Executive Summary
AI shopping tools are removing the human workarounds that have quietly covered for weak retail commerce systems.
Shoppers used to call the store when a product page was thin. Merchandisers fixed broken feeds in spreadsheets. An AI assistant does neither. It answers from whatever product, price, and availability data it can reach, and when that data is incomplete or inconsistent, it misrepresents your products or leaves them out.
So the question for your leadership team has changed. It is no longer whether your brand promises a good experience. It is whether your commerce systems can keep that promise with no person in the loop.
- AI channels remove the human workarounds, such as store calls and spreadsheet fixes, that have hidden weak commerce data for years.
- Product pages scored 66 percent for machine readability in Adobe’s April 2026 benchmark, the lowest of the retail page types it measured.
- AI-referred visits converted 40 percent better than non-AI channels on US retail sites during Prime Day 2026, according to Adobe.
- Four of the five customer promises in this article depend on product data, which makes product data the first place to look.
- Our client GSF Car Parts grew annual revenue by 140 percent after we rebuilt its commerce platform and product data layer, with human shoppers driving the gains before AI channels matured.
- Five checks show where your own gap sits. Jump to the five checks.
Picture a shopper who asks an AI assistant for a carry-on suitcase that meets airline size limits and arrives by Friday. The assistant checks two retailers. One lists the case with exact dimensions, confirmed stock, and a delivery date. The other sells the same case, but its dimensions sit in an image, and its stock figure is a day old. The assistant recommends the first retailer. Nobody calls the second retailer, and nobody there learns the sale existed. The gap behind that lost sale opened long before AI arrived.
People Have Been Filling In The Gaps So Far. AI Shoppers Cannot.
In 2005, Bain & Company surveyed 362 companies about the experience they delivered. Of those companies, 80 percent believed they delivered a superior customer experience. Their customers rated only 8 percent of them as superior.
That study is two decades old, and retail technology has changed completely since then. Yet the pattern it describes is still easy to recognize: leaders judge the experience by what their teams intend to deliver, while customers judge it by what the systems actually deliver.
So why did that gap stay tolerable for so long? Because people filled it. Shoppers and staff absorbed weak commerce data on the retailer’s behalf, usually without anyone noticing, and they did it in two places.
People Filled in the Gaps in Two Ways:
1. Shoppers Worked Around Thin Product Pages
The first workaround happened on the customer’s side. When a product page left a question unanswered, a shopper found another route to the answer. They:
- Called the store to check stock
- Read reviews to find a missing size or specification
- Zoomed into photos to confirm a detail the page never stated
- Made an informed guess and bought anyway
Each detour cost the shopper time, but it often saved the sale.
2. Staff Worked Around Systems That Disagreed
The second workaround happened inside the business, where your teams did the same thing at a larger scale:
- Merchandisers patched broken product feeds in spreadsheets.
- Teams reconciled product information across systems by hand to keep it correct.
- Store associates translated thin product pages into useful answers for customers.
This work rarely shows up in a job description, but it is often what keeps the storefront accurate.
AI Channels Remove Both Workarounds
Because neither workaround appeared on a dashboard, the systems underneath looked healthier than they were. An AI channel takes both away at once. It has no store to call and no merchandiser to fix the feed, so it reads your data exactly as it finds it.
| Data gap | What a person did | What an AI channel does |
|---|---|---|
| Thin product page | Called the store or read reviews | Answers from the page as it finds it |
| Conflicting attributes across systems | A merchandiser reconciled them by hand | Has no way to tell which record is right |
| Availability that cannot be verified | Store staff confirmed stock for the customer | Often leaves the product or the retailer out |
That shift is no longer hypothetical. The numbers that follow show how quickly AI is becoming the route by which shoppers reach your products.
AI Cannot Recommend or Buy What It Cannot Read
AI systems build their answers from the product information they can reach: pages, feeds, indexes, and APIs. You do not control every source a model may use, but you do control the quality and accessibility of your own product information. That control matters more as AI traffic grows, because a model’s answer offers no service recovery moment.
AI-Driven Retail Traffic Is Growing Fast
Start with the scale of the shift. Adobe Analytics recorded a 693 percent year-over-year increase in generative AI traffic to US retail sites over the 2025 holiday season, the largest rise of any industry it tracks. These are not casual visitors, either. In Adobe’s companion survey, 73 percent of shoppers who use AI for shopping said it is their primary source of product research.
Product Pages Are The Content AI Reads Least Well
Growing traffic would matter less if AI tools could read retail sites well. Adobe’s April 2026 analysis suggests they struggle exactly where it counts. It benchmarked how much of US retail websites machines can actually read:
- Homepages averaged a 75 percent machine-readability score.
- Category pages averaged 74 percent.
- Product pages averaged 66 percent, so roughly a third of product-page content is unreadable to the AI interfaces shoppers are starting to use.
Machine readability by retail page type. Adobe, April 2026.
Adobe machine-readability score, 0 to 100.
In other words, the pages that carry the most purchase intent on your site are the ones AI systems interpret least well.
AI-Referred Shoppers Already Convert Better
That weakness matters because AI-referred shoppers are unusually valuable. During Prime Day 2026, Adobe recorded an 89 percent year-over-year increase in AI-referred traffic to US retail sites. Those visits converted 40 percent better than non-AI channels, a sharp reversal from a year earlier, when the same traffic converted 23 percent worse.
Adobe notes that overall volume remains modest. Even so, the direction is clear. These shoppers often arrive after doing part of their research inside the AI tool, using whatever product information it could reach. If your data was incomplete during that research, they may never arrive at all.
Shopping Agents Cannot Make an Informed Guess
The next step raises the stakes further. An AI agent that shops on a customer’s behalf, which we cover in our agentic commerce coverage, cannot call a store or read between the lines of a thin page. So when product attributes conflict or availability cannot be verified, the safer action for the agent is often to exclude the product or the retailer. At that point, data quality stops being an internal hygiene topic and becomes a customer-facing commercial issue.
The question every retail leader now faces: can your commerce systems keep the promises your brand makes, with no person in the loop?
Five Customer Promises Now Depend On Systems, Not People
Answering that question starts with what your brand actually promises. Every retailer makes the same basic promises to its customers, and each one now depends on systems rather than staff.
Product truth is one authoritative, structured record of what each product is, what it costs, and whether it is available. Four of the five promises below rest on it, and the fifth rests on customer context.
Clean product data has always mattered, so none of this is a new agenda. What changed is the forgiveness: a person could compensate for a gap, and an AI channel cannot. That is why, for any retailer with discovery or AI-readiness problems, product truth is the first place I would look. The five promises below show where it carries the most weight.
Promise 1, “Show me the right product,” depends on structured attributes
The expectationA shopper asks an AI assistant for a waterproof hiking boot in a wide fit under $150 and expects relevant results.
What it depends onThat answer exists only if waterproofing, width, and price are stored as structured data fields, not as phrases buried in descriptive copy or a PDF specification sheet. In our commerce modernization work, we repeatedly find product information spread across several systems and kept correct only through manual reconciliation. We cover this pattern in more depth in product data quality as a commerce asset.
When it failsYour product drops out of the answer, and the sale goes to a retailer whose data was complete. Adobe's 66 percent product-page score suggests this risk is common.
Promise 2, “Tell me exactly what this product is,” depends on authoritative attributes
The expectationOnce the right product appears, customers expect every answer about it to be accurate, whether it comes from your site copy or from an AI assistant.
What it depends onHere, AI itself raises the risk. Some retailers are exploring AI-generated product content across very large catalogs. Generated copy, however, is only as accurate as the attributes underneath it, which makes accuracy an integration problem before it is a content problem.
When it failsAI tools produce confident, wrong answers at catalog scale, and customers act on information your systems cannot stand behind.
Promise 3, “Tell me the true price, stock, and delivery date,” depends on systems that agree
The expectationKnowing what a product is only helps if the customer can buy it on the terms shown. They expect the price, stock level, and delivery date they see to be the ones they get.
What it depends onThose terms come from your ERP, order management, commerce, and fulfillment systems, so all four must agree with one another. Product information describes the item; these systems decide whether you can sell it, at what price, and when it arrives.
When it failsEvery conflict between those systems eventually reaches a customer as a broken promise. For an AI agent, availability it cannot verify is a reason to leave the product out entirely.
Promise 4, “Give me the same answer everywhere,” depends on one shared source
The expectationCustomers also expect consistency. Your site, stores, marketplace feeds, service team, and any AI assistant should all say the same thing about a product.
What it depends onConsistency requires every channel to draw product and stock data from one shared source. When data sits locked inside the platform that holds it, each new channel needs a custom integration, and every release becomes slower and riskier.
When it failsChannels drift apart, and adding an AI channel becomes another IT project rather than a quick extension.
Promise 5, “Recognize me,” depends on connected customer context
The expectationThe last promise moves from the product to the person. A returning customer expects you to know them across web, store, loyalty, and service.
What it depends onThat recognition requires identity, consent, order history, loyalty activity, and service records to connect rather than sit apart by channel. We call this connected view a Loyalty Graph: one governed view of each customer's identity, consent, behavior, purchases, product affinity, and service history that your recommendation, service AI, and lifecycle marketing systems can all draw on.
When it failsYou can clean your entire catalog and still treat a loyal customer as a stranger.
None of these five dependencies is exotic. Each is ordinary data and integration work, and because the same data serves every channel, fixing it pays back across all of them at once. Our work with GSF Car Parts shows what that looks like in practice.
How We Rebuilt GSF Car Parts’ Product Data, and Why It Matters for AI Channels
GSF Car Parts is a UK car parts retailer with more than 200 branches and 400K unique SKUs. When GSF partnered with us, its five-year-old platform was breaking several of the promises above at once, and its customers and teams were absorbing the cost.
Where GSF’s Systems Were Breaking Promises
The problems sat in the data layer as much as in the storefront:
- Fitment-specific and long-tail searches frequently came back empty, so shoppers abandoned the site instead of browsing.
- A full product import took roughly 16 hours and often needed several retries, which slowed every catalog change.
- Stock and price updates ran only 4 times a day, which led to inaccurate stock levels, high returns, and cancellations.
For a parts retailer, each of these is a broken promise. Each one also forced a customer or an employee to work around the system, which is exactly the hidden cost this article describes.
How We Rebuilt GSF’s Product Data Layer
We replatformed GSF onto headless Adobe Commerce Cloud in phases, and we focused much of that work on the data underneath the storefront:
- We made Akeneo PIM the single source of truth for catalog data, so enriched product attributes flow into Adobe Commerce from one place.
- We engineered three backend interfaces in Adobe Commerce to reconcile vehicle fitment data from TecDoc and AutoGuru.
- We introduced compatibility-based search by make, model, and variant through Adobe Live Search, replacing the old year, make, and model pattern.
- We built automated API alerts that flag empty or dropped data feeds to GSF’s IT team before customers see the gap.
- We increased stock refreshes to 6 times a day and added real-time checkout validation against GSF’s regional and national distribution centers.
What Changed For GSF’s Customers And Revenue
The product data changes show up directly in GSF’s numbers:
- Product imports dropped from roughly 16 hours to 17 minutes.
- The zero-result search rate fell from 2.16 percent to 0.73 percent, a reduction of roughly 66 percent.
- Search-based conversion rose from 0.19 percent to 5.97 percent.
- Failed and invalid orders fell as stock accuracy improved.
Across the overhaul as a whole, which also covered checkout, fulfillment, and the user experience, GSF’s commercial outcomes included (GSF Car Parts case study):
Why GSF’s Results Matter For AI Channels
Human shoppers delivered every one of these results, and that is exactly why they matter here:
- The investment paid back in today’s channel, without waiting for agentic commerce to mature.
- The structured fitment data, single product record, and verified stock that now serve GSF’s shoppers are the same foundation an AI assistant or agent needs to recommend GSF’s parts with confidence.
GSF’s results also show the trade-off clearly. The work paid back, but it was a phased program rather than a single launch.
The Trade-Off: A Data Program Pays Back Later Than A New Front End
That trade-off applies to every retailer weighing this work. A product data program is slower and less visible than launching a new AI feature, so you should go in knowing what it costs and what it returns.
The Decision: Fix Product And Availability Data Before Adding AI Tools On Top
An AI assistant or search tool built on inconsistent data only exposes that inconsistency faster. That is why fixing the data first means committing to a program rather than a purchase.
The Gain: One Foundation Improves Every Channel
In return, the same work serves human shoppers and AI channels:
- Product data you structure for an agent channel also powers better on-site search for people, as GSF’s search results show.
- An availability service that an agent can trust also stops a store associate from promising stock that is not there.
The Risk: Results Arrive Later, And The Work Competes For Engineers
The cost is time and attention. A data program shows results more slowly than a front-end launch, and it competes with roadmap features for the same engineering time. Waiting carries its own cost, though. Every quarter spent on manual reconciliation adds data and process debt that the eventual fix must absorb. Meanwhile, retailers with structured data are better placed to capture AI-referred demand as it grows.
Before you weigh that trade-off, it helps to know how large your own gap is.
Run these this week
Five Checks Show Where Your Gap Is This Week
You do not need a program to find out. Your own team can answer each of these checks in days, and each one tests one of the five promises above.
- 01
Can your systems answer a specific shopper question from structured attributes alone?
Tests: Show me the right product
Worrying answer: Key attributes live in copy, images, or PDFs.
- 02
When AI writes or answers about a product, is there one authoritative record to check it against?
Tests: Tell me exactly what this product is
Worrying answer: Teams reconcile data by hand or in spreadsheets.
- 03
Do your site, stores, feeds, and service team show the same price, stock, and delivery date for an item?
Tests: Tell me the true price, stock, and delivery date
Worrying answer: Anyone on your team can name a recent mismatch.
- 04
Can a new channel get your product and stock data without a custom integration project?
Tests: Give me the same answer everywhere
Worrying answer: Every new channel starts as an IT project.
- 05
Can you recognize the same customer across web, store, loyalty, and service, with consent attached?
Tests: Recognize me
Worrying answer: Each channel holds its own customer record.
Honest answers to these five checks will tell you more about your readiness than any customer sentiment survey. And if they point to a gap, the next step is to measure it properly.
FAQs
No. Service and communication are earned strengths, and they still set retailers apart once a customer is in a conversation. They are now necessary but not sufficient, because more of the journey happens inside systems before any conversation starts.
Start with structure. Store the attributes shoppers ask about as data fields rather than descriptive text. Then keep one authoritative record for each product, and serve every channel, including AI tools, from that record. Adobe’s benchmark shows product pages are where most retailers fall short, so audit those first.
AI-referred traffic is already measurable and converting well, as Adobe’s data shows. Fully agent-led purchasing is less mature. Both depend on structured product information and reliable availability, so the preparatory work pays back before autonomous purchasing becomes common.
Start with the first source of truth that fails. For discovery problems, that usually means product and availability data: establish one authoritative source, measure how far each channel drifts from it, and fix the pipelines that feed the worst drift first. For service or personalization problems, start with identity, consent, and customer-history data, which is where Loyalty Graph work begins.
Find the gaps stopping AI channels from delivering your customer promises
The AI Readiness Assessment tests your product, price, availability, and customer data against the five promises in this article. You get a ranked list of the gaps most likely to cost you visibility and conversion in AI-led discovery, and the order in which to fix them.
