, counterfeit sellers, abandoned domains that someone else re-registered, and inactive domains that could be reused. The study measures exposure but doesn't confirm fraud. In other findings, when ChatGPT surfaced a risky site, it warned the user twenty-four percent of the time.

Paris: Small brands saw about double the exposure of large brands, and questions about regional storefronts did worst Google's response, they said they were happy with Gemini's performance. OpenAI said it's working on safeguards, and Perplexity says it doesn't claim perfect accuracy. So for smaller brands, which site is the official one is now a question that AI answers often wrongly Next story. Reddit global visits fell 13% year over year as per Similarweb. Business Insider using Similarweb data reports that Reddit's global visits fell 13% from October 2025 to October 2026.

Paris: Business Insider blames Google's AI Overviews and Reddit's shares fell 6% that day. So a 13% drop year over year based on Similarweb estimates causing a 6% decline in Reddit stock. US daily active uniques fell from 53 million in Q2 to 50 million in Q3, and all of the drop was logged out users. CEO Steve Huffman said AI Overviews are not traffic drivers So Reddit is one of the most cited sources in AI answers still, but fewer visits from Google could mean less fresh discussions feeding those answers And my advice here is don't plan on Reddit as a traffic source at all. Treat it as a channel for citations and how people perceive your brand.

Paris: The next story from Canada, Canadian publishers' copyright suit against OpenAI stays in Ontario. Ontario's Court of Appeal rejected OpenAI's bid to move a copyright suit by Canadian news publishers to the US. The case will be decided under Canadian law, which has no US-style fair use defense. The ruling was issued on Friday, November 13th. The plaintiffs include The Canadian Press, Torstar, The Globe and Mail, Postmedia, and CBC/Radio Canada, who filed in 2024.

Paris: This only settles where the case is heard. The merits, including where the training copying happened, are still open. Next steps, Tech Times reports the decision was unanimous, and that OpenAI has 60 days to seek leave to appeal to the Supreme Court of Canada So winning on fair use in the US won't protect AI companies everywhere, including even Canada, and licensing pressure is likely to come country by country. Next up, AI-generated images are back in AI Overviews for some users. Google seems to be generating images on the fly inside AI Overviews again for some searchers.

Paris: It tested this in July and paused it in August. Search Engine Roundtable reproduced it in some browsers, but not others. In August, Google called it a small experiment we're no longer running, and there's no statement from Google here. So generated images take up space that publisher images and links could otherwise fill And let's wrap up with some lightning round. Brave Search will encrypt its result links.

Paris: Brave says scrapers shifted to its search after Google's change to go-to redirect links, reaching several thousand requests per second. Clicks will now go through a privacy-preserving redirect. Next, Dejan mapped two hundred and thirty-nine thousand two hundred and sixty-six brands in Gemini 4 Hydrogen's built-in knowledge. It ran fifty-two thousand nine hundred and twenty-three queries with search turned off, and Apple, Microsoft, and Google ranked top. Next up, a new Google Ads tracking parameter flags clicks that came from AI Max, even when they were labeled exact match.

Paris: That's very interesting. And last, United Airlines website terms also bar using its data to ground AI Models. All right. That's a wrap for today's episode. Thank you all for tuning in, and see you on the next one.

See how AI answers describe your brand

GEOforge measures how often ChatGPT, Google AI Overviews and AI Mode cite you, then builds the content and citations that change it.