A shopper who leaves your product page has usually not rejected the product.
Everything in your analytics implies the opposite. The same moment, seen from both sides:
What analytics recorded: Product page. 1 min 40 s. Exit.
What the shopper was thinking: “The spec says 35.7 cm. My bag’s sleeve is 36 cm. Is that really enough clearance with a case on? … I’ll figure it out later.”
An exit reads as a verdict: they looked, they declined. But an exit records a behavior, not a reason, and the reasons are invisible because the page gave the shopper no way to voice them. Nothing about that laptop was wrong. The shopper left with the question, and “later” mostly never comes.
When a product page can answer back, shoppers type their reasons out. I have read thousands of those messages across the stores we serve, and they changed my mind about what abandonment is. It is mostly not rejection. It is unresolved uncertainty at the exact moment resolution was worth the most.
The mechanism: buying is a decision under uncertainty
A purchase is a small bet. The shopper trades certain money for an uncertain outcome: will it fit, will it work with what I own, will it do the job I am imagining. Every unanswered question widens the uncertainty, and past a threshold the rational move is not to bet. “Not buying” is the zero-risk option sitting one thumb-gesture away, and it requires no justification.
What makes product pages special is timing. The shopper on a product page is at the highest intent they will ever have for that item: they searched, clicked, and read the price. Uncertainty resolved in that moment converts at that moment’s intent. The same answer delivered tomorrow by email meets a different person, one whose tab is closed and whose enthusiasm has cooled. The contact form does not save the sale; it schedules a slower loss.
The industry’s aggregate numbers are consistent with this reading, in the part of the funnel where measurement exists. Baymard’s meta-analysis of checkout studies puts average cart abandonment around 70 percent, and the leading stated reasons are forms of uncertainty: unexpected costs, unclear delivery, doubts about returns and payment. And 43 percent of US shoppers say they abandoned simply because they were “just browsing.” I read that differently than most commentary does: “just browsing” is what an unresolved decision feels like from the inside. Nobody reports “I had a sizing question nobody answered.” They report not being ready.
A caveat the checkout data cannot give you: it measures people who put something in a cart. Product page abandonment happens earlier and is essentially unmeasured, which is exactly why the questions shoppers type are such unusual evidence.
What the questions actually are
They cluster tightly. Will it fit: bodies, bag sleeves, a 35.7 cm laptop in a 36 cm pocket, and mostly not missing data but missing confidence in the shopper’s own reading of the data. Will it work with what I own: chargers, mounts, model years, the intersection of the customer’s setup and your catalog that no static page can enumerate. What does the more expensive one buy me: comparison anxiety between two of your own products. And what happens if I’m wrong: shipping, returns, the exit cost of the bet.
What stands out in this list is how answerable it is. The data almost always exists in the catalog, the policy page, or the size chart. The failure is that the information cannot be assembled for this shopper’s specific doubt at the moment the doubt occurs.
Existing approaches, and the decision fatigue detour
Stores attack this with better pages: size charts, comparison tables, FAQ sections, reviews. All worth doing, and all limited the same way. A page is a bet on which questions will occur. The head of the distribution gets answered; the long tail (the customer’s specific bag) cannot be, because the author never met this customer.
The industry’s other favorite explanation is choice overload: too many options paralyze the shopper. The famous jam study says fewer options sell more. The research record is messier: a meta-analysis of 50 choice overload experiments by Scheibehenne, Greifeneder, and Todd found a mean effect near zero, with high variance across studies. Assortment size alone does not reliably hurt.
The conditions under which overload does appear are the interesting part: when options are hard to compare and preferences are not yet formed. That describes a product page comparison moment exactly. So the practical lesson is not “stock less.” The cost of choice is a cost of comparison, and comparison difficulty is something you can fix without touching the assortment.
A better model: the page as the end of a conversation
A great physical shop resolves uncertainty conversationally: the customer holds the product, asks the person at the counter two questions, and decides. The product page took that scene and kept only the product and the price. Everything conversational was stripped out and scattered: sizing to a chart, policies to a footer link, comparisons to the shopper’s short-term memory across seven tabs.
Abandonment, in this frame, is what happens when the stripped-out conversation was load-bearing.
The fix follows from the frame: put the conversation back at the moment of decision. An assistant on the product page, grounded in the catalog and policies and aware of which product the shopper is looking at, turns “good for a 1.90 m rider?” into “yes, take the L” in ten seconds, at peak intent. Two properties decide whether this works: page context, so “will this fit” needs no link pasted, and strict grounding, because on a product page one invented spec does more damage than a hundred good answers repay. When the honest answer favors the cheaper product, giving it is not lost margin; it is the trust that converts the next five visits.
The limit: an assistant resolves informational uncertainty. It does not fix a price that is in fact too high, a shipping cost revealed too late, or a product that is wrong for the customer. Some abandonment is the system working.
Practical implications
Read your ten most visited product pages as your most skeptical customer would, and fix what fails at a glance; recurring questions are page defects, not assistant work. Then let the pages answer back, and treat the transcript stream as the analytics layer you never had: a written record of every doubt standing between your pages and a sale, including demand for things you don’t stock, which no sales report can contain by definition. The mechanics of answering from vague, situational language are their own subject.
The shift
We optimized product pages as if shoppers leave because they were unconvinced, and built ever more persuasive pages. But a shopper with an unanswered question is not unconvinced; they are unresolved, and persuasion does nothing for them. The next conversion percent on most stores is not hiding in better copy. It is sitting in questions nobody can currently ask.