Say you run a land administration agent’s office in Hualien, the kind of firm people in Taiwan hire to handle property transfers and title registration, and you want to add a few articles to your website. So you ask ChatGPT to write one on “What you need to transfer a property, and how long it takes”. In under a minute, out comes a neatly laid-out, professional-sounding article. Looking at it, you’ll probably find yourself wondering: if I put AI-written articles on my website, will Google mark me down?

No, at least not for being written by AI. Google cares about the quality of content; and the documentation it gives website owners now also spells out two things: the facts in an article need to be checked by a person before publishing, and the named author mustn’t be made up. The way we read it, checking the facts means someone answers for whether the content is right, and a real name lets customers see who that someone is. Both come down to the same question: who answers for the article.

The trouble lies somewhere else. The AI has none of your office’s information, so nothing in that article belongs to you alone; an agent in Taipei or Taichung could ask for the same topic and get an article just like it. In our view, the biggest risk in using AI to write your website articles isn’t being penalised by Google. It’s ending up with an “average”: the usual things said in your line of work, averaged out until they fit any business in it. Customers now also put questions straight to AI tools such as ChatGPT and Gemini. The conclusion we’ve reasoned our way to is this: if your article leaves out the things only you know, AI has no reason to mention you when it answers their questions. How we got there comes further down.

AI writing isn’t penalised; Google wants someone to be accountable

Google set out its view of AI-written content on its official blog back in February 2023. The gist: Google focuses on the quality of content rather than how it’s produced; appropriate use of AI isn’t against its guidelines; and using AI doesn’t earn content any extra credit either: it’s just content like any other. If it’s useful and original, it may do well in Search; if it isn’t, it may not.

That doesn’t mean anything goes: Google has drawn a line. Its spam policies include one on “scaled content abuse”: generating lots of pages mainly to manipulate search rankings rather than to help users. It applies no matter how that content is created, whether by people or by AI. Its first example is, in fact, using generative AI tools (AI like ChatGPT that writes text on its own) to generate many pages. But the fault it picks out is doing so “without adding value for users”, not the use of AI.

Google’s current documentation for website owners also spells out two more things. The first is in Google’s AI content guidance, which tells website owners how to use AI-generated content. It says: it is critical to manually fact-check and review all AI-generated content before publishing. The guidance also says why: generative AI doesn’t look facts up; it predicts a likely sequence of words based on what it was trained on, so what it produces may contain inaccuracies.

The second is in Google’s helpful content guidance, which tells website owners how to create content that’s useful to readers. On authors, it says: fabricating creator profiles, such as with AI-generated headshots, made-up names or false credentials, to make content look as if it was written by human experts, is a form of deception.

Put those two together and the way we read them is this: the typing can be done by AI, but whether the facts are right, and whose article it is, has to be owned by a real person.

Even so, an AI-written article with every fact checked and a real author may still do little for you, because there may be nothing in it that’s actually about you.

AI doesn’t know your business, so what it writes sounds like your competitors

Take the office in Hualien. When you ask AI to write the article, you haven’t handed it your price list, and it doesn’t know what cases you handled last month; when it gets to “how agents’ fees work”, it can’t write how you actually charge. We can see two ways this goes. One is that it writes something that doesn’t match how you actually do things, which is exactly the kind of error Google wants you to catch before publishing. The other is that it writes lines like “fees vary depending on the complexity of the case” and “a professional team that takes care at every step”.

You can’t fault lines like those, and checking the facts won’t stop them either; but any agent could say them, and a customer who reads them can’t tell you apart from the office next door.

So what is it that only you know? Suppose most of your work is actually inheritance registration, and you often deal with several heirs living abroad, with documents that have to be posted out one by one, signed and posted back; and the question customers ask most on that first phone call is “If one of the siblings doesn’t agree, can it still go through?” You built up those things case by case. No competitor can write your version of them, and AI has no way of knowing them, because it has never read them.

Google has put this into its own self-check questions, too. The helpful content guidance lists questions for website owners to ask themselves, and a few of them are: Does the content provide original information or analysis? Does it offer anything beyond what everyone already knows? Does it provide substantial value compared with other pages in search results? Put an article that would work for anyone up against those questions, and it can’t honestly answer “yes” to a single one.

However many average articles you publish, AI still has no reason to pick you

Do articles that look much like everyone else’s in your line get written into AI answers less often? We haven’t found any study that has measured this directly, so this section is our own reasoning. What we base it on is how these AI tools answer questions. Customers now often put questions straight to AI tools such as ChatGPT, Gemini and Perplexity, for example, “What should I watch out for when hiring an agent for an inheritance registration in Hualien?”; for some questions, these tools look things up online before answering, pick a few of the pages they find to write from, and list them as sources.

Suppose ten agents’ offices in Hualien all have an AI-written explainer on inheritance registration on their websites, all much alike. Then it makes no difference to the AI which one it draws on; if it picks yours, that’s just luck, and there isn’t a single line it has to get from you. What gives AI a reason to take something from you is the few lines only you could have written: cases you have actually handled, how you actually charge, the things you always check for a customer first.

Even if you wrote ten articles yourself in one go, the result would be the same. Ten near-identical articles put together still add up to that one article, and AI gets nothing new out of them; and if they’re mass-produced for rankings without adding value for readers, they could also run into that “scaled content abuse” line from earlier.

The same goes when a customer asks AI about you by name. If they ask, “How does [your office] charge for an inheritance registration?” and the AI goes to your website and finds only “fees vary depending on the complexity of the case”, that one line is all it can pass on.

The author named on the article has to be a real person customers can look up

The facts in the article have to be yours, and the name on it has to belong to a real person on your side, too. Some articles leave customers wanting to know who wrote them, such as explainers on property transfers and inheritance, which involve money and people’s rights; for a shop, think of a food store explaining how to keep what it sells, or a garage writing about how often brake pads need replacing: before acting on it, customers will want to know who’s saying so. In that 2023 post, Google already advised considering an accurate author byline on content like this. The helpful content guidance also lists a question for website owners to ask themselves: can readers get from the byline to more about the author’s background and the topics they write about?

For the office in Hualien, the name on the articles should be that of a real land administration agent at the office. The name is real, how long they’ve been practising and which kinds of cases they specialise in can be spelled out, and a customer who wants to check can do so. A shop has no licence to look up, so put the real name of the owner or the person who makes the goods, and say how long they’ve been in the trade and what they’re responsible for. Which articles should carry a name first? Our suggestion: the ones customers will act on, or use to decide whether to spend money.

That doesn’t mean they have to type every word themselves. Among the helpful content guidance’s self-check questions is one asking whether content is written or reviewed by someone who clearly knows the topic well. Our suggestion is to label it truthfully: if they wrote it, name them as the author; if they reviewed it, say they’re the reviewer. Either way, the cases and fees in the article need to be ones they’ve seen and stand behind. That’s what a byline means: someone is accountable for the piece. If it’s wrong, someone fixes it, and someone owns it.

Let AI handle the sentences; the facts, experience and byline have to be yours

So the principle for dividing the work isn’t complicated. The AI content guidance opens by noting that generative AI can be particularly useful when researching a topic and for adding structure to original content. Here’s the split we suggest: smoothing out the sentences and tidying up the paragraphs and subheadings can go to AI; the facts in the article, the experience behind it and the name on it have to be your own.

Simple in principle, but doing it yourself tends to stall at the same point: you’re the last to see what only you have. Heirs scattered abroad, documents posted out one at a time: you deal with this every day, so it often feels like part of the job, and it doesn’t occur to you to write it up. And when you read your own article, “a professional team that takes care at every step” quietly gets filled in with everything you actually do, so it always reads as if there’s something there. In one case you don’t think to write it down; in the other you read it and assume it’s already there. When you’re the one inside the office, neither is easy to notice; someone who doesn’t spend every day in your line of work, reading through your material, is more likely to spot which things set you apart.

That outside reader can be us. We can write the new articles for you: our team picks out the things only you have from the material you give us (such as your price list, cases you’ve handled and the questions customers ask most) and writes them up; as for the byline, it works as described above: if your agent has reviewed the piece, say they’re the reviewer. If someone on your side would rather write them, that’s fine too: send us the draft and we’ll revise it before it goes on your website.

Writing articles is only one part of what we do. For the office in Hualien, the result we’re after is that when a customer asks AI “Which agent in Hualien should I use for an inheritance?”, the AI has your cases and how you charge to talk about, and gets them right. Articles alone won’t get you there: what’s written about you elsewhere, such as your office’s listing on Google Maps, has to match your website too. All of that together is our managed GEO service. GEO stands for generative engine optimisation, and the engines here are AI like ChatGPT that write answers on their own. Here’s how it works: 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.

Put that one-minute property-transfer article on your website as it is, and what AI reads is an article that would work with any office’s name on it. To find out whether your website is like that, start with how AI describes you. First settle on something only your business could say, such as which kind of case you handle most; if you run a shop, think of the item regulars keep coming back for, and what makes it different from everyone else’s.

Whether that one thing gets mentioned, you can see for yourself: open ChatGPT, type “Tell me about [your office]” with your own name in the brackets, and see what it says.

What’s harder to judge is the rest of the answer. Which lines are really about you, and which would work for any office? That’s hard to tell on your own: read “a professional team that takes care at every step” and you’ll quietly fill it in with everything you do every day, and think it’s covered you. This is the step where you can use a demo to take the AI’s answer apart. The demo is a session with one of our consultants where they show you live. The consultant asks ChatGPT to describe your business, or puts the question to it directly: “How does [your office] charge for an inheritance registration?” Once the answer is in, they go through it with you line by line and sort it into two piles: lines about things that are really yours count as yours; lines that would still make sense with the office next door’s name in them are the “average” we talked about.

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.

Whether you ask it yourself or at the demo, if ChatGPT can’t say much about you at all, that’s an answer too: most likely, AI hasn’t got hold of your information yet.

Further reading