Within less than a week, OpenAI and Google each published official documentation showing websites how to let AI agents in, and in the same few days ChatGPT’s agents started being able to check out for people. An AI agent is an AI assistant that, given a single instruction from a customer, goes online by itself to find things, compare prices and place orders for them. Our position is this: the platforms already treat these agents that run errands for customers as proper visitors wearing name tags, and are putting out their own manuals to show businesses how to let them in. Whether your website blocks them comes down to how your firewall or the company that manages your site has it set up right now, and you may never have seen those settings, so you need to find out now.
What happened
On 13 November, OpenAI launched ChatGPT Wallet, a wallet that stores your credit cards. A customer tells ChatGPT’s agent something like “Help me find a new sweater”, the agent goes off to search and compare, and at checkout the customer confirms with face recognition or a fingerprint, and the agent completes the payment for them. For now, the feature is open only to ChatGPT Pro subscribers who live in the US and use a US-issued credit card, so customers in Taiwan can’t use it yet.
According to the tech publication TechCrunch, OpenAI published a playbook for businesses at the same time. The playbook, which OpenAI has posted on its developer site, lists six things that businesses can do, and the second is to check whether your CDN or firewall is shutting out agents that have been verified as genuinely sent by ChatGPT. A firewall works like a security guard at your website’s door, stopping any visitor that looks like trouble; a CDN is a service that makes your website load faster, and many of them double as that guard. OpenAI writes in its guidance that a website doesn’t need to set up an exception for its agent in advance, since the agent can already use sites, but that a website’s security settings may block it unintentionally.
On 17 November, Google also added a page to the documentation that it writes for website managers, explaining how to allow its own agents through. The name tag that Google’s agents wear when they go online to get things done for users is Google-Agent, and it has existed since March; what’s new is this guidance on allowing it. Google writes that explicitly allowing Google-Agent in protections such as your CDN or firewall helps prevent legitimate requests from being blocked. Earlier still, on 4 November, Cloudflare, which provides website security, had already let site owners set whether to allow or block these agents one by one.
Why it matters to you
Say you run a pottery studio in Caotun, Nantou, and your website, which a web design company built for you a few years ago, takes online bookings for wheel-throwing sessions. The hosting, the CDN and the firewall were all handled by that company; you pay a maintenance fee every year, and nobody has ever told you what was set up.
Once Wallet reaches Taiwan, following the route that OpenAI’s playbook lays out, here is how it would go for you: a customer tells an agent, “Find somewhere in Caotun where two adults and two kids can make pottery together this Saturday afternoon”, ChatGPT lists a few options for them to compare, they pick you, and the agent opens your website, picks a time slot, enters the number of people and pays the deposit. Whether the agent gets in once it reaches your door depends on how your website is set up right now.
What stops it could be something quite ordinary. The troubleshooting list in Google’s documentation names several things: some protections block all connections coming from data centres, and while real customers connect from home or mobile networks, Google’s and OpenAI’s agents connect from computers in data centres; some throw up a CAPTCHA when a form is submitted, the “I’m not a robot” check that you have to tick; and some websites have added an allow rule, only for it to be overridden by a blocking rule that takes effect first.
An agent that’s stuck outside can’t read your website, so it has no way of booking a session with you for the customer. Cloudflare also pointed out in its announcement that blocking agents altogether can feel like the safest choice, but it can mean losing customers.
What to do about it
Move on this now: first find out whether your website is open or closed to these agents, and change whatever is blocking them. Allowing them doesn’t mean tearing down the protection that you already have, as both Google’s and OpenAI’s documentation make clear that what gets allowed is only agents whose identity can be verified. It makes no difference when these agents reach Taiwan, because whether your website is open or closed to them has already been set.
This isn’t something that you can settle by looking at a single switch. To get one task done, an agent has to go all the way from picking a time slot to paying the deposit, and Google specifically notes that allowing agents should cover actions beyond the homepage, such as search, the shopping cart and bookings; if any step is blocked, the whole errand fails. There’s also more than one agent, each with its own name tag. So the settings have to be checked step by step against the route that each agent will take, and this kind of check is safest in the hands of people who know these settings well.
Firewall settings sit in a back-end console that we can’t see in a demo, so a demo looks at the step before an agent reaches your door: on the route OpenAI’s playbook lays out, the customer first picks you from the businesses that ChatGPT lists, and only then does the agent head for your website. Book a demo with one of our consultants, and they will put what that customer said to ChatGPT, so you can see whether you’re among the businesses that it lists for the customer to choose from. The same session also brings up your website’s GEO audit score; the GEO audit is geoweb.tw’s free website check, which looks across 12 dimensions at whether AI can make sense of your website.
Whether you’re on ChatGPT’s list can go into managed GEO for the long run. We revise the content on your website, bring what other websites say about you in line with it, and regularly put the questions that customers ask to ChatGPT, Gemini and other leading AI engines, saving the answers so that they can be compared.
Further reading
- What is an AI agent, and how is it different from asking ChatGPT? — asking ChatGPT gets you a list, while an agent goes on to look, compare and book for you; this piece explains the difference from the start
- AI agents will browse websites like people, Google’s CEO says — what the Google-Agent name tag is, and why the robots.txt notice at your door usually doesn’t apply to it
- Big companies are laying the rails for AI agents to pay — which route the money takes: x402 and the card networks, and which of the two matters for a business’s sales