A client asked me last month whether the website was still worth the budget. Traffic was down. Conversions were flat. The line item looked expensive.
It was a fair question, asked with the wrong metric attached to it.
The website is not underperforming. It changed jobs, and nobody updated the job description.
The site is now an input, not an endpoint
Rand Fishkin made a version of this argument on the SparkToro blog recently, and the framing is worth taking seriously because he has spent two decades measuring exactly this shift.
His point is that a website remains the permanent, owned, centralized home for what an organization actually claims about itself. Everything else you publish sits on rented ground.
Then comes the part that reframes the budget conversation. Gemini, Claude, ChatGPT and Perplexity build their answers about your business from what they can read. If your own site does not state something clearly and currently, the model fills the gap with whatever else it found.
That is not a traffic function. That is a source-of-record function, and it runs whether anyone visits or not.
The cold data version: your website is now read far more often than it is browsed.
Fewer visits, wider reach
The zero-click numbers have been going one direction for years, and most teams responded by treating the decline as a content quality problem. Publish more, publish better, win the traffic back.
Some of that worked. Most of it treated a structural change as a performance dip.
Reach did not collapse. It decoupled from visits, which is the shift underneath survival in the zero-click era. Your material now reaches people through AI answers and search summaries that never register as a session, which means influence and traffic have quietly become two different measurements.
I have seen this play out with brands that were, on paper, losing. Organic sessions down by a third, inbound demo requests up, and sales calls opening with buyers who already knew the pricing model. Somebody had read the site. Nobody had visited it.
If your only instrument is Google Analytics, that quarter reads as a failure. It was one of their best.
What this changes on Monday morning
Three things follow from treating the site as a source rather than a destination, and all three are unglamorous.
First, currency beats volume. An outdated pricing page is not a stale asset sitting quietly in a corner. It is an actively wrong claim being repeated by four different models to people who will never see your correction.
Second, plain statements beat clever design. If a fact lives inside an image, a carousel, or an animation, it does not exist for the reader that matters most now. I have walked audit calls where every commercially important number on the homepage was rendered as a graphic. Beautiful page. Blank to a machine. This is the unglamorous work of making your site legible to machines.
Third, the conversion layer still has to work like a shop. This is the part people get wrong when they over-rotate on the zero-click story. The purchase still happens on your property. Checkout, pricing, onboarding, the demo form, the entire commercial mechanism is still yours to run, and it is still where brand experience is either delivered or quietly lost.
There is a fourth item that belongs on the list and rarely makes it. Dates. A claim with no date attached reads as unverified to a system deciding which of four contradictory statements about your company to repeat.
Put a visible last-updated stamp on every commercially important page and actually maintain it. It costs nothing and it is one of the few signals you fully control.
So the site does two jobs now. It feeds the machines, and it closes the humans. Those require different work, and most teams have staffed for neither.
The differentiation problem nobody wants to hear
There is a second-order effect worth sitting with.
If models synthesise the standard answer about your category, then publishing the standard answer buys you nothing. You become a source for a summary that mentions somebody else.
SparkToro’s own response to this was to rebuild their site around being noticeably human. Bright, odd, specific, with a voice no summarizer can reproduce without quoting it.
That is not a design preference. It is a visibility architecture. Generic material gets absorbed into the answer. Specific material with a point of view gets attributed, because attribution is how a model handles a claim it cannot generalize.
The practical test is simple. Take any page on your site and ask whether a model could produce the same content without you. If yes, that page is training data. If no, that page is a citation.
Most corporate websites are almost entirely training data. That is the real reason the budget question keeps coming up.
The website did not stop working. It stopped being a place, and started being a record.