A shopper asks ChatGPT for a running shoe that fits a narrow foot and can land on their doorstep by Friday. The model has hundreds of SKUs in front of it. Each one has a price. Each one has an in-stock flag. Almost none of them say anything about foot width. Only a handful expose a delivery window at the SKU level. That gap is the whole story: real-time intent data is quietly rewriting what a retail feed is supposed to do.
Key Takeaways
- Shopping agents filter and compare against the specific attribute named in the question, not the SKU as a whole, so a price and availability feed alone leaves most queries unanswerable.
- Retailers need to expose fit, compatibility, delivery windows, return terms and structured review sentiment at the product level, refreshed on a schedule the agent can trust.
- A feed can drive strong Share of Voice across the web while producing almost no Share of Citation inside shopping agents, because the two are measuring different things.
What Is a Shopping Agent Actually Asking For?
Nobody typing “narrow-fit trail shoe under 6,000 that ships this week” wants a category page. That single line has three filters buried in it: fit, a price ceiling, a delivery date. Every one of those has to resolve against structured data before the agent can answer anything. Or it guesses.
ChatGPT pulls from across the open web, so it’ll grab fit and delivery detail from wherever it happens to sit. A review site. A forum thread. The retailer’s own page, if that’s clearest. Perplexity works differently: it cites sources inline, which means it needs an attribute it can point to and name, not a paragraph it has to interpret first. Neither model hands the win to the lowest price automatically. Both hand it to whoever actually answered the filter that was asked.
The Fields That Decide the Answer, Not Just the Sale
Price and stock tell an agent whether something can be bought. That’s it. They say nothing about whether it’s the right one for this person, this question. The fields that settle that sit one layer deeper, and most retail feeds were never built to carry them at the SKU level:
- Fit and sizing specifics, meaning width, not just a size number, plus how a product runs against a standard size.
- Compatibility fields, for anything that has to work with a product the customer already owns.
- Delivery windows by SKU and by location, not one shipping estimate slapped across the whole site.
- Return terms and cost. A free 90-day return changes “which one should I buy” as much as price does.
- Structured review sentiment: what buyers actually said about fit or durability, not a star average with no context.
- Stock depth by location. “In stock” and “one unit, three states away” are not the same answer, even though most feeds treat them as identical.
A price feed answers what something costs. An intent feed answers whether it fits the question that was actually asked.
None of this is exotic. It’s what a good in-store associate would ask before recommending anything, ever. Would a salesperson hand you a shoe without asking about your foot? They wouldn’t. The only difference here is that an agent needs that answer sitting in a field it can parse, not buried in a sentence a human has to read and interpret.
growth in AI brand presence for NewDay USA, tracked weekly across five engines over six months.See the case study
NewDay USA didn’t get that lift from one content push. It came from structured signal, held steady, checked weekly across five engines for six months straight. Retail feeds compound the exact same way. Once an agent finds fit and delivery data reliably attached to a brand’s products, it starts checking that feed first, before it bothers with anyone else’s.
Daily Updates Are a Relic. Agents Don’t Wait
A feed refreshed once a day made sense when the customer journey ended at a search result and a click. It does not survive contact with an agent trying to answer “can this arrive by Friday” in real time. Stock and delivery promises move by the hour. A feed built on daily batches is already wrong by lunch.
There’s a real split here, between training-data presence and retrieval presence. Training-data presence builds slowly: wide coverage across the web that eventually gets absorbed into what a model knows. Retrieval presence is earned technically, through structured, current content the model can pull the instant the question is asked. Delivery windows and stock depth live entirely in that second layer. Stale feed? The agent either answers wrong or skips the retailer entirely and pulls a competitor’s numbers instead.
Share of Voice Won’t Save You. Share of Citation Might
A retailer can spend heavily on paid media, land press everywhere, and still be a ghost inside shopping agents. That gap has a name: Share of Voice versus Share of Citation. Share of Voice counts how often a brand gets mentioned across the web. Share of Citation counts how often a model actually pulls that brand’s data in and names it as the source. A brand can be loud everywhere and still sit at near-zero Share of Citation. Volume of mentions isn’t structured, checkable data. A model can’t cite noise.
Trust, in this context, isn’t one score. It’s a graph of agreement across independent sources. Does the retailer’s own fit data match what reviewers actually said? Does the delivery promise on the product page match what customers report actually happened? Once a model cites a retailer’s data and it holds up under a second check, the next citation gets easier. Feeds that expose consistent, verifiable attributes build that graph over time. Feeds that stop at price and a boolean stock flag never even get a shot at it.
What To Do About It
Start with the actual questions customers type into AI shopping tools for your category. Not the keywords your SEO team has been ranking for; those are a different dataset. Fit, compatibility and delivery timing show up in agent queries far more than they ever did in classic search terms. That gap, between what people ask and what the SEO team optimized for, is usually where the feed is weakest.
Then audit the product feed field by field against those exact questions. If width, delivery window, return cost and structured review sentiment aren’t sitting there as their own fields, price competitiveness buys you nothing. The product still doesn’t get recommended. This work runs alongside your existing SEO and PR effort, not instead of it. The same PIM and content team already maintaining your product data is the team that extends it into these fields. Build the additions once, at the schema level, and every engine querying that feed benefits from the same update. No repeat work per platform.
FAQs
Does a Rich Product Feed Guarantee an AI Agent Will Recommend It?
No. It removes the most common reason an agent skips a product: missing data on the specific attribute the customer asked about. Recommendation still comes down to price, stock, and how the retailer’s data stacks up against competitors on that same attribute.
Is This the Same Work as Structured Data for SEO?
It overlaps, but it goes further. SEO structured data was built for search ranking signals. Shopping agents need attribute-level fields, width, delivery window, per SKU, that most SEO schema implementations never included in the first place.
How Often Does an Intent-Ready Feed Need to Update?
As often as the underlying fact changes. Stock depth and delivery windows shift within hours, so those fields need near-real-time updates. Fit and compatibility data barely move at all, so they don’t need the same treatment.
Which Retailers Should Prioritize This First?
Any retailer where one wrong attribute, wrong fit, wrong compatibility, wrong delivery date, causes a return or kills a sale outright. That’s most of e-commerce, honestly, but it’s sharpest in footwear, apparel, electronics, and anything sold with accessories that have to match.
