Here's something most brands haven't clocked yet: on ChatGPT Ads, your product feed isn't a nice-to-have, it's the gatekeeper. We sat down with Ioannis Koukoulis, our VP of Marketing Technology, to unpack what "feed management" actually means in this brand-new, still-evolving channel.

Q1: First Things First: What Is Feed Management in the Context of ChatGPT Ads, and Why Does It Matter Now?

Ioannis: This year ChatGPT launched its own advertising platform where ads appear into user conversations with ChatGPT. Ads went live in the US since February 2026, and the self-serve Ads Manager opened to US advertisers in May. Also, ads began serving across 31 European markets last month. Feed management enters the picture here as well, because of product-feed campaigns. It works the same way the product-feed campaigns work for all the other advertising platforms: you connect to a catalog, set filters for which items are eligible, and the platform generates the placements from product names, images and attributes. However, since early this month, shopping placements in ChatGPT are reported to be limited to brands with a connected product feed. That means products that are not part of a product feed are absent from shopping results regardless of ad spend. The same catalog also powers the organic product recommendations users get when they ask what to buy, so one clean feed pays off twice. For most advertisers this is the first channel where feed quality is an eligibility prerequisite rather than a delivery optimisation.

Relevant article: Q&A: ChatGPT Ads for DTC Brands and Retailers: Cutting Through the Hype

Q2: How Does Product Feed Quality Influence Whether ChatGPT Surfaces and Recommends a Brand's Products?

Ioannis: Almost all product offers in ChatGPT originate now from direct merchant feeds. The model is not browsing your PDPs in real time and reassembling them, but it retrieves from structured data it has already ingested.

Which SKU is retrieved and shown by ChatGPT is intent-based, not keyword-based. A conversational request gets decomposed into constraints, so the attributes that decide inclusion are the descriptive ones: category, material, colour, size, gender, age group, condition, dimensions and weight, plus identifiers like GTIN or MPN. But what is also important is the completeness of the product information. This is the differentiator between two otherwise identical products. A row with fifteen populated fields can answer more question types than a row with five, so it enters more candidate sets. Lastly, trust signals also carry some weight: review aggregates, returns policy, seller identity and shipping information all feed the "can I confidently recommend this" judgement.

To sum it up, when product information is thin, the model does not abstain. It generalises. That is a brand risk as much as a visibility risk, because an inferred claim about compatibility or suitability lands on the merchant.

Q3: What Are the Most Common Feed Management Mistakes Brands Make When Preparing for ChatGPT Ads?

Ioannis: I would say that the dominant one is treating the product feed as a direct copy from Google Merchant Center. The attribute logic transfers well, but the delivery mechanics and the validation rules do not. There are some differences in the feed file format, as OpenAI doesn't support all the file formats and all the way Google supports. This makes sense, as OpenAI's merchant center is very new and lacks some functionality.

Secondly, there are some attributes that their content might need to follow slightly different rules.  Lastly, the in the GhatGPT  product feed there are some flag attributes like: "is_eligible_ads", "is_eligible_search" and  "is_elibilbe_checkout" that are very important. These flag values determine which products will be available to appear in searchs, in ads and which are available for the agentic checkout.

Listen to our podcast on this topic: Preparing Product Data for AI-driven and “Agentic” Commerce (ft. Marcel Hollerbach from ProductsUp)

‍Q4: How Should Product Feeds Be Structured Differently for Conversational AI Platforms Compared to Traditional Search or Social Feeds?

Ioannis: The commerce fundamentals are still unchanged: stable IDs, identifiers, price, availability, images, variants, shipping and returns. However, there is the additional layer, then one that contains the descriptive content whose job is to answer questions rather than to match query strings, and that layer is what most catalogs are missing.

More specifically, the product titles should read as a human would say the product out loud: brand name and product type first, then the defining variant. Keyword stacking is a net negative because it obscures what the item actually is. Descriptions should be factual and self-contained, covering use case, materials, compatibility, sizing behavior, care, and even the conditions the product is not suitable for. The product variants need to be explicit rather than collapsed. Also, the structured taxonomy still matters. A consistent category path plus attributes like gender, age group, material, pattern and size system lets the model classify the item across several different query types.

However, the trust and commerce context are now part of the content as well: review count and rating on matching populations, whether returns are accepted and for how long, the policy link, and the seller identity. Also, providing sample questions that are often asked by users, any sort of warnings or age restrictions, or providing the list of other eligible products that are available and buyers also buy. These attributes, till quite recently, were not part of most advertisers' product feeds, but now are making a real difference in the conversational context the products now appear through ChatGPT.

Q5: How Can Marketers Measure Whether Their Feed Management Strategy Is Actually Working for ChatGPT Ads?

Ioannis: For me, the most important aspect is to start by accepting that this is a test-and-learn channel. The bidding algorithm, the reporting interface, and targeting surface have changed several times in 2026, so the measurement plan should assume the platform will move under you. The same applies for the product feed it is important to measure the feed health as you measure other performance KPIs. That means, the advertisers should look at the share of SKUs accepted versus rejected per upload, share of SKUs flagged ads-eligible, attribute completeness by category etc.

Then, when running ad campaigns, don't forget to drill down to product-level reporting. OpenAI exposes product-segmented insights, so you can see which specific items received impressions and clicks, which is the closest thing to an inclusion rate you will get today. Afterwards, for any sort of feed optimizations and enrichments you do, treat them as a controlled test. Enrich one product group, hold a comparable group as control, and measure the change in impressions and inclusion rather than only ROAS, because volume on this channel is still small enough that ROAS alone is noisy. Lastly, keep in mind that an attribution gap is probably expected for ChatGPT Ads. The ad clicks land on the advertiser site and behave normally, but the ChatGPT conversations keep a large share of the discovery journey, so last click will understate the channel. Treat incrementality testing as a Q4 or Q1 2027 exercise once you have enough volume.

---

Unsure how to approach ChatGPT Ads? Reach out to us.

Relevant Insights:

· Case Study: Reddit Ads Case Study: Driving Full-Funnel Growth for a Global Fashion Retailer in Europe

· Article: Market Signals #9: Amazon Enters ChatGPT, Meta Launches an AI Shopper, and Gen Z Turns Out to Be More Loyal Than Anyone Expected

· Report: The Modern Paid Search Audit: What to Look for in an AI-First Google Ads Account

About Crealytics

Crealytics is an award-winning full-funnel digital marketing agency fueling the profitable growth of over 100 well-known B2C and B2B businesses, including ASOS, The Hut Group, Staples and Urban Outfitters. A global company with an inclusive team of 100+ international employees, we operate from our hubs in Berlin, New York, Chicago, London, and Mumbai.