Sixty percent of consumers now use AI tools at least occasionally while shopping, according to research Athos Commerce ran with Drapers.¹ So whose job is it to know what those tools say about your products? Gary Lombardo, CMO at Athos Commerce, was asked on The Agile Brand podcast what breaks first when product discovery moves somewhere a brand cannot observe. He named the reporting structure, before any system.²

Most teams respond by appointing someone to own AI search visibility, which answers who is responsible without changing what that person can see. They inherit a product catalog assembled one channel at a time, so they cannot answer the one question the job exists for: what does the AI say about our products?

The job nobody has been given

Marketing teams are organized around channels they can own and measure. At a large retailer that means paid media and the website each carrying a budget line and a named owner, with email and social close behind. At a 60-person brand it means two or three people splitting the same list between them, and a founder still in the weekly merchandising call. The split holds at either size, and it worked as long as the list of places a shopper could meet a product stayed short enough to divide up.

When a shopper asks an AI assistant which running shoe suits their gait, nobody’s objectives cover the answer. No tag fires, no dashboard row updates, and no one has an obvious claim on the outcome. What an assistant tells shoppers about your products sits outside every existing remit, so it goes unexamined. Smaller teams are not spared this. A brand where three people cover nine jobs has the same vacancy as a brand with four channel directors, and less slack to absorb it.

Blended reporting also conceals how unevenly that exposure falls. Thirty-nine percent of consumers say they would complete a purchase directly through an AI platform, and that figure reaches 50% among Gen Z shoppers while falling to 21% among Gen X.³ A brand that reads only the average will conclude it has a year or two in hand. If most of your customers are under 30, software you have never seen is already evaluating your products.

The org chart is a picture of the tool stack

Reorganizing alone fails for a reason that sits one layer below the reporting lines. Teams did not only organize around ownable channels, but they also bought their tooling the same way. Site search came from one vendor, product feed management from another, personalization from a third, each procured by whichever team owned that channel at the time and integrated only as far as that channel required.

Draw the org chart and the tool stack side by side, and you get the same diagram twice. Fragmented ownership and fragmented product data are one condition viewed from two angles, which is why a new title changes little on its own. Whoever takes the job inherits several systems that disagree about the same product record, and their first honest answer to “what does the AI say about us” has to be that it depends on which system you ask.

Lombardo traces the symptom back to that same source. The visible failure looks like a copywriting problem, a wrong price on a marketplace or a missing size on a social post, but the cause sits underneath the copy.

Most brands, he argues, run a product information management system that is 80% right, a product feed file that someone patches by hand for one channel, and a merchandiser’s spreadsheet more current than either. No copywriter can reconcile those three, and no owner can answer for a claim that changes depending on which system a machine reads.

The owner needs one product record to point at

Shoppers notice when a product record disagrees with itself, even when the brand does not. Only 14% of shoppers say product details always match when they move from social media or a marketplace to a brand’s own website. A further 36% say details often match, and the largest group, 40%, say they only sometimes do.⁴

AI systems read that inconsistency directly. When a model compares two jackets, it does not weigh your brand story against a competitor’s – it reads structured attributes and reconciles them across sources. Blank or conflicting fields become guesses, or grounds for leaving a product out of the answer altogether. The brand never sees the omission, because there is no impression to report on a recommendation that was never made.

Correcting the underlying product data also pays off on a timescale most marketing programs cannot match. Accent Group tightened product feed accuracy and data consistency across its brands and recorded a 60% revenue increase within eight weeks.⁵ Eight weeks is inside a single quarter, which makes this one of the few discovery investments a team can start and prove within one planning cycle.

Framing the problem this way settles the ownership argument that stalls most teams. Treated as a content problem, it belongs to marketing, and marketing cannot fix it. Treated as a product data problem, it becomes a joint mandate across marketing, ecommerce operations, and technology, with one team accountable for the product feed pipeline itself rather than the words sitting on top of it.

What the owner does in the first ninety days

The first quarter’s work is diagnostic, and every step below depends on reaching the product data underneath.

1. Audit presence against the product record you publish. Prompt the major AI assistants with the queries your customers ask, then check each answer against the product data you are syndicating. The audit is useful only when you can trace a wrong answer to the attribute that produced it and correct it at the source. Run it against six disconnected systems, and you generate a list of complaints nobody can act on.
2. Instrument the product feed itself. Attribution cannot observe an AI assistant, so stop rebuilding it and measure what stays visible. Attribute completeness, product feed accuracy, and channel-by-channel rejection rates are all measurable today, and they move before revenue does. Pair them with branded and direct search trends, where shoppers arriving already knowing a product name or SKU suggest something recommended you upstream.
3. Decide where approval stops being required, before an incident decides for you. Software that flags product feed errors and drafts attribute fixes for a merchandiser to approve is assistance. Software that reprices, reranks, or suppresses products across live channels with no review step has taken the decision itself. Early Settler saves more than ten hours a week through automated merchandising, and Aje saves two to three days a week, with a person still approving what ships live.⁶

Accountability does not transfer because a machine made the call. It stays with whoever owns the channel’s commercial outcome, which usually means the merchandising or ecommerce lead rather than the vendor or the model. A team that cannot explain why a product received a particular treatment has already crossed that line without deciding to.

The ownership question is about to get harder

Discoverability has a short shelf life as the industry’s organizing question. Lombardo’s read on the next twelve months is that the industry conversation moves from optimizing to be found toward optimizing to be transacted through, as a meaningful share of purchases start completing inside the AI platform itself.

That opens a harder set of problems than visibility ever posed. Someone has to work out who owns the customer relationship when a platform sits in the middle, how returns and loyalty function across that handoff, and how a brand protects both its margin and its own customer data when the platform holds the transaction. No one has settled answers, and all land on the same desk. Whoever owns AI visibility will inherit transaction ownership faster than any org chart can adapt.

Brands that already treat product data as the thing under evaluation will move first, because the work that makes a product legible to an assistant is the same work that makes it legible to an agent completing a purchase. David Jones ran a 30-day test on optimized titles and attributes within a single product segment and recorded a 66% performance uplift, with no creative changes involved.⁷

Answering the ownership question is cheaper while it is still only about being found. Once the transaction moves inside the platform, the same unanswered question costs margin.

Sources and further reading

1. Drapers and Athos Commerce. “Connected Consumer 2026.” Drapers, June 2026. https://www.drapersonline.com/insight/drapers-bespoke/connected-consumer-2026. Survey of 2,000 UK fashion shoppers aged 18 to 60.
2. Kihlström, Greg. “Athos Commerce CMO Gary Lombardo on Why Only 14% of Shoppers See One Consistent Brand.” The Agile Brand Guide, September 13, 2026. https://agilebrandguide.com/athos-commerce-cmo-gary-lombardo-on-why-only-14-of-shoppers-see-one-consistent-brand/.
3. Drapers and Athos Commerce. “Connected Consumer 2026.” Drapers, June 2026. https://www.drapersonline.com/insight/drapers-bespoke/connected-consumer-2026. Survey of 2,000 UK fashion shoppers aged 18 to 60.
4. Drapers and Athos Commerce. “Connected Consumer 2026.” Drapers, June 2026. https://www.drapersonline.com/insight/drapers-bespoke/connected-consumer-2026. Survey of 2,000 UK fashion shoppers aged 18 to 60.
5. Athos Commerce. “Accent Group.” Case study. https://athoscommerce.com/case-studies/accent-group/.
6. Athos Commerce. “Early Settler.” Case study. https://athoscommerce.com/case-studies/early-settler/; Athos Commerce. “Aje.” Case study. https://athoscommerce.com/case-studies/aje/.
7. Athos Commerce. “David Jones.” Case study. https://athoscommerce.com/case-studies/david-jones/. Figure reflects a 30-day test within a single product segment.