If your Prime Day dashboard showed sessions down and conversion rate up, your listing didn't get better. Your traffic got pre-screened.
That's the one thing to take from the first hard post-mortem dataset on Prime Day 2026, and it matters more than any single number in it: the shoppers who reach your product page are increasingly arriving after the comparison shopping already happened — inside an AI assistant you can't see. The product page is sliding from top-of-funnel to bottom-of-funnel, and every benchmark you've built on sessions and CVR moves with it.
What happened
CommerceIQ published its Prime Day 2026 first-look report, covered by PPC Land on July 7, 2026: across US brands, product page traffic (glance views) fell 10.3% year over year while unit conversion rose 17.1% to a blended 20.0%, ad spend fell 8.8%, and ordered revenue landed just 1.7% below last year. CommerceIQ attributes the conversion lift in part — carefully, as "likely" — to Alexa for Shopping, the assistant that replaced Rufus on May 13 and now handles a meaningful share of mobile queries.
Why most brand owners will read this wrong
The dumb take comes in two flavors.
Flavor one: "Conversion is up 17% — shoppers love my listing." No. A blended market-wide conversion jump during a traffic decline is a mix shift, not a merit badge. Fewer, hotter sessions mechanically raise CVR even if your page got worse.
Flavor two: "Traffic fell 10% — Prime Day is dying, Amazon is dying, panic." Also no. Units sold rose 4.9%. Revenue was basically flat. Demand didn't leave; the browsing left. It moved upstream into the assistant, which does the compare-and-eliminate work that used to generate eight product page visits and now generates two.
The real signal: this is the first event-scale receipt for something I've been arguing since the spring — the AI layer isn't a new traffic source, it's a new filter in front of your existing one. Two weeks ago I wrote about how Seller Central blends AI-path and grid-path shoppers into one CVR number. This report is what that blending looks like at Prime Day volume: a 17.1% conversion "improvement" that nobody's listing earned.
What actually changes for someone running $200K/mo on Amazon
Your CVR baseline just inflated, and your A/B tests are contaminated. If you're reading a post-May-13 CVR lift as proof your new hero or price test worked, some unknown slice of that lift is mix shift. A creative test that shows +8% CVR against a pre-May baseline might be a real +3% and a structural +5%. Directionally fine, magnitude unreliable. Anything you're paying for by the point of CVR lift — an agency bonus, a test vendor — deserves a re-read.
Sessions stopped being a health metric and started being a rationing metric. At 10% fewer glance views for flat demand, each session carries more revenue. A $200K/mo account at a 2% blended margin of error used to shrug off a weak slot-2 image; at compressed traffic, every pre-sold shopper who bounces was closer to buying than bounces used to be. The cost of a page that can't close went up even though CVR went up. Sit with that one.
The fight moved to eligibility, not persuasion. If the assistant eliminates you during the invisible comparison phase, you don't get a worse session — you get no session. The inputs that decide that phase are the unsexy ones: attribute completeness, review language, machine-readable deal structure. Your hero image doesn't render inside an Alexa recommendation. Your Supplement Facts attributes do.
The ad mix data is a quiet confession. Advertisers cut Sponsored Display 70% and Sponsored Brands 25% year over year, concentrated 88% of spend into Sponsored Products — and ROAS held, 4.78x versus 4.73x. During an event where the AI layer does the discovery, brands pulled upper-funnel ad formats and lost nothing they could measure. That's worth knowing before your Q4 budget meeting, with the caveat that one event doesn't prove display never worked — it proves the burden of proof flipped.
One more number for the road: out-of-stock losses rose 185% year over year. Compressed traffic means pre-sold shoppers, and a pre-sold shopper who hits a stockout is a completed sale you handed to the next ASIN in the consideration set.
What I'd do this week if I were them
- Re-baseline before you diagnose. Pull sessions and CVR separately, YoY, for your top 10 ASINs — event window and the two weeks after. Label May 13, 2026 as a structural break in your tracking. Any trend line that crosses it needs an asterisk.
- Recompute contribution per session, not CVR. Sessions are scarcer; value them like it. A SKU whose per-session contribution fell while CVR rose is losing the upstream filter game — that's an attribute and review-language problem, not a creative problem.
- Re-run any A/B test you launched before mid-May if a real dollar decision hangs on it. Same test, post-break window, clean read.
- Audit your stockout exposure like it's 3x more expensive, because it is. The 185% OOS-loss number plus pre-qualified traffic means days-of-cover thresholds set in 2024 are mispriced.
- Grade your top ASINs on assistant eligibility — attribute fill, review recency, structured deal data. Thirty minutes per ASIN. This is the new top of funnel and it doesn't show up in any report you currently read.
What I'd ignore
The "Prime Day is dying" discourse — units grew, read the whole table. The US-versus-EU divergence threads (EU revenue +20.3%, UK conversion +39% on a 34.6% traffic collapse) unless you actually sell there — interesting, not actionable. Anyone selling an "AI conversion optimization" service off the 17.1% headline this week — the number is real, the causation is CommerceIQ's own hedged "likely," and nobody selling you a retainer has the attribution data Amazon won't release. And the ROAS victory laps: flat ROAS on 8.8% less spend is efficiency, not growth, and you can always manufacture efficiency by refusing to grow.
The funnel compressed. The reports didn't. Until Amazon splits AI-path from grid-path traffic — and there's no sign it will — the operators who win are the ones who stop grading themselves on numbers that changed meaning in May.