28.5% of MacBook buyers at major retailers had a relevant AI chat within 30 minutes of purchasing, yet analytics attributed just 2% of those buyers to AI referrals, a 14x gap between what actually influenced the purchase and what your dashboards can see. Similarweb ran the data across Q1 2026, Modern Retail wrote it up, and most brands moved on, when they shouldn't have.

Shoppers are no longer moving through a list of options the way they used to. Increasingly, they get a recommendation based on what they are trying to solve, then make a decision from there. If decisions are being shaped upstream, growth is no longer just about capturing demand; it is about influencing how that demand is resolved before the click ever happens.

The buyer journey doesn't start where the transaction ends

44% of US online buyers start product research in an AI tool, or split it between AI and traditional search. That's Bain, 1,500 shoppers, September 2025, which puts this well past any early-adopter niche and into mainstream behavior across the buying population.

The shopper asks ChatGPT what to buy, ChatGPT names a product, and the shopper opens Amazon and types the brand name into the search bar. Amazon logs a branded search and books the sale, but the AI chat that named the product is something Amazon can't see, your analytics can't see, and your marketing mix model has no line for.

54% of ChatGPT referrals to retail apps land on Amazon, 86% of shoppers double-check AI recommendations before buying with 68% of them on Google, and 68% of US Google searches now end without a click. Read that sequence again: the consumer researches in AI, validates in Google, and buys on Amazon, three surfaces feeding one decision, with nobody tracking the first two.

Optimizing for Amazon AI is optimizing for the last step

I'm not saying skip Alexa for Shopping. It's the default assistant in Amazon's search bar for every shopper now, and its predecessor Rufus reached 300 million users in 2025, so anyone selling on Amazon has to be readable to it.

Here's the trap. Amazon blocks every major AI crawler, including OpenAI, Anthropic, Perplexity, and Google's shopping agents, so none of them can read your Amazon listing. When a shopper asks ChatGPT for the best creatine capsules, ChatGPT doesn't see your PDP; it reads your website, your third-party reviews, your Reddit mentions, and your YouTube, and then it names a product.

If it doesn't name yours, you aren't in the consideration set, no matter how well-optimized your Amazon listing is. The shortlist got decided before the shopper ever opened the app, and you can't win a price war you were never entered into.

The invisible growth channel

Adobe, May 2026: AI-referred retail visitors convert 54% better than non-AI traffic. Similarweb, June 2026: brands ChatGPT recommends see 2.5x more site visits within 7 days, and products with ChatGPT product feeds appear as the first offer 99.9% of the time.

Now look at how AI-influenced shoppers actually arrive on your site or listing: 55.9% via search, mostly branded, 19.9% direct, 13.5% referrals, and 8.8% labeled as AI referral.

That last number is the only channel your analytics attributes to AI, while every other bucket is where the AI influence actually shows up, with branded search rising, direct traffic rising, marketplace sales rising, and referral numbers flat. Inside Google Analytics, that pattern reads like nothing at all. Only 14% of marketers are tracking AI citations, which leaves the other 86% optimizing for a funnel they can no longer see.

What actually drives the recommendation

If AI can't read your Amazon listing, and search-ranked pages are declining as citation sources (38% of AI Overview citations now come from top-10 results, down from 76% a year ago), what's actually feeding the recommendation?

Ahrefs studied 75,000 brands, and the strongest signal correlated with AI visibility isn't backlinks; it's brand web mentions at 0.71 correlation, followed by YouTube. Reddit's citation share in commercial and shopping categories grew 73% in Q1 2026, and 80.9% of listicle citations go to third-party lists rather than brand-published ones.

AI decides who to recommend by reading the wider web, not your press release and not your paid search, but the conversations already happening about your product on Reddit, YouTube, third-party reviews, comparison content, and Q&A forums. If the internet isn't talking about you, AI can't recommend you.

Where the budget needs to go

Three things need to shift.

First, general-market LLM visibility is a top-of-funnel investment rather than an experimental line item. The dollars that go into being named in ChatGPT, Perplexity, and Gemini return through branded search on Amazon, direct traffic to DTC, and inbound at retailer sites. If your finance team is looking for the ROI on the AI referral line in GA, they're looking in the wrong place, since the return shows up two channels away.

Second, the discovery layer runs on the open web instead of the walled gardens, which means digital PR, Reddit and community seeding, YouTube presence, review and ratings programs, and third-party listicles all function as the input data for the AI that builds the shortlist. If a competitor has more third-party mentions, better Reddit sentiment, and stronger YouTube coverage, the AI names them, and you lose the sale before the shopper ever opened Amazon.

Third, the measurement framework has to span the whole path, covering AI citation share by prompt, branded search velocity, marketplace rank and Alexa for Shopping visibility, and community sentiment. Anyone telling you they can measure AI influence with a single attribution model is selling you the same iROAS story we already banished, because the right measurement is a system spanning the whole path rather than a single metric.

The strategic frame

Amazon is where they buy, AI is where they decide, Google is where they check the answer, and Reddit is where they compare notes with people who already bought it. Being visible in one surface and invisible in the others is a strategic error, not just a coverage gap. The brands only investing in Amazon retail media are paying to convert a shopper who was never going to arrive, because a competitor got named upstream, and the brands only investing in DTC SEO are ranking for pages nobody clicks anymore in a search environment where 68% of results end without a click.

The playbook isn't Amazon versus AI versus Google versus Reddit, it's all of them operating as one system, feeding one outcome: getting named when the shopper asks what to buy.

The 44% of shoppers starting in AI sit at the top of your funnel, and the 40.5% of ecommerce running through Amazon sits at the bottom, with the invisible middle where the decision gets made being the only piece most brands aren't funding.

Across the brands we work with, the ones gaining share are building connected visibility across AI, search, marketplaces, and the open web, so they're consistently recommended before the purchase begins rather than optimizing a single channel in isolation. That's the work we're doing now, mapping AI citation share by prompt against branded search velocity and marketplace rank so a client can see where the recommendation is actually getting made and fund the middle of the path most brands leave dark.

Be the answer everywhere they ask.