Something has quietly changed in how people look things up. A growing number of them no longer open Google, scan ten blue links and click around. They ask an assistant a question in plain words and read the answer it writes back. ChatGPT, Perplexity, Claude, Gemini — the shape is the same: one question, one answer, and sometimes a short list of sources underneath.
If your business is going to be found this way, the question is no longer only "do I rank?" It is "when the assistant writes its answer, does it name me — and does it cite me as where it got that?" That is a different game, and it pays to understand how it actually works before you try to win it.
How these tools actually answer
There are two broad ways a standalone assistant produces an answer, and most tools now mix them.
Some answer largely from what they absorbed in training — a vast, blurred picture of the public web as it stood months ago. Ask a general question and you get a confident paragraph with no links, because nothing was looked up; it was recalled. Others browse live. Perplexity and ChatGPT's browsing mode go and fetch pages as you ask, then write an answer that cites the specific pages with links. Most assistants now do a bit of both — recall for the shape of the answer, live retrieval for the specifics and the citations.
That split matters, because the two levers you can pull are different. Training-based recall is slow to change and reflects your long-run reputation. Live retrieval reflects what's fetchable right now on pages the tool can read. You want to be in good standing on both.
Why this isn't Google ranking
On Google, there is a ranked list. Pages compete for position, a person scans the list, and the click is theirs to give. The mechanics are well understood and, to a degree, gameable.
An assistant does not hand you a list. It assembles one answer and, if it cites, reaches for a handful of sources it can find and trust in that moment. There is no position 1 to win. There is only "did the answer mention me, and did it point at my page as a source?" Two things move that needle far more than they move a Google ranking.
The first is being clearly, quotably useful on your own pages — so that when a tool does fetch your site, there is a clean sentence it can lift. The second is your reputation off your own site — because a large part of what the model "knows" about you was never on your website at all. It is the sum of what the rest of the web says. (Google's own AI answer is a third surface again, with its own rules — I cover that in a separate piece on AI Overviews in Singapore, because it behaves differently from the standalone assistants here.)
Be quotable on your own pages
When an assistant fetches your page, it is looking for a sentence it can stand behind. Make that sentence easy to find.
- ✓Answer the real question in the first line. Not a welcome message, not a brand slogan. If the page is about a service, the opening sentence should say plainly what it is, who it's for and where you do it.
- ✓Write plain declarative sentences. "We fit and service air-conditioning for HDB flats across Singapore" is quotable. "Your comfort is our passion" is not — there's nothing in it to cite.
- ✓Put the facts in clean text. What you do, where you are, who you serve, and prices or ranges if you can share them. In real words on the page — not baked into an image, a video, or a PDF the tool may never open.
The test is simple: could a stranger copy one sentence off your page and use it, unedited, as the answer to a question? If yes, an assistant can too. If everything is vague or trapped in a graphic, there is nothing to lift. (There's more on structuring pages this way in my guide to showing up in AI answers.)
Your reputation off your own site
Here is the part people underestimate. The model's picture of you is roughly the average of what the web says about you — not just what you say about yourself. If ten directories, a handful of reviews and a couple of industry sites all describe you the same consistent way, that description becomes what the assistant "knows". If the web is thin or contradictory about you, the model has little to go on and reaches for whoever it can describe clearly instead — often a competitor.
So the off-site work is unglamorous but it compounds:
- ✓Consistent name and description everywhere. Same business name, same one-line description, same category, across every directory, profile and listing you appear in. Contradictions blur the picture.
- ✓Reviews and profiles that actually describe what you do. Not just a star rating — words a model can read and summarise.
- ✓Mentions on sites that cover your field. Being written about, listed, or quoted somewhere the tool already trusts does more than another page on your own site.
None of this is a trick. It is simply making sure that the true, useful version of you is the version the web repeats most often.
Freshness and what can be fetched
For the tools that browse, there is a blunt limit: they can only cite what they can actually fetch. A fact locked behind a login, rendered only by heavy scripting, or living inside a downloadable file is, for citation purposes, often invisible.
So keep the facts that matter — what you offer, where, current pricing, who to contact — in plain, crawlable text, and keep them current. A stale page gets quoted with stale facts, or passed over for a clearer one. Freshness here is less about publishing constantly and more about your key facts being right, readable and reachable on the day someone asks.
On the technical edge, some sites are starting to add a small file called llms.txt that points AI tools at their most important pages. It's worth knowing about, but it's a narrow, emerging thing and not the main event — I've written a separate explainer on llms.txt if you want the detail.
How to check where you stand
You don't have to guess. You can look.
Open ChatGPT and Perplexity and ask them the questions your customers would actually ask — the problem in plain words, not your company name. "Who does affordable bookkeeping for small businesses in Singapore?" rather than "tell me about [your firm]". See who gets named and cited. Do it from a clean, logged-out session, because a tool that already knows you will flatter you with your own history. Run each question a few times; the answers shift from run to run, and you want the pattern, not a single lucky hit.
If a competitor keeps coming up and you don't, that is your brief. Usually it means the web says more, and says it more clearly, about them than about you.
The honest limits
I'd be doing you a disservice to pretend this is controllable the way an ad campaign is.
You cannot dictate a model's output. Citations vary from one run to the next, even for the same question. Anything that depends on training-based recall lags by months, so improvements you make today may not show up in a model's "memory" for a while. And no one can sell you guaranteed placement inside an AI answer — there is no slot to buy.
What you can do is durable. Be the clearest, most useful, most consistently described answer to the question your customers ask. Do that and you are the easy thing for an assistant to reach for — whichever assistant it is, and however the mechanics change underneath. That is the whole of it: stop trying to game the answer, and become the answer.
Common questions
They do not run a ranked list the way Google does. The assistant assembles an answer and reaches for sources it can retrieve in the moment and trusts — pages that clearly answer the question, plus whatever the wider web consistently says about you. Being quotably useful on your own pages and being described the same way across the web both matter more here than keywords.
No. There is no ad slot inside the answer and no fee that makes an assistant name you. What earns a mention is being genuinely useful, clearly described, and consistently referenced on places these tools trust. Anyone selling guaranteed placement inside an AI answer is selling something that does not exist.
Usually because the web says more, and says it more clearly, about them. They may be described consistently across directories and reviews, mentioned on sites that cover your field, or simply have the plainer page that answers the question. The model's picture of a market is the average of what it can find — if you are thin or inconsistent, you get left out.
Ask it. Open ChatGPT and Perplexity in a clean, logged-out session and type the questions a customer would actually ask — not your company name, but the problem they have. See who gets named and cited. Run it a few times, because answers vary from run to run.
Yes. Ranking puts a link in a list and a person chooses. A standalone assistant writes one answer and may name a handful of sources inside it. Google's own AI answer is a separate surface again. The overlap is real — clear, trusted, useful pages help everywhere — but the mechanics and the payoff are not the same.
We build websites and run SEO & AEO for Singapore businesses. HeyAda is a Singapore web studio — design, copywriting, code and SEO, handled end to end. We write these guides to share what actually works, in plain English. If your site isn't getting found, say hi.
Want to be the answer AI gives?
I help Singapore businesses become the clear, trusted, quotable source an assistant reaches for. Tell us what your customers ask, and we'll show you where you stand today — and the plain steps to get named. That's our AEO service.