Similarweb found users are two to four times more likely to visit an AI-recommended brand than a competitor that was not recommended. Here is why that number turns AI visibility from a vanity metric into a commercial channel, and what a consumer brand should actually do about it.
AI Doesn’t Just Mention You, It Sends Buyers: Recommended Brands Get 2 to 4 Times the Visits
By Stephen Honight, Founder of Lmo7
Most people I speak to have quietly accepted a version of the same worry. AI search is interesting, the brand shows up in a few answers, but nobody can point to a sale. So it sits in the “watch this space” pile while the real budget goes to Amazon ads and Google.
There is now a number that makes that pile harder to justify.
Similarweb, in its 2026 Generative AI Landscape report, found that users are two to four times more likely to visit an AI-recommended brand than a competitor that was not recommended. That is the finding that moves AI visibility from a vanity metric to something closer to a channel. Not “we got mentioned”. Two to four times the downstream visits when the model puts you on the list and leaves the other brand off it.
I want to walk through what the number actually says, where it holds and where it does not, and what my recommendation would be for a consumer brand trying to decide whether any of this is worth funding yet.
What Similarweb actually measured
Start with the honest scope, because it matters. This is Similarweb’s estimate, based on their digital intelligence panel, US Desktop, over July to December 2025. Similarweb are clear that their data are estimations and extrapolations, not a census. So treat it as a strong directional signal from a large panel, not a guaranteed outcome for your brand.
What they did is compare brand pairs. When AI recommended American Express, Amex saw far more downstream visits than Capital One, and when it recommended Capital One the pattern flipped. Same shape in travel with Skyscanner against Kayak, and in beauty with Sephora against Ulta. Those are Similarweb’s category illustrations, not Lmo7 clients, so read them as examples of the mechanism rather than a league table.
Across those pairs, the recommended brand pulled two to four times the visits of the one that missed the cut.
The framing Similarweb use is the useful bit. AI is unlikely to decide every purchase. What it is doing is shaping the shortlist. The model narrows a messy category down to two or three names, the shopper carries those names forward, and the brand that made the shortlist wins the visit that follows. Rand Fishkin, quoted in the report, makes the point that this is the same kind of lift billboards, TV and radio were measured on last century. The job is not to prove AI can send traffic. It is to change how you measure and attribute it.
That reframe is worth sitting with, because it is where most brands get stuck.
Why the mention-to-visit gap is the whole story
Here is the mechanism, before any recommendation.
A shopper asks a model a real buying question. Which travel card is best for someone who flies a few times a year. What is a good beauty retailer for sensitive skin. The model does not return ten blue links. It returns a short, confident answer with a handful of named brands. The shopper does not click a source in that moment. They read the names, form a shortlist in their head, and move on.
The visit comes later. It comes through a branded search, a direct type-in, an app open, sometimes days after the prompt. By the time it lands in your analytics it looks like brand demand or direct traffic, not AI. The AI influence has already happened and left no obvious fingerprint.
So the two to four times lift is not really about traffic from the chat window. It is about influence over the shortlist that shapes everything downstream. If your brand is the one the model names, you inherit the visits, the branded searches and the direct sessions that follow. If your competitor is named and you are not, they inherit yours.
That is why “we can’t see any sales from ChatGPT” is usually the wrong test. You are measuring the click when the value is in the recommendation.
Optimise deep pages for citations, the homepage for conversion
The same report has a second finding that changes how you should actually do the work, and I think it is the most practical thing in there.
Similarweb split ChatGPT behaviour into two different things. Where AI pulls its evidence from, and where users actually land.
On the citation side, the pages models use as evidence sit deep. Around 65% of the URLs ChatGPT cited were two to three folders deep on a site. Deep guides, comparison pages, detailed product and category content. Not the homepage.
On the traffic side it is the reverse. Nearly 59% of ChatGPT referral traffic landed on the homepage. After ChatGPT’s May 2026 search update, the share of referrals hitting the homepage more than doubled, from around 25% to around 60%. So the model increasingly sends people to the front door of the brand, not to a specific page.
Aleyda Solis, quoted in the report, draws the obvious conclusion and it is the right one. Cited pages and traffic pages do different jobs. The deep pages are what feed the model’s answer, so they need to be clear, specific and genuinely useful. The homepage is where the visit lands, so it has to work as a conversion-ready next step for someone who arrives already half-sold.
My sense is most brands are doing neither well. The deep content is thin or missing, so the model has nothing strong to cite. And the homepage is a brand statement rather than a place to convert a warm arrival. Both need attention, and they need different attention.
There is no universal AI strategy, and the data proves it
One more layer from the report worth naming, because it kills a lazy assumption.
The mix of sources a model cites depends heavily on the category. In Similarweb’s ChatGPT data, beauty questions cited retail and e-commerce sites more than half the time, around 55%. Travel questions leaned on reviews and user-generated content, again over half. Finance questions leaned on specialist finance publishers, around 37%, with news close behind.
So the place you need to show up is not the same across categories. A beauty brand earns its way into answers largely through retail and e-commerce presence. A travel brand lives or dies on reviews and forums. A finance brand needs the specialist publishers. Running one generic “build some mentions” playbook across all of them wastes money in the wrong places.
This is a point we make a lot at Lmo7 and it is good to see it in someone else’s dataset. Different engines and different categories weight different sources. Citation strategy has to be specific to where your buyers’ questions actually get answered, not a single checklist.
Where I would be careful
I would not oversell this. A couple of caveats I would put on the table in any client conversation.
The two to four times figure is US Desktop and a specific six-month window. UK behaviour will not be identical, and mobile and in-app AI are not in this particular cut. The direction is very likely to hold, the exact multiple for your brand and market is unknown until you measure it.
And the deeper truth underneath all of this has not changed. Getting recommended is not a content trick. It runs into domain authority, which is still the hardest constraint on whether models cite and recommend you at all. On our own site we score well on AI readability and are close to invisible because our authority is low. You can write the cleanest comparison page in the category and still not get picked up if the wider web does not treat you as a credible source. So I would frame this as two tracks, same as always. Quick wins on content and structure that you can ship in weeks. And the long game of authority, third-party presence and reviews that takes months and sits partly outside any content project.
Anyone who tells you the content track alone gets you the two to four times lift is selling you the easy half.
What this looks like when it works
I will keep the proof honest and only use our own client results.
With Trip Drinks we ran a 60-day AI search engagement. Average position across the major models moved from 7th to 3rd, and AI referral traffic rose 33% over the period. That is the shortlist mechanism in practice. Move up the answer, earn more of the visits that follow.
With Haleon we ran Share-of-Model analysis across Voltarol, Sensodyne and Centrum on ChatGPT and Gemini. Voltarol came back with a 100% mention rate and the number one average position in its category. When the model is asked about topical pain relief, that brand is on the shortlist every time. On the Similarweb logic, that position is doing real commercial work, not just sitting in a report.
Neither of those is a promise of a specific multiple for your brand. They are examples of the same lever the Similarweb data describes, pulled on real client work.
So what should you do next
Here is where I would start, in order, depending on where you are.
If you have no read on your AI visibility yet, get the baseline first. You cannot manage a channel you cannot see, and right now most brands genuinely do not know whether models recommend them or a competitor. Our DaaS tier gives you direct data access across Amazon, Meta, Google and Shopify plus the upskilling to read it, from £250 plus VAT a month. That is the low-friction way to find out where you stand before you spend anything on fixing it.
If you already know you are being left off the shortlist and you want it fixed, that is Challenger. The Agentic Stack covers tracking and content optimisation for AI search visibility, and the content module builds the deep, citable pages the models actually pull from while making the homepage convert the arrivals. This is the two-track work run properly, quick wins on content and structure alongside the slower authority build. We run this hands-on so your team does not have to.
If you are a larger business with several brands and stakeholders who need to align before anyone acts, that is Enterprise. A workshop plus multi-brand Share-of-Model tracking shows you, brand by brand, who owns the shortlist in each category and where the gaps are, so the commercial case is on the table before the budget conversation starts.
The one thing I would not do is keep treating AI visibility as a curiosity while the shortlist quietly gets decided without you. The number that changes the argument is now on the record. Recommended brands get two to four times the visits. My recommendation would be to find out which side of that you are on.
Stephen Honight is the Founder of Lmo7, an AI-native agency helping consumer brands win in AI-powered discovery and agentic commerce across Amazon, ChatGPT, Gemini, Google AI and Alexa for Shopping. Lmo7 works with brands including Trip Drinks, Veloforte, Brown-Forman, Haleon, Pelotan and Symprove.