GenAI is already shaping product decisions, even when it does not get the click. And retailer AI assistants like Amazon’s Rufus are starting to influence how people convert, not necessarily how they discover.
If you work in eCommerce, you’ve probably heard two equally unhelpful takes in the past year:
“ChatGPT is going to replace Google.”
“AI shopping is hype, it’s not driving real sales.”
The latest Similarweb webinar lands somewhere far more useful. The speakers (Ben Parks and Marta Sukvich) did a good job of separating what’s actually happening from what we all assume is happening.
Webinar: Similarweb, From Search to Shop: How ChatGPT and Rufus Are Rewriting eCommerce
[Watch it here](https://www.youtube.com/watch?v=3IZhaWayQgE)
Here’s what stood out, what it means for brands, and what we at Lmo7 think you should do next.
**1) The click is not the journey anymore**
One of the strongest points in the webinar was how misleading last-click thinking can be in an AI-influenced world.
Similarweb showed that if you only measure referral traffic from ChatGPT to retailers, you see a small number. But when you look at journeys (people who use ChatGPT, then leave, then go to Google or directly to a retailer later), the impact can be multiple times bigger than referral clicks suggest.
That matches real behaviour.
People use ChatGPT like a shopping companion, then they:
open a retailer app
search the product name
compare options
and buy later
So if your reporting is only counting “ChatGPT sent X visits”, you will undercount AI’s influence and you will make the wrong decisions.
What to do: start treating AI as a consideration layer, not just a traffic source. Your measurement needs to include assisted journeys, not only direct clicks.
**2) Walmart’s ChatGPT growth is real, but context matters**
The webinar highlighted that ChatGPT referral traffic to Walmart has grown, and that this traffic can convert well.
But they also added an important reality check: even if ChatGPT becomes a meaningful share of Walmart’s referral traffic, it can still be a tiny portion of Walmart’s total traffic.
That matters because it keeps teams grounded.
The right conclusion is not “drop everything for ChatGPT”.
The right conclusion is: AI-driven behaviour is growing and it is already shifting category dynamics, but it’s early.
What to do: watch where AI influence is rising by category and intent, then prioritise the areas where it can actually move commercial outcomes.
**3) Purchase intent in ChatGPT is smaller than Google, but still commercially meaningful**
Similarweb shared a key comparison:
Google still dwarfs ChatGPT in top-of-funnel shopping behaviour
A smaller percentage of ChatGPT prompts show purchase intent versus Google
But then the important part: even a “small” percentage of ChatGPT’s scale can still translate into a meaningful number of high-intent shopping moments, particularly in complex categories.
And those complex categories were a theme throughout the session: electronics, laptops, tools and home improvement.
What to do: if you sell products where shoppers need reassurance, compatibility checks, comparisons, or “best for my situation” answers, you should treat AI visibility as a core part of your digital shelf.
**4) A better way to think about Rufus: conversion agent, not discovery engine**
Amazon’s Rufus came up a lot, and the framing was useful.
Similarweb’s read is that Rufus is currently better understood as a conversion agent rather than a discovery engine. People use it to narrow choices, confirm decisions, and reduce doubt.
That means Rufus is less about “show me new things” and more about:
“which one should I pick?”
“summarise reviews”
“is this compatible with X?”
“is this good for my use case?”
They also noted that the overall volume of Rufus-driven journeys is still relatively small compared with Amazon’s full site traffic, but it is growing.
What to do: optimise for decision support. Rufus rewards clarity, context, and content that helps someone say “yes”.
**5) The surprise: on-site search is still growing**
With ChatGPT on the outside and Rufus and Sparky on the inside, you might assume the search bar is dying.
The webinar showed the opposite. On-site search is up across major retailers year to date (Walmart, Target, Amazon).
The lesson is straightforward:
AI assistants are not replacing search yet. They are changing what happens around search.
So the fundamentals still matter:
keyword coverage
titles and bullets that match how people search
retail media alignment with organic shelf performance
What to do: keep retail search as a core growth lever. Build AI optimisation alongside it, not instead of it.
**6) The most practical insight: conversational content wins**
The most actionable example in the webinar was around Sketchers and Rufus.
Similarweb’s view was that Rufus appears to respond well to conversational, contextual content, particularly in reviews. The example reviews were packed with use cases and constraints, like:
“I’m a parcel driver, I walk 11,000 steps a day”
“extra wide feet”
“cold weather and heavy rain”
“used for 300km of walking”
That is basically how people speak to AI.
They don’t say: “waterproof shoes”.
They say: “I need something waterproof for long shifts, wide fit, comfort matters, what should I buy?”
What to do: rewrite your digital shelf content to answer those prompts. Add context that mirrors how customers explain their needs.
The new KPI stack: you are not replacing metrics, you are adding them
Marta put it well: brands won’t swap old KPIs for new ones, they will track more.
The new layer includes:
- AI visibility by topic (use cases, ingredients, concerns, categories)
- sentiment (how AI describes you)
- sources that AI seems to rely on (reviews, retailer PDPs, brand sites, third-party content)
- retailer SKU-level traffic and share
- on-site search keyword share and click share
At Lmo7, we see this as the expansion of the digital shelf into an AI shelf. If you’re not measuring it, you’re guessing.
What we recommend doing next (simple and realistic)
If you want practical next steps without creating a brand new department, start here:
-Build a prompt library
List 25 real customer-style prompts that describe needs, constraints, budget, and context. Check whether your brand appears and how.
-Upgrade PDPs for “decision language”
Add “best for…” use cases, compatibility notes, comparison-style FAQs, and plain English outcomes, not just features.
-Treat reviews as performance content
Encourage detailed reviews that mention use cases and constraints. This helps humans and helps AI assistants interpret value.
-Keep retail search tight
Improve keyword coverage and align retail media with the terms that matter most for conversion, especially around peak trading periods.
**Final thought**
The Similarweb webinar didn’t argue that AI is replacing retail fundamentals. It argued something more useful:
Shopping journeys are being re-routed.
AI is shaping what people consider. Retailer assistants are helping them decide. Search is still the backbone.
The brands that win in 2026 will be the ones that become the most confidently recommendable option for specific customer contexts, and then make the purchase easy wherever the shopper lands.
If you want help turning this into a measurement and optimisation system, that’s the kind of work we do at Lmo7. Not hype, not guesswork, just a clearer shelf across AI, retail, content, and media.