AI search is taking your homepage visits. So what is your website for now?

You're sat looking at a graph that says one thing while the phone says another. A lot of business owners are in that position right now, and most of them are drawing the wrong conclusion from it.

Here is the shape of it. The Pew Research Center tracked 68,879 Google searches across 900 American adults through March 2025. Where an AI summary sat at the top of the page, people clicked an ordinary search result on 8% of visits. Where there was no summary, 15%. Links inside the summary itself got clicked about 1% of the time. [1]

One caveat, because plenty of people quote that figure without it. The drop only applies to searches that actually trigger a summary, and in Pew's March 2025 data that was around 18% of them. By the autumn, Conductor was measuring Google AI Overviews on roughly 25% of searches. [2] So this isn't half your traffic gone. It's half the clicks on a growing slice of your searches, and the slice keeps growing.

Same study, far less quoted: people ended their browsing session altogether on 26% of pages carrying an AI summary, against 16% of pages without one. [1] They got their answer and went and did something else.

The customer still turns up, though. And that's the half nobody puts in the doom posts.

Why your traffic is falling and your enquiries are not

There's a tidy explanation doing the rounds, and you'll have seen the number attached to it. Semrush published research in June 2025 putting AI search visitors at 4.4 times the conversion value of ordinary organic visitors. [3]

I was going to lead this section with it. Then I read the methodology. Semrush drew the figure from 500-odd high-value topics in digital marketing and SEO, which is to say our own industry, and it has been repeated across the web ever since as though it described everybody's customers.

It might not describe yours. Amsive ran a paired test across 54 sites with real form fills and purchases behind them, six months of GA4 data, and got organic converting at 4.6% against 4.87% for LLM referrals. The p-value was 0.794, which in plain English means the gap is noise. [4] Filtering down to the higher-volume sites widened the gap and it still didn't hold up.

Split that data by business model and something more useful falls out. B2B sites did show a real difference, 2.17% against 1.16%. B2C came out level. [4]

So here's the honest version. If you sell something people research properly before they commit, the pre-qualified visitor is probably real and you should expect fewer, better enquiries. If you sell on price or impulse, don't bank on it. And whichever you are, don't take the 4.4x into a board meeting.

Keep the scale honest too. Conductor put AI referral traffic at 1.08% of sessions across 3.3 billion of them, though that came from enterprise domains, which get cited a good deal more than the rest of us do. IT companies saw 2.8%; most manufacturers will see less than the average. [2]

Which leaves you back where you started, watching your own two numbers. Traffic down with enquiries flat is a completely different problem from traffic down with enquiries down, and if you only look at sessions you can't tell which one you've got. No benchmark answers that question for you.

Joost de Valk wrote the piece I would have written

On 22 July, Joost de Valk published an essay called The future of the website . He founded Yoast, the SEO plugin running on millions of WordPress sites, and he's spent most of his working life on this exact question. If you've got six minutes, go and read it, then come back.

His argument, in one line: the website becomes the home base of your business, a place that humans and machines can both read, act on, and check.

I agree with nearly all of it. What follows is his framing plus what I see week to week from inside a Leeds agency that builds and hosts these things for a living. Where a point is his, I've said so.

What generative engine optimisation actually means

Quick definition, because the acronyms are multiplying. Generative engine optimisation, or GEO, is the work of getting your business accurately described and cited inside AI-generated answers, rather than only ranking in a list of links. Some people call it AEO. Joost makes the sensible point that the renaming is mostly cosmetic: the work is what it always was, but there's more than one surface to do it on now, and Google is no longer the only one that pays.

Your website is now source material

The old job worked like this. Someone searched, they landed on you, and your job was to win them over once they arrived. Everything you spent money on served that moment. The hero image. The video. The clever scroll animation on the services page.

That still happens. Just less.

What happens more is that a machine reads your website and tells somebody about you. They never see the hero image. They get a summary, built from whatever your site said, delivered inside a chat window to a person you'll never meet.

You are writing reference copy now. That's a different brief from the one most sites were built to.

Out-of-date content used to sit there quietly. Now it gets read out.

This is the bit I see most, and it's mine rather than Joost's.

Old, wrong content used to just sit there. Nobody visited that page, so no harm done. The opening hours you changed in 2023. The service you stopped offering. The bloke who left two years ago and still smiles out of your team page. The price list you keep meaning to sort. The accreditation that lapsed. The case study for a client who's since gone under.

All of it is live source material now, and something is going to repeat it to a customer with total confidence and no idea it's wrong.

What makes this worse than a stale page ever was is that you don't get to see it happen. A wrong price on a page you'd forgotten about used to produce an awkward phone call, and at least you learned something from the awkward phone call. Now it produces an answer in somebody's chat window, they decide you're too expensive, and you never hear from them at all. There's no bounce rate for that. Nothing shows up in Search Console. The first sign is a quiet quarter you can't explain.

I run our hosting migrations, so I get a proper look inside a lot of other people's websites before we take them on. It's the same story most times: a site nobody has touched since launch, full of things that were true once.

Nobody owns it, is the root of it. The agency that built the site finished and left. The marketing person who used to update it went part-time. So it drifts, quietly, for four years.

That didn't cost much when the only visitors were people who already knew you. It costs now.

AI can check what Google could only guess

This is Joost's sharpest point. Google spent twenty years working out whether you're trustworthy from signals: links, reviews, how long people stayed on the page. Clever guesswork at enormous scale, but guesswork all the same. It's what E-E-A-T was always reaching for.

An AI can go and look instead. It reads your website, your Google Business Profile, your LinkedIn, Companies House and your reviews, then checks whether the story matches. Do the people on your about page exist? Is the address the same everywhere? Did you really do the thing you say you did in 2019? Line up and you look real. Don't and you look shifty, and nobody will ever tell you which one happened.

Which is why your about page might now be the best page on your site, as Joost puts it. Anyone can produce a good-looking website in an evening. Our CEO Rory built his own speaking site in about a week in January, no code, a couple of hours a night after the kids went down, and wrote up what he learned . Looking good has stopped being worth much. Being real can't be generated: actual people with actual faces, a story that holds up against every other trace of you online, customers who'll vouch for you by name.

One catch, and it's the bit with a budget attached. Your own website is the only place you fully control, which is exactly why a machine won't take its word for it. The cross-referencing has to find you somewhere else. Press coverage. Directory listings that match. Reviews with detail in them. Other people's sites mentioning yours by name. Call it link building, call it digital PR , call it what you like. It's the same job it's been for fifteen years and it just got a second reason to exist.

Your website is turning into a set of interfaces

Here's the part of Joost's piece that got the least attention and will matter the most, so I'll give it its own section.

A website used to be HTML, and sometimes an API bolted on the side. His argument is that if your website matters to your business, it now needs machine-readable surfaces as well: an MCP server at minimum, so an agent can interact with you directly rather than reading a page and guessing. Check availability. Get a price. Book the slot. Place the order.

Follow that through and the conversion stops happening on your website. It happens in somebody's chat window, through an agent acting for them, and your site's job is to be the thing that agent can talk to reliably. Thin, stale or machine-hostile sites simply don't get picked.

Nobody should rush at this. Joost is blunt that the answer isn't a second website for the robots, because we've done that before with m-dot sites and AMP and parallel versions always drift out of sync and die. One source of truth, rendered more than one way. That's an architecture decision, and it's a good reason to think carefully before your next rebuild .

Own the long game, rent the short one

Being the source an AI answer is built from is worth a great deal. It's also slow to earn, and you don't control the timing.

Meanwhile the person who wants what you sell is out there today, ready to buy, and one route to reaching them has narrowed.

The answer to that is not simply to spray money at Google. Work out which half of your visibility you own and which half you rent, then be deliberate about both. Paid search didn't get less useful when AI arrived. For plenty of businesses it's the one slot they can reliably occupy while everything else moves around underneath them.

Own the long game. Rent the short one. Most people are doing one of the two and calling it a strategy.

Not every business needs the same website

Joost makes this point about tradespeople where he lives in the Netherlands, and it travels.

If you're a good electrician in Leeds with a full diary and a decent reputation, your website isn't the thing keeping you busy. It needs to be accurate, quick, and convincing enough to close the people who've already heard your name. That's a modest job and it should cost a modest amount. Could a better site make you busier? Probably. Do you want that? Plenty of people don't, and that's a perfectly sensible answer.

If you're a brand, it's the other way round. There are only so many high streets, and your buyers live everywhere. Most of them will decide how they feel about you without ever speaking to one of your people. Your website sells more than any shop you could open, and it deserves the budget that implies.

Rory made a version of this argument in January about the market splitting, and from where I sit the split is getting cleaner rather than blurrier. Be honest about which side you're on, because it changes the size of the job. Plenty of people buy a website far bigger than their business needs. A few buy one that's nowhere near big enough and then wonder why nothing happens.

Should you rebuild your website in 2026?

More often than people expect. Rarely for the reason that prompts the call.

Most rebuilds get commissioned because the site looks tired. That's the weakest reason going now, because good-looking is cheap and everyone has it. Reskin a site with the same thin content and the same muddled structure and you've bought a new coat of paint on a building with no foundations.

Worth adding Joost's nuance here: technical correctness is getting cheaper too. Point a competent agent at a spec and a lot of what used to need a specialist gets done. He thinks there's still real advantage in being one of the few who bother, and I'd agree, but the bar is moving and the durable value sits in the questions a machine can't answer for you. Does a website even matter to your business? Have you got anything real to say on it?

The rebuilds that pay back fix what's underneath:

• Structure a person and a machine can both follow without guessing
• Real depth on every service you sell, in plain words, on pages that exist rather than in a PDF people have to ask for
• Named people with faces, and a story that matches every other platform you appear on
• Content that answers the questions your customers actually ask, rather than the ones you find flattering
• Somebody maintaining the thing, so it's still true in eighteen months

That's a bigger job than a reskin. It's also the one worth paying for, and it's the difference between a website that looks nice and a website that gets you recommended.

A test you can run in five minutes

Open ChatGPT. Ask it what your business does. Ask who runs it, what you charge, where you're based. Do the same in Gemini and Claude, because they don't all read the same things. Then put the three answers next to your own website.

If it's wrong, that's not the AI being stupid. That's your source material, and it's telling you exactly where to start.

Your website used to be the shop window. It's now the filing cabinet that everything else reads from. Less glamorous, considerably more important, and worth knowing what's in yours.

Want to know what AI is saying about you?

Our free digital marketing audit covers exactly this: what your site is telling machines about you, where the story stops matching, and which fixes are worth doing first. We'll look at the technical side and the AI search side together, because separating them stopped making sense a while ago. If you'd rather just talk it through, have a word with us .

Book your free digital marketing audit

Sources

1. Chapekis, A. and Lieb, A. (2025) Google users are less likely to click on links when an AI summary appears in the results . Washington DC: Pew Research Center, 22 July. Based on the browsing data of 900 US adults across 68,879 Google searches in March 2025, of which 12,593 returned an AI summary.
2. Conductor (2025) The 2026 AEO / GEO Benchmarks Report . 13 November. Based on 13,770 enterprise domains and 3.3 billion sessions, May to September 2025. The 1.08% referral figure derives from the enterprise client subset; the AI Overview prevalence figure from 21.9 million searches, September to October 2025.
3. Semrush (2025) We Studied the Impact of AI Search on SEO Traffic. Here's What We Learned. 9 June. Drawn from 500+ high-value digital marketing and SEO topics, which limits how far the 4.4x figure travels to other sectors.
4. Amsive (2025) Does LLM Traffic Convert Better Than Organic? A New Data-Backed Study . Six months of GA4 data across 54 sites with validated macro conversions. Organic 4.6% against LLM 4.87%, p = 0.794. Thresholded to 33 higher-volume sites, p = 0.376. B2B 2.17% against 1.16%.
5. de Valk, J. (2026) The future of the website . Published 22 July 2026, updated 28 July 2026.
6. Mason, R. (2026) What I learned from vibe coding my own website in 2026. 21 Degrees Digital, 12 January.