AI Search Changed Where Consumers Look — Not What Makes Marketing Work

“How do we show up in AI search? Everyone is using it — what do we need to do to be there?”

That’s the question I’m getting most often right now, and it’s the right one to be asking.

Here’s the answer: you show up in AI search the same way you show up in Google search. Strong web vitals, strong technical SEO, and content that’s actually worth showing up for in either one.

Now the logic, because that answer sounds too simple to trust.

AI search has to find your site somehow. It’s a program reading code and variables, the same way Google finds your site. If Google shows your site prominently, they’ve read the code and the signals and categorized your site to show to users. The main point is they’ve categorized it. There’s already a global categorization of websites and web pages from Google, so we’d be naive to think AI isn’t using it — but we don’t have to be naive about it, because every AI model has a setting that lets you check a box for web search. Logically, these things are interconnected.

The behavior backs that up. The single biggest place ChatGPT sends people is Google — 21.6% of the traffic it passes along, more than any other website gets.¹ And Google’s own instructions to website owners say that optimizing for its AI features is optimizing for search. It’s still SEO.²

So the question was never SEO versus GEO — generative engine optimization, the new label for getting yourself into AI answers. The question is whether your site is clear enough to get categorized correctly in the first place.

That’s the part that hasn’t changed. Here’s the part that has.

What actually changed in how people search

Your customers genuinely are starting their research somewhere other than where they started three years ago. If you measure your marketing by traffic, that shift looks like a slow bleed. Start with the behavior, because that part isn’t debatable.

Pew Research Center tracked what 900 U.S. adults actually did across nearly 69,000 Google searches. When Google showed an AI summary at the top of the results, people clicked a normal search result on 8% of visits. When there was no AI summary, 15%. They clicked a link inside the AI summary itself on about 1% of visits. And they closed out and stopped browsing entirely on 26% of pages with an AI summary, compared with 16% of pages without one.

Read that last number again. People aren’t just clicking less — they’re finishing their search entirely, getting what they needed and closing the tab.

Zoom out and the trend agrees. Searches that end without anyone clicking anything at all reached about 68% in the U.S. in early 2026, up from roughly 58% in 2024 and around 50% in 2019.³ Notice that number was climbing for years before AI answers ever showed up in Google.

AI didn’t create this trend. Consumer behavior was already changing, probably because Google’s own result summaries kept getting better while websites optimized their content toward those summaries. And if anything, AI is solving a real problem for the consumer — it pulls several sources together, summarizes them, and then suggests what’s actually worth digging into. That’s a better experience than opening eight tabs and sorting it out yourself. The cost of that better experience just happens to land on you: those consumers never reach your site, which also means they never enter the audience you could have remarketed to later.

At the same time, AI models became a real source of traffic in their own right. ChatGPT dominates that traffic at more than 90% of what we can actually track, with Gemini a distant second and Claude still small but growing off a tiny base — roughly 1% of it, up something like 64x over eighteen months.⁴ Visits to U.S. retail sites coming from AI grew 138% year over year as of May 2026, and more than 14x since October 2024.⁵

So: a lot of growth, off a small number. For most businesses, AI still sends a sliver of your total traffic. Anyone telling you otherwise is selling something.

Which means the useful question isn’t how much traffic you lost. It’s which traffic you lost.

The traffic you’re losing and the traffic you’re gaining aren’t the same traffic

Adobe, working from more than a trillion visits to U.S. retail sites, found that people arriving from AI converted 54% better than people arriving from everything else. They spent about 53% more time on the site and looked at 23% more pages.⁵

Think about why that would be. Someone who lands on your site from ChatGPT has already had the browsing, the comparing, and the question-asking done for them. The AI handled the early research. What shows up at your door is a person who’s already been filtered, and who’s much closer to a decision than your average visitor from Google.

That reframes the whole conversation. If your traffic is down but the visits you still get are worth more, you’re measuring the wrong thing. The prize was never traffic. It was intent — and it’s always been intent, which is why the ROI math here usually looks better than the traffic chart does.

Here’s the part nobody wants to hear: the reason you aren’t getting more of those high-intent visits probably has nothing to do with AI.

Most sites simply can’t be read

Before you spend a dollar on GEO, ask a much cheaper question. Can an AI model read your site at all?

Adobe scored how much of a page’s content AI models can actually read. Cosmetics sites came in at 63%. Electronics, 56%. Sporting goods and apparel, 51% each. Grocery, 48%. Furniture and home, 47%.⁵

So on a category average, something close to half the content on these sites can’t be read by the very thing that increasingly decides what your customers see.

Now put that next to what Google itself published in 2026 in its instructions for showing up in AI features. No special code to add. No new file to upload to your server. No chopping your content into little pieces for the robots. No rewriting your pages in some special way for AI. Google names all of that as unnecessary, and says some of it can actually get you in trouble under its spam rules.²

What Google says does matter: it needs to be able to reach your pages, store them, and be allowed to show them in results. Your pages need to load fast and work properly on a phone — that’s what Core Web Vitals measures, Google’s scores for how quickly and smoothly a page loads and settles. Clean page code. JavaScript that doesn’t hide your content from search engines. And content with a real firsthand point of view instead of a rewrite of what everyone else already published.⁶

None of that is exotic work. It’s ordinary technical housekeeping that most businesses never got around to, which is why the gap between sites AI can read and sites it can’t is as wide as it is right now.

Which brings the logic back around. AI has to find sources, and finding sources is still a program crawling your website. Proper technical SEO structure, plus content relevant enough that people actually share it and link to it, gets you the same result with AI that it always got you with search engines. It all runs on pages a search engine can reach, store, and show — the same gate it has always been.²

One of our clients builds automation systems for corrugated box plants. Not a category anyone would mistake for a content play. We spent the last year rebuilding the content on their site alongside their internal team, and ChatGPT now shows up as a traffic source in their Google Analytics. Nobody optimized for ChatGPT. We made the site clearer and more genuinely useful about a narrow industrial specialty, and AI found it.

There’s a strange wrinkle here too. One analysis of millions of AI-referred visits found ChatGPT sends nearly 29% of its traffic to the search results page on the site’s own website.⁴ That’s unflattering but useful — it means the AI trusts your company and can’t work out which of your pages answers the question. So ask yourself: if AI landed on your site already knowing you were the right company, could it find the right page? That’s a site structure problem rather than an AI one — your pages aren’t giving it enough to work with about how they relate to each other.

But don’t let “it’s just SEO” make you complacent

This is where I part ways with how the industry is repeating Google’s line.

Saying “GEO is just SEO” is naive as a blanket statement. Technically, there are differences between the two. Foundationally, there are far fewer. The more useful way to look at it is that the whole environment is changing, and changing very fast. Understanding what’s changing and why is genuinely harder now because of that pace. But the foundational principles that built this landscape in the first place are still what every new layer gets built on top of.

What I’m watching change is which sources are in play. From what I’m seeing, open platforms are getting pulled into what these models read in a way they weren’t before — Reddit being the obvious example. That’s the real opening for brands: extra places to build your presence where you previously had no reason to bother. And yes, doing that also helps your SEO, which is the point. So it’s less that you need a different playbook, and more that the field you’re playing on got wider.

It also matters more than it used to, because AI isn’t always looking at the live internet. One study of real browsing data found only about a third of ChatGPT questions trigger an actual web search, down from nearly half in late 2024.¹ So a good share of what AI says about your industry comes from what it already learned when it was built, not from reading your site this morning. Which raises a question most business owners have never had to ask: what does the internet already say about you, in the places AI reads?

Technology is changing faster right now than at any point I’ve worked in this industry, and it’s changing how people shop and search more than anything since search engines were invented. This is the TikTok moment again — a new place to be seen opens up, the businesses paying attention get cheap attention there for a while, and the ones who wait until it’s an established best practice pay full price later.

Staying on top of that is genuinely a lot of work. Which is the ironic part: the most practical use of AI for most marketing teams right now isn’t writing content. It’s keeping up with how fast AI is changing where their customers look.

Google Ads, Bing Ads, and Meta Ads all got rebuilt the same way

All three major ad platforms have handed ad delivery over to AI. Google runs AI Max and Performance Max, which write your ad copy and pick your landing pages for you.⁷ Microsoft has been putting ads inside Copilot conversations since 2024, and letting other websites run the same kind of ads inside their own chat tools.⁸ Meta rebuilt the system that decides which ads even get considered — it calls it Andromeda — so it can narrow tens of millions of possible ads down to a few thousand before the auction happens.⁹

Different platforms, one requirement: these systems only perform if you feed them clean data.

So here’s the question to ask about your own accounts. Forget what you told the platform your goals are — what results have you actually fed back into it? Does it know which of those form fills turned into money?

Performance Max was rough when it launched. The return was genuinely bad and plenty of advertisers got burned. It’s improved a lot since — partly Google refining it, partly the system having more to learn from. But what separates a Performance Max campaign that scales from one that quietly wastes money isn’t a setting. It’s whether your conversion data is clean and connected: tracking what happens on your website, tracking the sales that close off it by phone or in person, and sending all of that back into Google, matched to the right type of conversion.

None of that is specific to Performance Max. The same principle drives Google’s automated bidding on regular search campaigns — the settings where you tell Google what a lead or a sale is worth and let it do the bidding. Feed it accurate results and it goes after your actual business outcome. Feed it form fills that never became customers and it will happily buy you more of those. Building those data pipelines properly is a critical step Flywheel takes on every Google Ads account we run, and it’s the biggest single reason our clients see the results they do.

This is also coming whether you opt into it or not. Google is upgrading Dynamic Search Ads — the older campaign type where Google writes your headlines off your website — into AI Max. Voluntarily now, automatically starting September 2026, after which you won’t be able to create a new one at all. Google says AI Max campaigns average 7% more conversions or conversion value at a similar cost per conversion, though that’s measured against using only part of the feature set, not against having AI Max switched off entirely.¹⁰ Read the fine print before you quote that number to your board.

The same logic runs Meta Ads. When the system is choosing from tens of millions of possible ads, your job is giving it enough creative variety and enough accurate data about what actually converted for it to have something worth picking. Meta reports gains from its own automation and AI creative tools, but those are company-wide internal numbers — treat them as a direction, not as a prediction for your account.⁹

Bing Ads gets a more measured answer from me than most agencies will give you. Microsoft Advertising is a much smaller share of total searches, but depending on your industry and how much you’re spending, it delivers a strong return. Think about who’s actually using Bing, and you can work out which industries it serves well.

Still, it trails Google on capability and on sheer volume of searches, so we put more budget into Google. Not because Bing is hard to manage — the AI tools we’ve built for reviewing ad accounts cover Microsoft Ads about as well as they cover Google, with close to round-the-clock coverage either way. The real reason is return. Consolidating spend into a channel where the return is proven also concentrates the conversion data feeding that channel’s bidding, which makes the whole thing compound. Spreading into Bing makes sense at scale, or in specific cases where the customer intent is genuinely there — not as a reflexive hedge.

Worth noting that Microsoft’s own numbers point the same direction on AI search adding to search rather than replacing it: back in 2024 it reported better click-through rates overall when someone used both Copilot and regular search on the way to a decision.⁸

The attribution crisis is mostly a tracking problem

You’ll read a lot right now about AI destroying attribution — meaning you can no longer tell which marketing actually produced a customer. Traffic down, leads unexplained, nobody knows what’s working.

If your traffic is down and your leads are flat, which number is lying to you? Usually neither. Usually the two were never properly connected in the first place.

We’re not seeing the crisis. Traffic moves around, and lead volume still makes sense, because the tracking was built properly from the start — clean data flowing into Google Analytics and Google Tag Manager, into the ad platforms, and the closed-sale data going back in so we can see what actually turned into revenue.

I’m not claiming this is easy. I’m claiming that most of what gets blamed on AI search is a measurement gap that already existed and is now just impossible to ignore. If you can’t explain where your leads came from, AI isn’t your problem.

And then there are Google reviews

When someone finally does find you — what do they see before they ever reach your website?

I’ll be straight about this one, because the tidy version of this article would tell you reviews drive your AI visibility. We are not seeing Google reviews show up prominently in AI answers.

The importance of Google reviews still cannot be overstated. Maps is where people go when they’re looking for something local, and across the accounts we manage, reviews are what moves the needle most in both directions — whether you show up in Maps at all, and whether anyone gives you a second look once you do. Google’s own instructions on local ranking say how well-known you are is shaped partly by how many reviews and what ratings you have, that replying to reviews helps you stand out, and that there is no way to request or pay for a better local ranking.¹¹

That last point is the whole reason reviews belong in a conversation about AI. Nearly every other input in your marketing can be bought, automated, or generated. Earned trust can’t be. Which makes it the one piece of the foundation that gets more valuable as everything else gets easier to fake.

What this actually means for your marketing

Nothing on this list is new. That’s the point.

Fix the technical foundation. If half your content can’t be read, no amount of GEO strategy makes up for it. Can search engines reach your pages and store them, is your site organized sensibly, do your pages link to each other in a way that makes sense, and do they load fast. This is the same work that made you visible in 2015, and it’s now the price of admission to an AI answer.

Make the site work after the click. Visitors who come from AI arrive closer to a decision than anyone else you get. Losing them to a slow page or a confusing path to contacting you is a much more expensive mistake than it used to be — which is why a conversion audit tends to pay for itself faster than another push for more traffic.

Keep earning reviews. Not for the AI. For Maps, and for the humans reading them before they call you.

Get your conversion data right. It’s the limiting factor on every AI-driven ad platform you’re buying. Google Ads, Meta Ads, Microsoft Ads — all of them chase whatever you tell them success looks like. Tell them accurately.

Stay curious about the new places to show up. Not because the fundamentals changed, but because where you apply them is expanding faster than it ever has.

So go back to the original question — how do we show up in AI search? You show up the same way you’ve always shown up: a site that can be read, content worth reading, and a business people vouch for. AI search moved where your customers look. It didn’t change what earns their business once they’ve looked. Businesses treating this as the reason to finally fix their fundamentals will do fine — you can see how that plays out across our client work. Businesses hoping to skip the fundamentals with a clever new acronym will find out that AI is, unhelpfully for them, reading the same website everyone else is.

If you want a straight read on where your own foundation actually stands, that’s a conversation worth having.

Sources

1. Semrush analysis of over one billion lines of U.S. browsing data from a 200-million-user panel, October 2024 through February 2026, reported by Danny Goodwin in Search Engine Land. Google accounts for 21.6% of all traffic ChatGPT refers, the largest share of any single website, and only 34.5% of ChatGPT queries triggered a web search, down from about 46% in late 2024. ↩↩
2. Google Search Central, guide to optimizing for generative AI features on Google Search. Trade coverage documenting that this guidance contradicts a good deal of prevailing GEO advice: Matt G. Southern, Search Engine Journal, May 2026. ↩↩↩
3. Figures on searches ending without a click come from a SparkToro study using Similarweb browsing data covering U.S. Google searches January–April 2026, reported by Danny Goodwin in Search Engine Land. The study notes that year-over-year comparisons draw on different data providers and methods and should be read with that caveat. The same analysis found Google’s newer AI Mode accounted for only 0.34% of searches during the window — the shift described here is driven by AI Overviews, the AI summaries in normal results. ↩
4. Share of AI-referred traffic by model, Claude’s growth rate, and the internal-site-search finding come from a Previsible analysis of 6.77 million visits across 166 Google Analytics properties, November 2024 through May 2026, published by David Bell in Search Engine Land. ↩↩
5. Adobe Analytics data on AI-referred traffic growth, conversion and engagement lift, and how much page content AI models can read by category, based on more than one trillion visits to U.S. retail sites, as reported by Abbas Haleem in Digital Commerce 360, June 2026. Readability scores are produced by Adobe’s own AI Content Visibility Checker. ↩↩↩
6. Google Search Central, understanding Google page experience. Google’s own framing is worth being precise about: relevance comes first, and page experience contributes where plenty of comparably helpful content already exists. It’s a tiebreaker, not a trump card. ↩
7. Google Ads, “Steer performance with new AI Max features”, April 2026. Google publishes no conversion or cost figures for AI Max in this announcement, and none are implied here. ↩
8. Microsoft Advertising, “Unlocking the power and potential of generative AI for advertisers and publishers”, April 2024. The click-through figure is Microsoft’s own reported number and has not been independently verified; the post also predates Microsoft’s later Copilot ad expansions. ↩↩
9. Meta Engineering, “Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine”, December 2024. Meta’s reported gains are internal and company-wide. ↩↩
10. Google Ads, “We’re upgrading Dynamic Search Ads to AI Max”, 2026. Automatic upgrades begin September 2026, when creating new Dynamic Search Ads campaigns ends. Per Google, the 7% figure compares the full AI Max feature set against using its search term matching alone — not against a campaign with AI Max switched off. ↩
11. Google Business Profile Help, tips to improve your local ranking on Google. ↩