Chapter 5 of 11
Product Page Optimization
The product page has one job. Here is how to work out whether yours is doing it, what to fix first, and which popular PDP advice reliably does nothing.
Contents
- The product page has one job
- Above the fold
- Social proof that reads as evidence
- Variants, quantity, and the cost of a bad selector
- Descriptions, specs, and the accordion problem
- Cross-sells, and where they belong
- Shipping, returns, and the two facts that move the number
- The page before the page
- What to test first, by traffic level
- Mobile failures that do not show up on your laptop
- The advice we have stopped following
The product page is where most ecommerce conversion work happens, for a simple reason: it is where the decision is made, and on most stores it sees 60–80% of sessions.
It is also where the most nonsense advice accumulates, because it is the page everybody has an opinion about.
The product page has one job
Not “sell the product.” Not “communicate brand values.”
The product page’s job is to answer every question standing between a visitor and the decision, in roughly the order they ask it, without making them leave the page to find an answer.
That framing does more work than any checklist, because it converts a vague design problem into a research question you already answered in Chapter 2. The questions are in your reviews and your support inbox. Your page either answers them or it does not.
The universal ones, in rough order:
- Is this the thing I was looking for?
- What does it actually look like?
- Will it fit / work for me / suit my situation?
- Is it worth this price?
- Do other people like it, and are those people like me?
- When will it arrive and what does shipping cost?
- What happens if I am wrong about it?
Walk your own PDP with that list. Count how many require a scroll, a click, an accordion, or a trip to a different page. Every one of those is friction on a question the customer needs answered before they will spend money.
That is the audit. It takes fifteen minutes and it outperforms most paid tools.
Above the fold
You have one screen, and on mobile that screen is about 640 logical pixels tall. Roughly 74% of ecommerce traffic is mobile, so design this for the phone and let desktop inherit.
Four things earn the space:
The image. Large, tappable, opening to a gallery. Product photography does more conversion work than any copy on the page, and the single most common PDP failure we see is images that require pinch-zoom to evaluate texture, scale, or finish. If you sell anything where material matters, a zoom that actually works is not a nice-to-have.
The name, and what the thing is. “Cloud Runner” is a name. “Cloud Runner, lightweight daily trainer, 8mm drop” is a name plus an answer to question one. Stores with brand-led naming lose people who cannot tell what they are looking at.
Price, and any comparison price, honestly. If there is a subscription price, show both. Hiding the higher one-time price to make subscription look default is the kind of thing that converts once and refunds twice.
One claim that is not marketing language. The single differentiating fact. Not “premium quality” or “thoughtfully designed.” A fact: “Ships in a 4-inch box that fits a mailbox.” “Rated 4.7 by 2,300 buyers.” “Made from a single piece of steel.”
What does not earn above-fold space: a hero banner, a breadcrumb trail taking three lines, an announcement bar stacked on a cookie notice stacked on a promo strip, and a set of trust badges nobody has ever looked at.
On sticky add-to-cart buttons: they usually help on long mobile pages, where the decision and the button otherwise end up hundreds of pixels apart. But sticky elements eat vertical space permanently. If your sticky bar covers 12% of a 640px screen on every scroll position, you have paid for that button with content. Test it, do not assume it.
Social proof that reads as evidence
Every store has reviews. Most stores get almost nothing from them, because they present them as decoration rather than as evidence.
The star average is nearly useless on its own. Every product on the internet is rated 4.6. A visitor has learned to discount it. What they have not learned to discount is a specific review from someone recognisably like them, describing the thing they are worried about.
Three things that consistently outperform a star widget:
Reviews sorted or filtered by the objection. If fit is your top review theme, let people filter reviews by “runs small / true to size / runs large” and show the distribution. You are answering question three with customer words instead of your own.
Customer photos over studio photography, in the review block. Studio images sell the idea; customer images sell the reality, and the reality is what people are checking for at that stage of the decision. Swapping the order is a cheap test and frequently a winning one.
Count over rating. “2,341 reviews” does more than “4.7 stars” because volume is harder to fake and reads as evidence of a real business.
What reliably does nothing: generic trust badges, “As seen in” logo rows for publications nobody recognises, and live purchase notifications. That last one deserves a specific mention because it is still widely sold. In our own testing it has been flat more often than positive, and on stores with a considered purchase it has been negative. It reads as pressure, and pressure is not trust.
Variants, quantity, and the cost of a bad selector
Variant selection is where good product pages quietly leak.
Show unavailable variants as unavailable, not as absent. Removing a sold-out size makes the customer think you never made it. Greying it out with “back in stock soon” and an email capture turns a dead end into a lead.
Swatches beat dropdowns for anything visual. A dropdown for colour makes the customer open, read, close, and imagine. A swatch answers the question in place.
Update the image on variant selection. This sounds obvious and is broken on a startling number of stores. If someone selects “forest green” and the photo stays navy, you have introduced doubt at the exact moment of decision.
Quantity selectors are usually the wrong tool. For most stores, the customer wants one. A quantity stepper adds a decision without adding revenue. Bundles do the job better, because they price the decision instead of just permitting it. The Men to Mars starter kit is built on exactly that idea.
Descriptions, specs, and the accordion problem
Below the fold, most product pages become a wall of brand copy followed by an accordion nobody opens.
Lead with the specification, not the story. People scrolling past the buy button are usually looking for a fact: dimensions, materials, ingredients, compatibility, care. Put a plain specification block early and put the brand narrative after it. The customer who wants the story will scroll; the customer who wants the dimension will leave if they cannot find it.
Accordions hide things from search engines’ users and from scanners. Collapsed content is still indexed, so the SEO objection is largely obsolete, but a customer scanning a page does not read section headers as an index. If a piece of information is needed by most buyers, it should not be behind a tap. Accordions are correct for the long tail: full ingredient lists, care instructions, technical certifications.
Write specs as a table, not as prose. “The bag measures 42cm tall by 30cm wide with a 12cm depth and weighs 890g” is worse than four table rows, because nobody reads a sentence to extract a number.
Answer the comparison question. For any store with more than a handful of products, the customer’s real question is often “which one of yours do I want,” not “do I want this one.” A comparison table between your own variants, or a two-line “choose this if / choose that if,” resolves the decision instead of leaving them to open four tabs.
Cross-sells, and where they belong
Cross-sells on a product page are a tax on the decision you are trying to help someone make. They are also revenue. The resolution is placement.
Above or beside the buy button: no. Anything that offers an alternative before the customer has committed splits the decision. “You might also like” next to the add-to-cart is an invitation to start over.
Immediately after add-to-cart: yes. The decision is made, and a genuinely complementary item is now helpful rather than distracting. This is the highest-value cross-sell position on most stores.
Bottom of the page: mostly ignored, harmless. Useful for people who did not want this product and are still browsing.
The test for whether a cross-sell is complementary or competitive: would a customer plausibly buy both? Filters with a coffee grinder, yes. A different grinder, no. Recommendation widgets left on default settings routinely serve the second kind, because “similar products” and “products bought together” are different algorithms and only one of them belongs on a product page.
Shipping, returns, and the two facts that move the number
Two pieces of information belong on the product page, near the button, in plain text, on every store:
When it arrives and what it costs. Not “free shipping over $75” as a badge in the header. A specific line near the add-to-cart: “Free shipping. Arrives Tue, Sep 12 to 48201.” Delivery-date estimates near the button are one of the most reliably positive changes in ecommerce, because they convert an unknown into a fact.
If you have a free shipping threshold, say how far away the customer is from it. A progress-bar version tends to lift average order value rather than conversion rate, which is exactly why you watch both metrics, and why the free-shipping threshold calculator models the margin as well as the basket.
What happens if it is wrong. Return window, who pays, how long the refund takes. Risk reversal near the button does more than a testimonial does, because it removes the actual barrier rather than arguing against it.
The pattern: these are the two questions in the list that most stores answer somewhere else on the site. Somewhere else is not good enough. They are decision inputs, so they belong at the decision.
The page before the page
Product page work is limited by what happens upstream. If your collection pages send people to the wrong product, a perfect PDP cannot recover it.
Show enough on the card to make a real choice. Price, a second image on hover or swipe, star rating with count, and any variant range. A collection card with only a photo and a name forces people to open products one at a time to compare, and most will not.
Make the sort and filter options match how people actually decide. Filters for the attributes that appear in your review mining, not the attributes in your product database. If people choose by fit, filter by fit.
Do not hide out-of-stock behind a filter default. Show it, marked, with a back-in-stock capture.
Watch the collection-to-product click rate as its own metric. It sits above the funnel step everyone stares at, and on stores with large catalogues it is frequently the real constraint.
Internal search deserves a mention of its own: searchers convert at multiples of browsers on most stores, because search is a statement of intent. Read your search logs. The queries that return nothing are a list of products customers expect you to have, in their own words, and the ones that return the wrong thing are a synonym problem you can fix in an afternoon.
What to test first, by traffic level
This is where Chapter 4 meets reality. The right first move depends entirely on how much traffic the page gets.
Under ~20,000 monthly sessions on the page. Do not test. Audit against the seven questions, fix everything that fails, ship it all at once, and watch the trend over the following eight weeks. You will not get clean attribution. You will get a better page, which at your size is worth more than attribution.
20,000–75,000. Test one thing at a time, and only changes big enough to produce a 10%+ relative effect. Above-fold restructuring. Adding delivery dates. Moving reviews into the buying decision rather than below it. Do not test button copy at this traffic level; you cannot detect it.
75,000+. You can run a real programme. Sequence it: structural changes first, then content, then the smaller refinements. Keep a running record so you stop re-testing eliminated branches.
For raw candidates once you know your level, we keep 30 product page tests tagged by impact and difficulty. Filter by difficulty first if you have no developer time, by impact if you have traffic and want the biggest swing. And if you want to see this applied to a real page rather than described, the Fieldnote Hydration teardown is an annotated PDP with the specific fixes marked.
Mobile failures that do not show up on your laptop
Three quarters of your traffic is on a phone and essentially all of your team reviews the page on a desktop. The gap is where the damage lives.
Tap targets that are too small or too close. 44px minimum, with real space between them. Variant swatches packed tightly are the usual offender, and a mis-tap on a swatch is worse than no tap at all because the customer now thinks they selected something they did not.
The keyboard covering the thing you are typing into. Any input near the bottom of the viewport, on iOS especially. Test every form field on a real phone, not in a simulator.
Sticky elements eating the screen. An announcement bar, a header, a sticky add-to-cart, and a chat bubble can between them occupy a third of a 640px viewport. Each was added by a different person for a good reason. Add up what is left.
Images that require pinch-zoom. Covered above and worth repeating because it is the most common serious PDP fault we find.
In-app browser rendering. Traffic from Instagram, TikTok and Facebook renders inside their own webview, where custom fonts sometimes fail, position: sticky sometimes misbehaves, and some payment sheets do not launch. If a meaningful share of your traffic is paid social, this is not an edge case, it is a primary rendering target.
Horizontal overflow. One wide table or one oversized image and the whole page scrolls sideways. It looks broken, because it is.
Run the device walkthrough from Chapter 2 specifically against this list. It takes an hour and it consistently outperforms a month of opinions about layout.
The advice we have stopped following
Four things that get repeated constantly and have not earned it in our accounts.
Button colour. Contrast matters for accessibility and for being findable. Which specific colour does not. The famous red-versus-green case studies were mostly contrast changes reported as colour changes.
Reducing page length for its own sake. Long pages convert well when the length is answers. The problem is never length, it is ordering. A 6,000px page that answers questions in order beats a 2,000px page that answers them out of order.
Countdown timers on evergreen products. Covered in Chapter 1. Real scarcity works. Manufactured scarcity trains people to distrust everything else on your page.
Chat widgets as a conversion tool. They are a support tool. On a mobile PDP they occupy the bottom-right corner where thumbs live, and they cover content. If you keep one, delay it and make it small.
Next: the cart and checkout, where the customers who already decided to buy find reasons not to. If you would rather have someone do this page-by-page on your store and hand you a ranked list, that is the audit we sell.