I launched my first Amazon FBA brand in 2017.
The ecommerce playbook was pretty clear then:
Get found in search. Win the click. Build a strong product page. Convert the customer.
I later sold that brand, co-founded ecommerce A/B testing platform ProductPinion, worked on scaling Amazon brands inside an aggregator, and more recently advised major CPG companies on AI visibility.
Across all of those chapters, we spent enormous amounts of effort optimizing what happened after the shopper started searching.
I think the more interesting battle is now moving one step earlier.
What if your next customer barely searches for your product, barely browses your listing, and still buys from you?
Because AI already helped them decide what deserved consideration.
The data changed very fast
In March 2025, Adobe found that visitors arriving at U.S. retail sites from generative AI sources converted 38% worse than non-AI traffic.
One year later, they converted 42% better.
By May 2026, AI-referred shoppers were converting 54% better and generating 53% more revenue per visit.
Adobe’s analysis covers more than 1 trillion visits to U.S. retail websites.
I don’t think the important conclusion is simply:
“AI traffic is better traffic now.”
There is a more interesting explanation.
The click may be happening later in the buying decision.
A shopper can already have asked questions, compared products, narrowed the options and resolved objections before visiting the brand.
The referral is simply where we finally observe that journey.
AI is increasingly doing the product evaluation
I like keeping the AI shopping journey simple:
Discovery → Consideration → Purchase Intent
For an omega-3 shopper, that might look like:
-
Discovery: “What is the best omega-3 for women?”
-
Consideration: “OceanBlue vs Nordic Naturals?”
-
Consideration: “Is OceanBlue third-party tested?”
-
Purchase Intent: “Which omega-3 should I buy?”
Historically, a shopper might answer those questions across Google, Amazon listings, reviews, Reddit, YouTube and comparison articles.
AI can increasingly synthesize much of that research into one conversation - on demand.
McKinsey found that among European consumers using AI for shopping:
-
63% use it to compare brands, models, prices and reviews
-
55% use it to learn about a product or category
-
46% use it for product discovery and inspiration
More importantly, McKinsey found that AI adoption is happening faster for decision influence than for actually executing the transaction.
AI doesn’t need to complete the checkout to matter.
It only needs to influence which products make the consideration set.
Amazon and Walmart show why this matters commercially
Most AI visibility conversations focus on ChatGPT, Gemini or Perplexity.
Those matter.
But I am particularly interested in what happens when the AI recommendation engine and checkout already exist inside the same marketplace.
Amazon is probably the clearest example.
Rufus is now part of Alexa for Shopping (AFS), and Amazon says more than 350 million customers used it during the previous 12 months.
In Q2 2026:
-
active users nearly doubled year over year
-
interactions increased more than 5x
-
U.S. customers using Alexa for Shopping spent over 40% more per order
Amazon previously attributed nearly $12 billion (with a B!) in incremental annualized sales to Rufus during 2025.
Those are Amazon’s own numbers, so I would treat them as first-party evidence, not independent proof.
Fortunately, independent data points in the same direction.
Sensor Tower tracked 60,000 dedicated U.S. Amazon shoppers and found that shoppers interacting with Rufus were 2.74x more likely to buy. Across web and app, Rufus sessions delivered roughly double the conversion rate of non-Rufus sessions.
Their interpretation is useful:
“Rufus doesn’t create demand; it reveals it.”
Someone asking detailed shopping questions is likely already demonstrating higher purchase intent.
So I would not claim the assistant caused the entire conversion difference.
But now Walmart is reporting a strikingly similar pattern.
On its Q2 FY2027 earnings call, Walmart CEO John Furner said the number of customers using Sparky was up 70% year over year, while customers and members who use Sparky for shopping spend 40% more per order than those who do not.
That gives us Amazon and Walmart independently reporting essentially the same commercial signal:
Shoppers engaging with AI shopping assistants are showing materially higher value and intent.
Again, correlation is not causation.
But for a brand, there is a practical implication either way:
AI is increasingly present very close to the buying decision. Already.
If Alexa for Shopping or Sparky repeatedly recommends three competitors when shoppers ask an important category question, traditional search rankings no longer tell you the whole visibility story.
The new metric brands are missing
For most of my ecommerce career, visibility was primarily about best seller or keyword ranking.
Position one.
Position three.
Page one.
AI introduces another type of shelf.
At ARTAN AI, we define the AI Shelf as:
The set of brands and products an AI system surfaces or recommends in response to commercially relevant shopper questions.
And one of the KPIs I increasingly care about is AI Share of Voice, or Recommendation Share:
AI Share of Voice = the percentage of relevant shopper questions where AI recommends your brand.
The critical word is relevant.
Running twenty random prompts and producing an “AI visibility score” doesn’t tell a brand much.
First determine which questions actually influence the buying journey in your case. We rely on Amazon data here.
Then ask:
Which commercially relevant questions does AI recommend our competitors for, but not us?
That turns AI visibility from an abstract marketing concept into something a brand can investigate, improve and measure. This is the basis of the AI Visibility Map and Share of Voice approach we use at ARTAN AI.
What can brands actually influence?
Being machine-readable matters. For both organic marketing and sponsored ads.
The AI Shopping Revolution: How Google’s Latest…
Tim Zawislack
1 year ago
AI Is Becoming the New Storefront Across Amazon… Will Haire 3 months ago
What is Agentic Shopping? And what does it mean for… NIQ Brandbank 10 months ago
Adobe found that the average U.S. retail product page had a machine-readability score of only 66% in its analysis.
But “make your website readable by LLMs” is not a complete strategy.
A product’s digital identity exists across many information surfaces.
And relying too heavily on one supposedly LLM-friendly source is risky.
AI source preferences can change fast, and differently by platform. In August, Promptwatch saw Reddit citations in ChatGPT fall 86%, while Searchable found Reddit usage rising across Google AI Overviews, AI Mode, Perplexity and Gemini.
The takeaway isn’t “Reddit is dead.”
It’s that no single-source GEO playbook is durable. Build multiple reasons to be recommended, then measure what each platform actually uses now. It will change.
I would rather use Reddit to understand customer questions, objections and comparisons, then make sure strong evidence exists across the surfaces that matter.
For Amazon-first brands, we currently organize the major brand-controlled surfaces into four practical levers:
-
Backend attributes: Can Amazon clearly understand what the product is and who it is for?
-
Listing content: Does the PDP answer the questions shoppers actually ask AFS?
-
Editorial and web content: Is there credible external evidence supporting the product and positioning that AFS currently cites?
-
Customer reviews and community Q&A: Does real customer evidence reinforce the relevant claims and use cases?
This is an ARTAN AI operating framework, not an Amazon-published ranking formula.
The principle behind it is simple:
Build multiple, independent reasons for AI to recommend the brand. Make sure key data points are consistent everywhere.
That is much harder to algorithm-update away.
What we’re seeing at ARTAN AI
In one Amazon-first brand engagement, we measured Alexa for Shopping Share of Voice moving from approximately 1% to 9% within one month while work was being done across product attributes, listing content, web content and customer evidence.
This is one observational example.
It does not prove that any individual intervention caused the increase, and I would not present it as a general benchmark.
But it illustrates why measurement matters.
Without the baseline, the brand would know content had changed.
With measurement, we could also see whether recommendation visibility changed alongside it.
That is the difference between doing AI optimization and actually evaluating whether it worked.
The framework I would use
I wouldn’t begin an AI visibility strategy by producing dozens of “GEO-optimized” articles.
I would start by identifying the shopper questions that actually matter.
Then:
Measure → Find the gap → Fix it → Re-measure.
Measure which questions your brand wins.
Find where competitors consistently beat you.
Determine what information, content or evidence may be missing.
Improve what the brand can actually control.
Then ask the same questions again.
One question gap can inform listing copy, images, backend attributes, FAQs, creator content, YouTube, expert content, PR, comparison pages and even product positioning.
That is much more useful than simply telling a brand to “do GEO.”
The bigger shift
When I started selling on Amazon in 2017, the search result was the shelf.
Win the keyword.
Win the click.
Win the PDP.
That shelf isn’t disappearing.
But another layer is forming before it.
A customer asks a question.
AI interprets the need.
It evaluates available evidence.
Some products make the personalized answer.
Others don’t.
Only then might the shopper search, click or buy.
The most interesting thing about Adobe’s conversion data isn’t simply that AI-referred shoppers now convert better.
It is what those numbers may be telling us about when the product decision is being made.
The click used to start much of the evaluation.
Increasingly, it may be the outcome of it.
Which gives ecommerce brands a new question worth measuring:
When your customer asks AI what to buy, how often does it recommend you?
Find that number.
Find where competitors beat you.
Fix the gaps.
Measure it again.
That’s where I would start.
Sources
Adobe Digital Insights, April 2026: AI traffic conversion reversal, engagement and retail machine-readability. Adobe Digital Insights report
Adobe Digital Insights, June 2026: May 2026 AI-referred retail conversion and revenue-per-visit data. Adobe May 2026 update
McKinsey, 2026: Consumer use of AI for comparison, product research and discovery. McKinsey: AI to browse but not to buy
Amazon, Q2 2026: Alexa for Shopping adoption, engagement and spend-per-order data. Amazon Q2 2026 results
Sensor Tower, April 2026: Behavioral analysis of 60,000 U.S. Amazon shoppers and Rufus usage. Sensor Tower: From Scroll to Sold
Walmart, Q2 FY2027 Earnings Call, August 20, 2026: John Furner’s comments on Sparky adoption and customer spend. Walmart Q2 earnings event
Promptwatch, August 2026: Analysis of Reddit citation share in ChatGPT Search. Promptwatch: Reddit citations are dropping in ChatGPT