Here is the sentence I keep repeating in buyer meetings this quarter. The visitor who arrives on your product page from ChatGPT or Gemini is already further down the funnel than the one who typed your category into Google, and that visitor converts about ninety percent better than the rest of your traffic. The window between discovery and purchase has collapsed, and the product page is now the first page a serious shopper sees.
That is the part everybody in retail finance can get behind. The part the same executives tend to miss, because it lives two floors down from them, is that a third of what sits on the average retail page is invisible to the model that just sent the buyer. The AI reads what it can read. On a product page it can usually read about two thirds of the content.
I have been building WooCommerce and WordPress stores long enough to recognise the shape of this problem. It is the same shape as the mobile-browser story in 2012. The traffic is already changing, the money follows a few quarters later, and the sites that get the engineering right between those two points capture a disproportionate share of the shift. The engineering this time is unglamorous: clean attributes, honest copy, a crawler that can read your page without running your JavaScript.
What follows is the published data on the gap, the operators doing something about it, and the first ninety days of that work on a mid-market WooCommerce estate. The numbers are Adobe’s. The operators are on the record in June and October 2026.
Two thirds of a retail product page is a lot to be invisible
Adobe Digital Insights built a tool it calls the AI Content Visibility Checker and ran it across more than a trillion U.S. retail visits earlier this year. The director of Adobe Digital Insights, Vivek Pandya, has been publishing the readings in two-month increments on Adobe’s own blog and in press briefings since April. The headline from the first pass, released at the launch of the checker, is the one everyone has quoted: homepages score around 75 percent, category pages around 74, and product pages around 66, the lowest category of page on a retail site. Thirty-four points of your product page, on average, cannot be parsed by the model that is trying to summarise it for a shopper.
Pandya expanded the sample over the summer, and the score fell. In the August update covered by Chief Marketer, he put apparel at 76 percent, electronics at 70, cosmetics at 68, sporting goods at 67, furniture and home at 64, general merchandise at 63, and grocery at 59. His framing is restrained. “These figures demonstrate the ongoing value AI provides in the ecommerce sector,” he told Chief Marketer on 19 August 2026. The thing he does not say out loud, that the operators I talk to say for him, is that the category with the highest conversion lift from AI traffic is also the category whose pages are weakest at being read by it.
Adobe’s October update made the gap uncomfortable. AI-referred traffic to U.S. retail sites in September 2026 converted 90 percent better than non-AI traffic and generated 83 percent more revenue per visit. The same month, Pandya’s team reported AI traffic was up 71 percent year on year and 1,772 percent since Adobe began tracking it in October 2024. Holiday forecasts carry the direction forward: 159 percent growth projected for Thanksgiving, 95 for Black Friday, 88 for Cyber Monday, all year over year. A channel growing that fast with a 90 percent conversion premium is not a channel you can afford to be two thirds readable on.
Before I walk through the levers, one caveat worth keeping. Readability is necessary and not sufficient. A perfectly machine-parseable page does not force an LLM to cite you, any more than a perfectly indexed page in 2012 forced Google to rank you. What it does is move the question from “can the model see this at all” to “has the model been given a reason to pick this over the alternative,” which is a much better question to be arguing about.
Lever one: model your attributes, do not write them into paragraphs
The cheapest, highest-leverage thing a store can do this quarter is to stop hiding product specs in prose. Material, dimensions, weight, compatible models, country of manufacture, battery life, warranty, washing instructions. Every one of those is a discrete attribute, and every one of them belongs in structured data and in a visible spec table rather than inside a marketing paragraph. The model sees prose as prose. It sees a table as rows.
AJ Ghergich, global vice president of AI at Botify, made the point bluntly in Modern Retail‘s 16 June 2026 piece by Mitchell Parton on how AI is forcing retailers to rethink product pages. “If you go look at a lot of product pages, there’s JavaScript everywhere,” Ghergich said. “Carousels, floating things.” Those are the first candidates for simplification. A carousel that reveals the fifth attribute on the third tap is a carousel the model never reaches.
Target has done something concrete here. Its vice president of digital product management, Ranjeet Bhosale, told Shoptalk last fall that Target has rebuilt its site to be “machine-readable” for its own AI tools and for outside engines pulling from it. Modern Retail paraphrased the point rather than quoted it at length. The direction is clear, and consistent with traffic data its own analysts have been publishing. A retailer that attracts a disproportionate share of AI-driven traffic is worth watching on the structural side of this problem.
On WooCommerce the concrete move is to promote free-text product descriptions into typed attributes. Material, size, colour, compatibility, warranty, country of origin. Each one becomes a product attribute, each becomes a facet in the catalogue, each flows into the Product schema that the store emits on every PDP. The merchandiser’s paragraph stays for the human reader; the model reads the attributes.
Lever two: teach the model what the category is actually for
The second lever is content on the category and product pages that does not exist on most retail sites because it was never useful for Google: a short, honest, model-facing answer to the question the shopper actually asks. “Is this suitable for a six-month-old.” “Will this fit a 2019 Ford Focus.” “Does this ingredient keep me awake.”
The example Mitchell Parton used in Modern Retail is the supplement brand Olly. Jennifer Peters, Olly’s director of DTC, marketing technology and digital compliance, has had her team add a frequently-asked-questions section to its product pages that explains ingredients and their efficacy in plain language. The intent, as Parton reported it, is to guide LLM users asking about sleep difficulties toward Olly’s sleep gummies. The move costs almost nothing. It changes what the model has available to recite.
Max Sinclair, founder and CEO of the discovery platform Azoma, put the shape of the query well in the same piece: “Does this ache 20 miles in while running a marathon.” That is not a keyword, it is a sentence, and the product page that answers it in a sentence is the one that gets lifted into the answer. A spec table alone will not do it. A spec table plus two paragraphs of plain prose addressing the three questions your category actually gets asked, that will.
Lever three: render the page, do not hide it behind JavaScript
This is where the readability score mostly lives, and where the mid-market has the most room. Adobe’s August category figures reward retailers whose pages are flatter and more editorial: apparel and electronics score at the top, general merchandise and grocery at the bottom, and the difference is largely a difference in how aggressively the page uses client-side rendering, tabs, modals, and lazy-loaded blocks that only expand after a user gesture. The crawler is not a user. It does not tap. If the price only appears after a script has run and a region has filled in, half the AI surfaces will record your page as a page without a price.
Blaine Nielsen, president of the retail division at Rithum, framed the state of the practice in Modern Retail with the dry line this market needs more of. “It’s pretty early innings,” he said. “People don’t have real hard numbers on how effective it is, but they are seeing positive trends.” That is the honest read, and it is why I would not spend the quarter chasing a specific LLM. I would spend it making sure every important fact on the page is in the HTML that gets served, not in the JavaScript that gets executed.
The lever that does not work yet
There is one lever in this market that nobody has got running at scale, and it is the one the CFO will ask about first. You cannot cleanly attribute a sale to a specific LLM citation of a specific product page. You can measure AI-referred sessions at the channel level, and GA4 is now reasonable at it. You can see the lift in revenue per visit. You cannot ask, “which forty-three product pages did ChatGPT quote verbatim last week, and which skus moved as a result.” The crawlers do not leave that trail behind, and the browsing LLMs that do pass a referrer rarely pass the query that produced the citation.
Ghergich’s own caution in Modern Retail is worth keeping on the wall of the war room. “I’ve yet to see anyone who has north of 5% of their revenue attributed to AI,” he said. “You can’t just go crazy caring just about AI, when 95% of your revenue’s flowing from traditional.” The channel is growing fast off a small base, and the measurement is immature. Build for it anyway, because the base will not stay small, but do not let the readability project crowd out the old-fashioned SEO that still pays the bills.
What we would build, in the first ninety days
Here is how we would spend the quarter on a mid-market WooCommerce estate with two engineers, a merchandiser and a product owner. It is not a transformation plan, it is a sequence of unglamorous moves in the order that produces a readable product page without breaking the human one.
- Sprint one, audit the top hundred. Rank SKUs by last-twelve-months revenue, then fetch each page with JavaScript disabled and check whether price, availability, five key attributes, warranty, and the primary image’s alt text are in the HTML. Half a day of engineering; a list that drives every later decision.
- Sprint two, promote attributes. Migrate free-text specs into WooCommerce product attributes with proper taxonomies. One week for the merchandiser, three to four days of engineering to expose them in Product schema and in a visible spec table.
- Sprint three, three questions per category. Give the merchandiser a brief: for every category, the three questions buyers ask that the current pages do not answer in plain prose. Write them in sentences under the spec table. A paragraph each. The Olly move, applied one layer up.
- Sprint four, render what you serve. On the same top hundred, move price, availability, variations, review counts and JSON-LD out of client-side rendering and into the server response. A healthy block theme is probably already doing this; a decoupled stack is where this sprint earns or does not earn its keep.
- Sprint five, instrument honestly. Turn on referrer-based AI channel tagging in GA4, log raw referrers for anything you do not recognise, and ship a weekly report of AI-referred sessions, revenue per visit, and the top twenty landing SKUs. The readable figure is revenue per visit, not “LLM citations.”
- Sprint six, hold the line against decoration. Marketing will come back with a hero carousel that pushes the spec table below a scroll-triggered animation. That is the sprint where someone says no. Governance is the cheapest lever in the plan, and the one that quietly fails without attention.
Rough totals: six to eight weeks of concentrated work for two engineers and the merchandiser, a governance line in the sprint ritual, and a quarterly readability re-score. On a healthy WooCommerce estate running a modern block theme and a sensible schema plugin, the floor is already where Adobe puts apparel, not where it puts grocery. On an older theme with carousels, lazy tabs, and a product data model living in Advanced Custom Fields textareas, sprint one goes long and sprint two pays for everything.
I would not frame any of this as an AI project to the board. The honest frame is that the store is upgrading its product data layer. The AI payoff is a side effect of an upgrade retail should have done anyway, and that is an easier conversation than one that depends on a channel still at five percent of revenue. If you want to see what your top hundred SKUs score against a cleaned-up pass, we run that as a scoped two-week engagement on WooCommerce estates.
Last modified: November 20, 2026