Millennials now name ChatGPT as their second most-used product-discovery tool, at 41%, ahead of Amazon. AI product discovery has passed the largest marketplace in retail without acquiring the one thing every other channel on that list acquired early: an ad market at the point of consideration. Google, Amazon, YouTube and Instagram were all monetized long before they reached that share of the decision.
- PYMNTS Intelligence's July 2026 study puts ChatGPT at 41% for millennial product discovery, behind Google at 57% and ahead of Amazon at 37%, YouTube at 29%, and Instagram and Gemini at 26% each.
- Among millennials using generative AI for shopping research, 42% say it has fully or mostly replaced their previous research methods.
- Every neighbouring channel sells inventory inside the moment of consideration. AI sells a slot beside the answer, not a place inside the shortlist.
- Retail and shopping advertisers are rotating spend out of ChatGPT's own ad product exactly as retail discovery moves in. That is a format mismatch, not weak demand.
- The publisher reviews and buying guides feeding those shortlists are priced at zero, which is why the channel looks cheap to the buy side and looks like a loss to the supply side.
What AI product discovery actually is
AI product discovery is the stage where a shopper asks an assistant what to buy and receives a synthesized shortlist instead of a page of ranked links. The consideration set narrows before the shopper reaches a merchant, a marketplace or a shelf. The inputs are third-party pages: reviews, comparisons, category guides. The output is three or four brand names, with no auction, no bid and no line item anywhere in the process.
Paid search, retail media and the shortlist nobody sells
In the open-web economy, every point of discovery share arrived with a matched ad market and a paid supply side. Google's 57% is monetized by search advertising. Amazon's 37% is monetized by sponsored products and a retail media business. Social's share is monetized by the feed. The publishers whose review pages powered the research step were paid through programmatic and affiliate revenue for hosting that decision.
The agentic web keeps the discovery function and removes both halves of that structure. Brands cannot buy their way into the shortlist, and the publishers supplying the evidence behind it are not compensated for the role. A 41% share of consideration with no transactable unit attached is not a mature channel. It is an unpriced one.
What PYMNTS found, and what the traffic data says
PYMNTS reported on 8 September 2026 that AI is moving upstream in the millennial journey faster than purchasing is moving online. Its own data supports the split: 77% of millennials made their most recent grocery purchase in a physical store, and only 16% bought groceries online. Discovery is digitizing faster than checkout.
One number needs framing. The 41% is survey-reported usage, not measured traffic; Similarweb data published by SparkToro in June 2026 shows AI tools sending under 1% of all web traffic outward, alongside 68% of Google searches ending without a click. Both figures are correct, and we use both, because the distance between them is the actual finding.
Why retail budgets are leaving the channel retail discovery is entering
The strongest case for waiting
A disciplined CFO has real grounds to hold. Measured referrals stay under 1% of web traffic. OpenAI's $1bn annualized run rate, announced on 31 August 2026, is a pace of roughly $83m a month rather than booked revenue, as Digiday noted. And Sensor Tower reported on 1 September 2026 that ChatGPT's advertiser mix is rotating away from shopping and retail toward financial services, software, travel and wellness. The category with the most AI discovery is the category pulling back.
Why a channel that decides without referring is mispriced, not small
Read the rotation as a verdict on the format, not the surface. Retail advertisers tested a click-billed slot under the answer and found it does not touch the shortlist decided above it. Adobe Analytics reported in July 2026 that AI-referred retail visits convert 60% better and produce 53% more revenue per visit than non-AI traffic, an eleventh consecutive month of outperformance. The channel is doing the persuasion upstream and handing over buyers already decided. In a click-based ledger that work is invisible, which is exactly what mispricing looks like.
What this means for brands and for publishers
For CMOs, media buyers and agencies: fund shortlist presence, not sessions
Stop scoring AI as a performance channel and stop waiting for session volume that structurally will not arrive. Buy presence on the publisher pages engines actually consult in your category, and measure mention share by query cluster rather than referrals. Categories that close offline, grocery, automotive, telecom, banking, need this first, because there is no click to attribute in the first place.
For publishers: buying guides are shortlist inventory now
Your commercial-intent archive, the reviews, the comparison tables, the best-of guides, is the raw material of the millennial shortlist. It earned money as a destination and earns nothing as an engine input. Price it as decision inventory rather than as pageviews, and treat every citation in a shopping answer as an ad impression on a product you have not yet sold.
Three signals to watch before the 2027 planning cycle
First, whether retail spend that left the on-platform slot reappears in an off-platform lane priced on citation instead of clicks. Second, whether agentic checkout standards, including Google's Universal Commerce Protocol launched at NRF 2026 with Shopify, Etsy, Wayfair and Target, close the transaction leg while the discovery leg stays free. Third, whether mention share becomes a currency buyers transact on rather than a dashboard metric. The channel that decides the shortlist will not stay unpriced for two more planning cycles.
Conclusion
Hold on to this: AI product discovery reached second place in the millennial journey before anyone built a market for the moment it owns, and the gap between 41% of consideration and under 1% of referrals is the size of the mispricing. Smalk AI closes it with Generative Engine Advertising, native ads for AI agents placed on the publisher pages engines cite most, so brands can buy shortlist presence and publishers can finally be paid for supplying the answer. Watch the retail advertisers who just rotated out of the slot beside the answer: where they spend next tells you when this category prices in.
