Say you run a campsite in Taitung, on Taiwan’s east coast. A few years ago, someone planning a camping trip would most likely type “Taitung campsite pets” into Google, then click through the list of pages that came up, one by one, to see whether they could bring the dog and whether there were rain shelters. These days some customers search a different way: they open ChatGPT, or use “AI Mode” in Google Search (it writes you an answer and lets you keep asking follow-up questions), and type the whole thing in: “Is there a dog-friendly campsite in Taitung? It needs rain shelters, a power socket by the pitch, and hot showers in the evening.” The AI goes and finds information on the various campsites, replies with a few of them, and adds the links.

You may have heard this shift described as search moving from a “keyword era” into an “intent era”. In the keyword era, customers boil what they need down to a few words and do the rest of the looking themselves; in the intent era, they say the whole thing, the AI works out what they actually want (that’s what “intent” means here), and then does the looking for them. This piece sets out to make two things clear: what has actually changed, and what it means for your website. The short version: whatever your line of business, the better AI understands what customers want, the more it goes by their conditions and drops the businesses that don’t spell things out.

Google has long been guessing at intent, but customers still had to find the answers

Google has been talking about “intent” for a long time. Its official page on how Search works says that to return relevant results, it first needs to establish what you’re looking for, the intent behind your query. Its example: search for “pizza” and you’ll see nearby businesses that deliver. The way we read it, you type a single word and it guesses you probably want to order one.

In October 2019, Google announced it was using BERT in Search, an AI technique that helps computers read a whole sentence, and its official blog said that because BERT looks at the words that come before and after a word, it’s particularly useful for understanding the intent behind searches.

The same post on BERT also admitted that Google sometimes still doesn’t get it right, particularly with complex or conversational queries. That, Google said, is one of the reasons people often use “keyword-ese”: typing strings of words they think Google will understand, which aren’t how they’d naturally ask a question. The “Taitung campsite pets” our camper typed is exactly that kind of search.

However well it guessed, though, what Google handed back was still a list of pages. Looking back in a 2013 post at what searching was like in 1998, Google’s official blog put it like this: you typed in some keywords and got 10 blue links to websites that had those words. Search has become a lot smarter since then, and the results page has gained images, videos and maps; in our view, though, the main thing on it is still that list of pages. That’s why we say search used to do only half the job: it did its best to guess what you wanted, then handed you a list of pages; which campsite takes dogs and which has rain shelters, the customer still had to click through page by page to check.

AI hears the whole sentence and checks the conditions itself

That second half is what AI takes over. In May 2024, announcing new capabilities for AI Overviews (the AI-written summary at the top of ordinary search results), Google wrote that rather than breaking your question into multiple searches, you can ask your most complex questions, with all the nuances and caveats you have in mind, all in one go.

Back to our camper. That one sentence sets four conditions at once, and in ChatGPT or AI Mode the customer doesn’t have to break them up and search for each one: the AI will usually go online, find information on the campsites and sort out a few that meet the conditions.

So this is what’s genuinely new about the “intent era”. Working out what the customer wants only gets done completely once AI is involved: it doesn’t stop at understanding the question, it also checks the conditions and gives the answer for the customer. Bear in mind that which era you’re in has nothing to do with which company you use; it depends on what ends up in the customer’s hands: a list of links, or an answer. Even within Google, ordinary search gives you a list of links, while AI Overviews and AI Mode give you an answer written by AI.

Keywords are still there; AI types them behind the scenes for the customer

So are keywords useless now? They’re still around; they’ve just moved behind the scenes. Google says AI Overviews and AI Mode may both use what it calls a “query fan-out” technique: breaking your question down into subtopics, issuing multiple related searches at the same time on your behalf, and using what they find to put together a response.

ChatGPT does something similar. According to OpenAI, ChatGPT may search the web automatically when your question would benefit from current information. It sometimes works with partner search providers to find information, and when it does, it typically rewrites your question into one or more targeted search queries first and sends those. OpenAI gives an example: someone asks “What are some good restaurants near me?”, ChatGPT works out from their rough location online that they’re in the San Francisco area, and may rewrite the question as the search query “top restaurants San Francisco”.

In other words, someone is still typing keywords; it’s just a different someone. Before, the customer typed “Taitung campsite pets” themselves; now AI types for them. Which words the AI actually types behind the scenes for our camper’s question, the customer can’t see on screen, and neither can you; we can only imagine they’d be along the lines of “dog-friendly campsite Taitung” or “Taitung campsite rain shelter”. And since you can’t see them, you can’t go after rankings for those words one by one the way you used to.

So should your website still do SEO (the work that gets web pages ranking well on Google)? Yes. Google is clear: a page has to be indexed by Google (meaning Google has stored that page in its records) and eligible to appear in its search results before it can be shown as a supporting link in AI Overviews or AI Mode. Google also says SEO best practices remain relevant for AI Overviews and AI Mode. So the parts of SEO that make sure Google can index and understand your pages are still worth doing; that’s what lets AI find you in the first place. What changes is something else, in our view: getting a handful of words to the number one spot is unlikely to win you AI’s recommendation; once AI has found you, whether it recommends you comes down to what your website says.

The better AI understands customers, the faster it weeds out businesses that aren’t clear

This is where what your website says starts to show. Suppose your campsite has rain shelters, a power socket at every pitch and hot water around the clock, and dogs are welcome, but the line about pets on your website says: “If you’d like to bring a pet, please message our Facebook page first.”

In the keyword era, that line didn’t do much harm. A customer clicking through would see rain shelters, power and hot water all there, with only the dog question left, so they’d send you a message; you’d reply “Yes, just keep it on a lead,” and the booking would most likely go ahead.

When AI does the reading for the customer, in our view, it plays out differently. The customer wants “dogs allowed”, the AI can’t confirm that on your page, and the campsite next door says “Pets welcome, up to two per tent”, so the one that makes it into the answer is most likely that one. AI won’t send that message on the customer’s behalf, so you never get the chance to say “yes”.

That’s the part of the intent era that matters most to business owners. Before, the message a customer sent filled in whatever your website had left unsaid; now, it has to be said on the website first. AI is the one that understands the customer, but it still has to take the answers from information it can find online, and the part of that you control most is your own website. Whatever you haven’t made clear yourself, however smart the AI is, it can’t say for you.

The question for owners becomes “which conditions will customers set?”

So the question business owners ask needs to change. It used to be: “Which keywords do I need to rank for?” Now it should be: “Which conditions will customers set, and does my website clearly answer them?” In your line of work, the conditions are probably the things customers most often ring or message you to ask about. If you run an optician’s, for example, the questions customers ask most are probably whether they can collect their glasses the same day and whether you’re open on Sundays.

Answering clearly means the AI can be sure of the answer once it has read it: yes or no, there or not, how much, open until when. When AI reads a line like “Feel free to message us”, it may well treat that condition as having no answer.

The intent era brings one more change, and businesses that sell products will feel it most: customers no longer need to know what they want to buy. Say your brand sells lip balm. In the keyword era, a customer had to know they wanted lip balm before searching “best lip balm”, then pick from the brands Google listed. Now a customer might just ask AI: “My lips keep cracking. What should I do?” Once the AI has covered how to look after them, it may go on to recommend a few lip balms. If the customer adds that their lips react easily and that winters where they live are cold and dry, the AI can narrow it down further, to the one that suits that situation.

For the AI to pick that one, someone has to have written the answers down: which balm is fragrance-free, which suits cold, dry weather, which is meant for overnight. If every product on a brand’s website just says “Moisturises and protects your lips”, the AI can’t confirm a single one of the conditions the customer set, and has no reason to pick you for that answer; the brands that spell out what each product suits are the ones with a chance of being picked by AI, and of having the right product picked.

Whether you sell lip balm or run a campsite, spelling out the conditions sounds simple, but when you do it yourself, it’s easy to miss things, because every customer sets a different combination of conditions. Take the campsite in Taitung: one person has a dog and is worried about rain; another has two kids and needs both power and hot water; someone else only cares how close the site is to town. If you put a question to AI yourself and see your campsite in the answer, all that tells you is that you passed that one set of conditions; with a different set, the result may change. And when you’re not on the AI’s list, it usually won’t mention which condition tripped you up.

On top of that, AI doesn’t only read your website: your business details on Google Maps, booking sites and bloggers’ camping write-ups may all be used to check conditions, and what they say has to match your website too. This is ongoing work: finding the conditions customers will set, one by one, and writing them up on your website as answers AI can confirm; tidying up what’s said about you elsewhere to match; then going back to AI regularly to see whether it can answer them.

This kind of work is called generative engine optimisation; take the initials of the three words and you get GEO. It’s aimed at AI tools such as ChatGPT and Gemini that go online, read up and write answers on their own. You can hand all of this to geoweb.tw: with our managed GEO service, 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.

Our camper set four conditions in one go. To find out which ones your customers set, and how AI answers right now, put it to the test at a demo. The demo is a session with one of our consultants where they show you live: you name the three or four things customers most often message you about, and the consultant strings them together into one full question and puts it to ChatGPT there and then. Which businesses it lists, whether you’re among them, and whether it can answer those conditions when it talks about you, you’ll see it all on screen. One question only tests one set of conditions, but seeing AI pick businesses with your own eyes once usually shows where you might be getting stuck; trying different combinations and checking back regularly is the ongoing work described above. 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