Your ads are fine. Your product data is the problem.

Product feed optimization is the work of improving the product data you send to shopping channels, rather than the ads or bids that sit on top of it. The feed is the file that tells Google, and increasingly every AI shopping surface, what each product is: its title, description, identifiers, category, price, availability, images, and the relationships between products. Optimization means structuring that data so products are approved, matched to the right queries, and legible to the systems deciding what to show. It is usually the highest-leverage work in a Shopping account because it changes what you are eligible for, where bid changes only change what you pay.

Feed management is the operational half: getting a clean, current file to every channel, handling the sync, and keeping products approved as policies and catalogs change. Feed optimization is the performance half: deciding what goes into the fields so the products win. Most tools sell management, which is genuinely useful plumbing and mostly a solved problem. Very few merchants have anyone doing the optimization, which is why a well-managed feed can still be quietly uncompetitive. We do both, and if you already have a feed tool that handles the plumbing well, we will tell you to keep it.

In most catalogs we look at, the same handful of causes account for nearly all of it. Identifier gaps, where a product carries neither an accurate GTIN nor the combination of MPN and brand that Google accepts in its place. Price or availability in the feed that no longer matches the landing page, which happens most often during a promotion when the feed is pulled on a daily schedule. Images with a watermark, logo, or promotional text overlaid on the product. Missing required attributes for a specific category, which surface only once you sell into that category. Worth knowing what is not on that list: a plain white background and a product filling most of the frame are Google best practices, not requirements, and no background colour will get an item disapproved. Anyone reporting those as violations is inventing one. The real causes are all structural rather than one-off, which is why they recur until the pipeline changes rather than the product.

Google added conversational attributes to the Merchant Center specification in 2026 and states they support discovery on AI-driven surfaces, including AI Mode in Search. Those include structured question and answer pairs and declared relationships between products such as accessories, substitutes, and required parts. Here is the part you will not get from most agencies: Google publishes no ranking or inclusion effect for any of them, for AI surfaces or for Shopping, so treat any specific percentage uplift attributed to a feed attribute with suspicion, including if we ever quote you one. What is sayable without inventing anything is the structural point. A feed carrying none of these attributes serves Shopping perfectly well while giving a conversational surface almost nothing specific to work with, the attributes cost nothing but the work of populating them, and most catalogs have none of them today.

Yes, and this catches almost everyone doing AI-assisted feed work. Google requires titles created using generative AI to be submitted through the structured title attribute rather than the ordinary title attribute, with the digital source type declared as trained algorithmic media. The same pairing exists for descriptions. There is a trap inside the trap: if you submit both the structured version and the plain one, Google uses the plain title and ignores your declared version entirely, so a feed that populates both is not compliant, it is just quietly overwritten. If you are generating titles at catalog scale and have never touched structured_title, that is worth checking this week.

They solve different problems and the honest answer is often both, or sometimes neither. A tool is the right answer when the difficulty is mechanical: many channels, frequent catalog changes, complex transformation rules. It will not tell you that your titles are structured wrong or that your category assignments are costing you eligibility, because those are judgement calls about your catalog and your buyers. If your feed is small, stable, and already well-structured, you may need neither, and we would rather say so than sell you a retainer for maintenance work that does not exist.

Faster than most marketing work, which is one of the reasons we start here. Disapproval fixes restore eligibility as soon as the products are reprocessed, usually within a few days. Title and category restructuring changes which queries you match, so it typically shows within two to four weeks as the channel gathers data against the new structure. Conversational attributes are the exception and we will not pretend otherwise: adoption across AI surfaces is still uneven, so that work is a position taken ahead of the demand rather than a change you will measure next month.