Jehan Hamedi, Founder of Vizit, created Visual Intelligence, AI that predicts how people respond to visual content.
A few months ago at the Digital Shelf Summit, between the third and fourth session on agentic commerce, I got talking with the guy who runs analytics for one of the world’s largest chocolate makers. I asked him what his ChatGPT traffic actually looked like. He laughed before giving me the number. There was a rounding error next to every other channel.
That gap between what the industry says is happening and what shows up in the dashboard is creating some very expensive decisions. The business case gets underwritten on a story, not the numbers.
You may not have a line item called “agentic commerce,” but you have an AEO vendor, an LLM visibility dashboard and somebody whose objectives mention it. I think I know why. This was never about agents. It’s about media.
Reaching high-intent shoppers isn’t getting cheaper. A 2025 eMarketer report projected U.S. retail media spend to reach around $70 billion in 2026, with nearly 90% of the incremental spending going to Amazon and Walmart. The rent for your digital shelf goes up every year, and two landlords collect almost all of it.
I get it. When somebody tells you AI discovery is free organic traffic, your eyes light up. Maybe you can stop funding so many sponsored placements. Then fear sets in. If discovery really is moving and you’re not there, you’ll pay more tomorrow to reach the same shopper.
Let me be straight with you. The real opportunity is bigger than the one you’re being sold, and it’s not where you’re looking.
Where We Actually Are
Gartner Inc. placed agentic AI at the Peak of Inflated Expectations on its 2026 Hype Cycle. Checkout.com’s June 2026 report found that 3% of transactions involved an AI agent, while 89% of merchants were preparing for agentic commerce. Nine in 10 companies are building for something touching three in 100 orders.
Then came the test. In November 2025, Walmart put 200,000 products into OpenAI’s Instant Checkout. By March 2026, it pulled out after in-chat purchases converted at a third the rate of shoppers who clicked through to the site. People could buy without leaving the chat, but they wanted to see more about the product first, so they went to the PDP.
That’s not a glitch. That’s a person behaving like a person.
Found Isn’t The Same As Chosen
In October 2025, the Interactive Advertising Bureau watched this in live shopping sessions. Shoppers averaged 1.6 steps before AI entered the journey and 3.8 steps after. AI is adding steps, and every one is a chance to lose the sale. As a result, your PDP carries more weight than ever.
The AEO work is worth doing. Fixing the structured product data gets you in the result set. However, it doesn’t get you chosen. Visibility is a marketing problem. Choice is a conversion problem.
Follow The Money
Trace a single dollar from your media plan to your P&L and see where it stops.
AI referral traffic is growing off a small base. Retail media is still expanding. One of those funds your volume next year, and it’s not the free one. You’ll keep buying paid traffic because it works. If 95% of your volume next year comes from a funnel you already pay for, where would a rational person spend their time?
The most powerful thing in that funnel is the one nobody measures. Your product images drive your conversion rate, which drives your ROAS (and your ACoS if you’re an Amazon advertiser). Paid, organic or agentic doesn’t matter. Every page view lands on the same handful of images, which is why fixing them raises the return on every dollar you spend.
A hero SKU pulling 100,000 sessions a quarter at 10% conversion does 10,000 orders. Move that to 11.3% and it does 11,300. Same traffic, same spend, 1,300 more orders. Apply that same math across 100 SKUs, and your entire media plan is 13% more productive.
At most large CPGs I’ve worked with, executive incentives come down to some version of three things: organic sales growth, net profit or EPS, and free cash flow. Improving conversion on traffic you already bought moves all three because revenue arrives without incremental media spend.
Three Things I’d Do Today
1. Get your AI benchmark. How much traffic is AI actually sending you at your price point? For some, it’s growing fast. For others, it’s noise. Find out before you spend on more people, process and technology.
2. Don’t spend another dollar on retail media until your team shows you how your PDP images convert. You’re about to commit another year of budget to images nobody has scored. Some will be fine. Some are costing you real money. You can’t tell which is which today.
3. Decide who owns the image. Every other line on your plan has a name and a number attached. Media has an owner. Trade spend has an owner and a whole team auditing it. The one asset every page view lands on gets approved by a few people in a room who like it. Nobody has ever lost a bonus over the wrong product image. When did that become acceptable?
Looking Ahead
There will be categories where agents take the human out. Sell 3/8-inch stainless hex bolts, and an agent should buy that without anyone looking at a photo. That’s not what gets sold online the most. Walk through your pantry, your closet or your laundry room, and almost none of what you see got there because a machine decided it should.
Think about the last 10 things you bought. You looked at nearly every one first. That won’t change because vision is the oldest purchasing interface there is. What has changed is it can finally be measured.
Soon, publishing an unmeasured image will seem insane. We measure every dollar of media. Don’t keep guessing at the thing the shopper actually sees.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
