A few weeks ago I shared my screen with a private equity principal and pulled up the search visibility for one of his portfolio companies.
The line fell off a cliff around last September and never came back. I said what anyone says looking at that. Well, that looks pretty terrible. He agreed.
I told him what I’d been seeing elsewhere, which is that his company is not alone in this. I’ve watched similar declines at other businesses whose marketing leaned heavily on content, and they seem to be taking it worst: the ones whose pipeline was built on top and mid-funnel articles. Companies that spent five years publishing helpful articles to catch people early, watching that machine produce a fraction of what it used to.
He looked at it for another second. Then he did something I didn’t expect. He didn’t ask me to fix the SEO. He asked me to make AI visibility foundational to the brand build we were about to start. Not a phase two. A foundation.
He got there in about ten seconds. Most companies haven’t got there after two years of watching the same graph, and that gap is worth more than it sounds.
But first, the question in the title, because it’s the one people actually ask.
Yes, and that’s the easy part
You need a website.
There’s an honest exception. If you’re a two-person service business running on Facebook Marketplace and word of mouth, and it’s working, you may genuinely not need one, and this isn’t written for you.
Everything below assumes something specific: an established business selling a considered service, where buyers compare two or three providers before they contact anyone. If that’s you, the question was never whether you need a website. It’s what the thing is for now, because that changed, and most sites are still built for the old job.
The decision moved off your website
You already know this part, so I’ll be quick.
Someone with a problem used to search, land on a few sites, and work out who could help. Discovery happened on your website. Every convention your site follows exists because of that assumption.
Now they ask an assistant. Or they see three names in an answer. Or they read a thread. Or someone in their network says a name over lunch. Or they look at a directory a model happens to trust for that industry.
By the time they type your name, they’ve often narrowed twenty firms to three. You might already be on that shortlist. You might already be off it. None of it happened on your website.
I’ve written elsewhere about how people actually find businesses now, so I won’t repeat it.
What I want to talk about is what it does to the thing you own.
They’re not discovering you. They’re confirming you.
If the shortlist gets made somewhere else, then by the time someone reaches your site they aren’t asking “who could do this?”
They’re asking “is this real?”
They arrive with a hypothesis. Something told them you might be the answer, and they’ve come to check. Do you actually do this. Have you done it for someone like them. Are you the size they expected. Is there anything here that makes them uneasy enough to go back to the other two names.
I saw a clean example of this recently. A B2B software company filled in our contact form with a brief more specific than most discovery calls produce. They knew what had broken, roughly when, which parts of their marketing had stopped working, and what they thought they needed. I don’t know where they learned it, or whether any of it came from an AI answer. What I can say is that they arrived informed, and the conversation started somewhere well past the beginning.
That’s a different visit from the one most websites were built to receive. It makes your website a credibility engine rather than a lead engine.
What the research says, briefly
The conversion signal is consistent. Adobe Analytics, across more than a trillion visits to US retail sites, found AI-referred visits converted 31% better than other traffic over the 2025 holiday season. Ahrefs, on their own site, found AI search delivering 0.5% of visitors and 12.1% of signups. Different worlds, same direction.
The engagement signal is not consistent, and anyone telling you it is hasn’t read both. Adobe found AI visitors bounced 33% less and stayed 45% longer. Ahrefs’ study of roughly 82,000 sites found the opposite, slightly. So don’t let a bounce rate decide whether your website is working.
Ahrefs offered a reading of their own numbers, and I’ll keep their hedge:
“One interpretation of this is that AI users have clearer intent. They’ve already done their research via AI, so they land with specific goals.”
That’s a proposed explanation, not a proven cause. But confidence up, returns down, and clearer intent all point the same way. More of the deciding is happening before anyone reaches you.
One thing your dashboard can’t help with either. Someone hears your name in an answer, checks a directory, reads a thread, mentions you to a colleague, and types your URL eleven days later. Your channel report says Direct. GA4’s data-driven attribution does spread credit across what it can see. The problem is everything it never saw. Keep the reports, stop treating them as the account of how people found you, and ask buyers directly.
What a confirmation surface has to prove
If someone arrives already half-decided, the page’s job is to remove doubt fast. That’s a different brief from the one most sites were built to. A discovery page attracts and explains, assuming it has to earn attention and teach someone from zero. A confirmation page assumes attention was earned elsewhere, and its only job is to not lose it.
The visitor is running a short checklist, mostly without noticing.
Do they actually do this? Not “a full range of services.” The specific thing, named the way the buyer names it. If someone came looking for a firm that rebuilds ecommerce sites on a particular platform, and your services page lists eleven capabilities in alphabetical order, you’ve made them do work they won’t do.
Have they done it for someone like me? Same industry, same size, same complexity. When this fails it fails silently. Nobody emails to say your case studies were all consumer brands and they’re industrial. They just go with the other name.
Are they the right size? Too small and they can’t handle it. Too big and I’m a rounding error. Most sites are coy about this, and the coyness itself reads badly, because the buyer assumes you’re hiding whichever answer would disqualify you.
Is the money going to be a problem? You don’t have to publish a rate card. You do have to signal a range, because otherwise they guess. People guess wrong in both directions, and the expensive version is the one where a good-fit buyer assumes you’re out of reach and never asks.
Is there anything here that makes me hesitate? A dead link. A case study from 2019. A team page with three people who’ve left. A claim that’s obviously boilerplate. Confirmation is a fragile state, and it breaks on small things, because the visitor is already looking for a reason to stop deliberating.
Almost none of that is persuasion. It’s resolution. The visitor is trying to close an open question, and every second they spend not closing it is a second they might spend going back to the other two names.
A useful exercise: take your last five sales calls and write down the questions you had to answer that your website could have answered. That list is your priority order, and it’s usually shorter and less glamorous than a redesign.
So look hard at where the money goes
If the job is confirmation, proof carries more weight than presentation.
Most website budgets are weighted toward the parts that win attention. The homepage hero. The scroll animation. The brand film. Those absorb the most time, the most opinion, and the most money, because they’re the parts everyone in the room has a view about.
But the visitor already gave you their attention. Something else earned it. They aren’t on your homepage to be impressed.
The parts doing the actual work tend to get built last and cheapest. The case study that names a real situation and what went wrong in it. The page that speaks to one industry properly. The pricing signal. The team page that reads like humans rather than job titles.
I’m not arguing against design, and I’d distrust anyone who did. Design is often what makes proof land. Real photography of real work is more convincing than a stock image of a handshake. Clear hierarchy is what lets someone find the one thing they came for in ten seconds. A well-built case study is a design problem as much as a writing problem, and a site that looks careless fails the confirmation test before anyone reads a word.
The question isn’t design versus proof. It’s whether the design is serving the proof or decorating around it. Those look similar in a mockup and behave completely differently when someone arrives with a question.
Everyone is feeling this at once
In conversations with clients and prospects over the last few months, one thing keeps repeating. I talk to companies in industries with nothing to do with each other and hear the same three sentences in the same week. Traffic is down. Impressions are down. We don’t know what to do about it.
I can’t tell you how widespread that is. I’m describing conversations, not a study. But I can tell you what I have never once heard, which is a company saying they had already worked out what to do and adjusted.
What I do hear is two responses. Cut the budget, because the numbers look bad. Or do more of what stopped working, harder. Both assume the game is the same one and they are losing it.
That is the opportunity, and I want to be plain about it rather than coy.
Your competitors are looking at the same falling graph you are and drawing the wrong conclusion from it. Some of them are cutting spend on the exact thing that would fix it. When a whole category retreats at once, the cost of showing up drops for whoever doesn’t.
And the new game is cheaper to enter than the old one. Winning on links took years and a budget. Becoming the name that comes up for a specific problem takes being genuinely good at that problem, saying so in public, and being present in the handful of places your buyers and their models actually look. That is not easy. It is achievable for a company your size in a way that outranking a competitor with ten times your domain authority never was.
This window is open because of a misreading, and misreadings get corrected. In two years the people who moved will look prescient and the people who cut will be buying their way back in at a worse price.
Your read on AI traffic may be older than the evidence
A lot of people looked at AI referrals a while back, decided there was nothing there, and moved on.
If you evaluated it in late 2024, Adobe’s retail data gave you a reason to be cautious. Revenue per visit from that traffic was sitting below everything else.
By the 2025 holiday season the same source reported AI referrals converting better than other traffic. That’s a reason to revisit your own results, particularly if the last time you looked was more than a year ago.
I’d resist the tidy version of this, which is that the channel was worthless and then became valuable. A lower revenue per visit doesn’t mean nobody good was arriving. It means the average was lower, at a point when the volume was tiny. Plenty of businesses were getting a handful of excellent customers from it the whole time and couldn’t see them in the average.
What’s actually worth taking from it is duller and more useful. The numbers underneath a decision you made about a channel can move without anybody telling you, and most people never go back.
Retail is not a considered B2B service and none of those figures transfer to your pipeline. The shape is what transfers.
What actually works now, and what still does
The technical basics still matter. AI search products don’t run on training data alone. They issue searches, retrieve live pages, and depend on those pages being indexed and readable. If your site can’t be crawled, or your pages are slow and thin, you have the same problem you had three years ago, and no amount of reputation fixes it.
What’s changed is what sits on top of that.
For twenty years the currency was the link. Who points at you, with what anchor text, from how much authority. That still counts. But a model’s sense of what exists and who is credible in a category is also shaped by plain text that carries no link at all. A trade publication naming you next to a problem, a thread where somebody asks and somebody answers, a conference listing, a directory entry, a client mentioning you in a post.
Those mentions can now contribute to whether you turn up in an answer, in a way that a link-based view of the world would not have predicted.
Which produces an odd conclusion after two decades of search getting steadily more technical: the work that pays best for most businesses is being genuinely known for something specific, in enough places, that a description of the category is incomplete without you. Not visible. Known.
That’s uncomfortable if you’ve been buying technical audits as your whole strategy. It’s good news if you’ve been building an actual reputation, because that work now compounds somewhere it didn’t used to.
None of this means stop publishing
“Your website is a confirmation surface” is not “so stop writing things.”
What changed isn’t the value of publishing. It’s what you should expect from it. Publishing used to be a traffic play. Rank, get visits, convert a few. That’s still real, and it’s smaller than it was.
Now the same writing does three jobs. It’s what a model reads when it decides who to name. It’s the proof a half-decided visitor finds when they arrive to check whether you know anything. And it’s what makes you the kind of company that comes up in the first place.
Which is why this article exists, and I’ll show my work on it.
The phrase “do websites still matter” gets, by keyword-tool estimate as of September 2026, about 20 searches a month in the United States. Its results page opens with an AI Overview. Most of the related phrasings return effectively zero clicks. So this is a long piece on a topic with almost no search demand, published into a results page that answers the question before anyone clicks, against companies with far more authority than I have.
I don’t expect it to rank. I wrote it anyway, and in six months I’ll look at three things: whether anyone turns up to a sales call already arguing with it, whether it gets cited when someone asks a model what a website is for now, and whether people I respect send it to someone else.
Judged on sessions this was a bad idea before I started. That’s rather the point.
Where I’d start
Ask a model who it recommends in your category. Not once. Five or six phrasings of the question a buyer would actually use. Write down whose names come up and, where you can see it, where the answer came from. Twenty minutes, and usually clarifying in an unpleasant way.
Take your last five sales calls and list the questions your website could have answered. That’s your priority order, and it’s usually not a redesign.
Pick the thing you want to be known for, and get specific. Not “web design.” Not “consulting.” The particular problem, for the particular kind of company. Broad positioning was survivable when discovery happened on your site and you had room to explain yourself. It’s harder when a model has one sentence to decide whether you’re relevant.
Then get that association into more places. Your own writing is one, and it’s slow. Interviews, podcasts, trade press, communities, partner sites, the directories that actually matter in your category. The mention does work whether or not it carries a link.
Ask your last five closed deals where they actually heard about you. Not the attribution field. Ask them. You’ll find a source you didn’t know existed.
Recheck anything you concluded about AI before mid-2025, especially if you concluded it didn’t matter to your business.
I won’t pretend there’s a playbook, and I’d be careful with anyone selling one. Every industry, every buyer, every query is a universe of its own, and what matters for a manufacturer with a ten-month sales cycle isn’t what matters for a clinic.
But the direction is clear enough to act on. The companies that come out of this well won’t be the ones with the most pages or the cleanest technical audit. They’ll be the ones who are genuinely known for something, in enough places that a model can’t describe the category without them, with a website that closes the loop when somebody finally comes to check.
