What happened

The next visitor to your website may not be the customer, but an AI agent the customer sent: an AI assistant that looks things up, compares prices and places orders for people. In an episode of the interview show Sources released on 21 October, Google CEO Sundar Pichai said agents will reach websites in two ways, and both will keep developing.

The first is the way a person does it: the agent opens a browser and clicks through page by page to check timetables, compare prices and fill in forms. Pichai called this a good approach, because it’s what users do themselves anyway, and it makes it easy for them to keep an eye on things along the way. The second is a dedicated channel that a website opens for agents, where an agent can pick up information and complete tasks directly without clicking through pages. He thinks this route scales better, but it will take longer to build, and it will be messy for a while before it settles on standards everyone can use.

Google has already prepared for this. Since March, the documentation it publishes for website owners has included a new name badge (Google-Agent): agents that Google runs on users’ behalf wear it when they come to your website to get something done. Pichai also said in the interview that the agent features in Gemini, Google’s AI assistant, are mainly available to paying users for now and will open up to more people over the next few months.

Why it matters to you

Say you run a Pilates studio in Neihu, Taipei, and your website has your class timetable, the price of a trial class and an online booking form. On a Thursday evening, an office worker who lives nearby tells his AI agent: “Find me a Pilates trial class in Neihu on Saturday morning, under NT$1,000, book it and let me know.” The agent behaves just as he would, opening several studios’ websites, checking timetables and comparing prices, then going into the booking form of one whose times fit and filling it in field by field.

At that point, the agent acts on what your website says. Suppose the timetable on your site is still last season’s, and the Saturday morning class was dropped long ago. In the past, a customer who saw the old timetable would ring first, you’d say “that class has stopped, how about Sunday?”, and you could still win the booking. An agent won’t make that call. It books the class the page shows and tells the customer “Saturday morning is booked”, while the booking form you receive is for a class that no longer exists. We think that hurts more than not being picked: miss out and you lose one sale; get booked on outdated information and the customer’s first impression of you is “their website isn’t accurate”.

This visitor doesn’t behave like the programs Google usually sends to read pages so you show up in search (known as search crawlers), either. The door of a website carries a notice called robots.txt that says which programs may come in and read. Google’s documentation says its common search crawlers always follow that notice when they come to read pages automatically, while the kind a user sends out to get something done (Google-Agent belongs to it) generally doesn’t, because the fetch was requested by the user. Some owners are tempted to use that notice to keep AI programs out; but an agent arrives carrying a customer’s booking, so even if you could keep it out, the business would go next door. What to look after isn’t the door, but what the agent reads once it’s inside.

What to do now

Act now. Pichai expects the dedicated-channel route to take a long time, but agents are already using the browse-like-a-person route, and Google’s own agent features are already available to paying users; what they act on is whatever your website says today.

Start with the things an agent will act on: timetables, prices, what a trial class includes, and how a booking gets confirmed. The trouble is that these are spread across several pages, written by different people at different times: the timetable may have been updated by the front desk last month, the price page put up by the web agency years ago, and last year’s promotion page never taken down after the offer ended. Outside your website, your opening hours on Google Maps and the timetable on a class-booking platform say the same things too. A person who lands on an outdated page notices something is off and rings to ask; an agent may not be able to tell which page is out of date, and may act on whichever version it reads.

Going through your website page by page and bringing it in line with how you work now, and keeping it that way, is something you can hand to our managed GEO service (GEO stands for generative engine optimisation, and the engines here are AI tools like ChatGPT and Gemini that write answers on their own): our team fixes your site’s content, brings what’s said about you elsewhere into line, and regularly checks how each AI engine describes you; you just review the results.

Before an agent can act on your website, it has to find you and understand you. Whether AI can find you, and whether what it reads about you is up to date, you can already see by asking ChatGPT, which may also go online and read businesses’ websites before answering. To see it live, book in for a demo: a session with one of our consultants where they show you live. In the tone of the office worker at the start, the consultant writes the kind of instruction a customer in your line of business would give, puts it to ChatGPT, and checks which businesses it lists and whether the times and prices it gives for you are the current ones. In the same session, the consultant will also show you your website’s GEO audit score. The GEO audit is geoweb.tw’s free website audit, which checks how well AI can understand your site and scores it across 12 areas.

Further reading