Do we still need a website if AI answers the question first?

7 min read · Updated 1 September 2026

What is actually dying.

Not search. The answers appearing above the results are generated from the same index that produces the results — Google says so plainly, describing its generative features as rooted in the core Search ranking systems. Nothing has replaced the index. If you are not in it you are not in the answer either.

What is being removed is the click. For most of the web's commercial history a search engine's job was to hand you a list and let you choose; its business was the visit. Increasingly it answers the question itself, and the visit becomes optional.

We can show that on our own site. Searching a competitive thermography query in Malaysia in incognito on 1 September put our page high in the organic results, while the answer above it named two other companies. Treat that as an observation rather than a ranking — these results are personalised and generated fresh each time, and Search Console puts the same page at an average position of 23.3 across August. The distance between those two figures is itself the point. Where you sit in the list and whether you are the sentence quoted above it are different things, measured differently, and the second one is not in your reports at all.

Only what is stated survives.

This is the part worth sitting with, because it inverts a lot of ordinary web advice.

When a machine summarises your company, everything about how you present disappears. The design, the photography, the tone, the care taken over a headline — none of it is carried into the answer. What is carried is the claims: the sentences specific enough to be lifted and repeated.

A company that publishes a price gets its price quoted. A company that routes every price question to a contact form gets described as offering tailored or bespoke pricing, which is accurate, unhelpful, and sits next to a competitor whose actual figures are being repeated on their behalf. The same is true of scope, of turnaround, of what is not included.

We know that one from the receiving end. A summary of our own market described our pricing as tailored and quoted a competitor’s ringgit figures, for the simple reason that they had published theirs and we had not. We have since published ours as bands, which is the entire fix. It required no new technology.

So the thing being devalued is presentation, and the thing being revalued is explicitness. That is an uncomfortable trade for an industry that mostly sells the former.

Which changes what a website is for.

If the click is optional, a website stops being mainly the thing that introduces you and becomes mainly the thing that confirms you. Somebody hears your name in a meeting, sees it in a summary, gets it from a colleague — and then looks you up to decide whether to take you seriously.

Those are two different jobs and they reward almost opposite things. Acquisition rewards reach, volume and repetition. Verification rewards specificity, consistency and the willingness to state a limit. A site built to be found will use broad claims and superlatives. A site built to be checked will publish the number, name the exclusion, and say what it cannot do.

Most company websites in Malaysia are still built for the first job. When we scored 400 contractors on their websites, the weakest dimension across every site we assessed was proof of completed work — the single thing a verifying buyer is looking for.

The measure has to change with it.

This has an awkward consequence for anyone who takes measurement seriously, ourselves included.

It is worth being precise about what is and is not visible. A click from a Google AI Overview arrives as ordinary organic search and cannot be separated out. A referral from an assistant such as ChatGPT can be: analytics now reports those under their own AI Assistant channel, and ours shows four such sessions between June and early September. What neither will ever record is the visit that never happened because the answer was sufficient. So the top of the funnel is becoming partly unobservable, and a low traffic number can mean two opposite things: nobody is reading you, or people are reading about you somewhere you cannot see.

Our own site had 19 clicks in its first three weeks of data. Read as acquisition, that is a poor number. Read as verification, it is close to meaningless — because the question it answers is not how many people arrived, but whether what they found when they did was worth finding.

Who this actually affects.

It would be convenient for us to tell you every business is equally exposed. They are not, and the difference is worth judging honestly before spending anything.

A buyer commissioning a substantial technical job — where the decision is considered, the sum is meaningful and being wrong is expensive — does not choose from a summary. They ask colleagues, check a procurement list, request a proposal and meet someone. The machine layer influences the shortlist at most. If that is your market, this shift is real but slow.

A small business owner comparison-shopping a routine, repeatable service is at the other end. Lower consideration, less specialist knowledge, more willingness to accept a confident answer. That buyer may form a complete impression of you without visiting anything you built.

We are exposed on both sides, differently, and we do not treat it as one problem.

What we would do, and what we would not.

We would not buy anything sold specifically as optimisation for AI answers. Google states in its own guidance that no special file, markup or format is required to appear, and that files created for that purpose neither help nor harm. We have set out what does and does not matter separately.

We would write down the answers. Go through the questions a buyer asks in the week before choosing — price, scope, what is excluded, how long, what happens if it goes wrong — and put the actual answers on the page rather than deferring them to a call. That is the same work that makes a site convincing to a person, which is why it is worth doing whether this shift continues or stalls.

And we would put real weight into the parts of a pipeline that do not depend on being found at all: the people who already know you, the customers who can refer you, the places your buyers already gather. If discovery is becoming less controllable and less measurable, the rational response is not to spend more on discovery. It is to depend on it less.

What we cannot tell you.

We cannot quantify what this is costing you. There are credible published estimates for the web at large, and we have been through them separately, but every one of them is an estimate drawn from a panel, none of them is Malaysian, and the click that did not happen is recorded nowhere. We cannot tell you the trend continues at this pace; the answers are two years old as a mainstream feature and the behaviour around them is still forming.

What we are reasonably confident of is narrower. The click is becoming optional. What survives being summarised is what you have stated plainly. And a website is now more often read by someone deciding whether to trust you than by someone discovering you exist — which means the most valuable thing on it is no longer the thing that attracts attention, but the thing that survives scrutiny.