The feed is Google's model of your product range

In Shopping, the feed replaces the keyword list

If you treat the feed as an export file, you leave your keyword research to the shop's interface. According to Google, in Shopping ads it is Merchant Center product data, not keywords, that determines how and where an ad is served. No retailer would take a keyword list for a Search campaign from IT without checking it. With the feed, that check is easily skipped, because it looks like a technical file.

A good feed therefore does two jobs that have little to do with each other. The first I call accuracy: every detail matches the shop and the product. The second is reach: the product is described so that it qualifies for the searches you want to pay for. Most feed checklists throw both into one list of attributes. In practice, though, separating them is the whole point, because the two jobs need different owners and different metrics.

Which attribute does which job

The table sorts the fields by what an error costs. The assignment in the two right-hand columns is my assessment, the consequences in the middle column are Google's own statements.

For serving, this means: required attributes open the door, and the title does most to determine which searches the product fits. According to Google, optional attributes are submitted to help boost performance, they are not a prerequisite. Which fields are required also depends on the product and the target country. Colour, for example, is required for apparel targeted at Germany. If you add a new country or a new product group, you should therefore check the list again.

A deadline is approaching: from 31 January 2027, according to Google, 500 x 500 pixels is the minimum resolution for product images. Merchant Center already flags affected products as a warning under "Needs attention", and Google optimises some of the images that are too small itself. New and optional are video_link for product videos and question_and_answer. I would only tackle them after the products that drive revenue, and I know of no performance data on them.

In my view, the title is where most work pays off: it is explicitly matched against the search, and it is the field you have most freedom to word. Seer Interactive also calls it the single most important attribute. 150 characters are allowed, at most 70 are visible depending on the screen. What customers need in order to choose therefore goes at the front. Google can also move key attributes to the start of the title itself, so the ad does not necessarily show your wording. How to handle this in the shop is covered in the article on product titles for Google Shopping.

I treat the category as a correction: as a rule, you only touch it when Google's own assignment is visibly wrong. Labels do not affect matching against the search, only how campaigns are divided. So they bring no reach, but they are where commercial knowledge such as margin tier or sell-through gets into the account. Which of these are worthwhile is covered in the article on custom labels.

How complete does a feed need to be?

Required attributes must be 100% filled. Beyond that, Google sets no threshold. In my view, a high fill rate as a target is even dangerous, because Google treats wrong values more harshly than empty fields. On GTIN, Google says that when in doubt you should leave it out and not guess a value, and a wrong value leads to disapproval. If you give the team 100% GTIN coverage as a metric, you reward exactly that guessing.

The decision rule I derive from this has three levels:

• Value known and verified: enter it.
• Value exists but is not available: get it from the manufacturer or from purchasing, because leaving it out can cost visibility.
• Value uncertain: leave it empty until it is clarified.

An empty field may cost you performance, a wrong one costs you the product. With price, the stakes are higher still, because there the whole account is at risk. One cause Google names is simply timing differences between shop and feed updates. That is a question for the interface, and the article on wrong prices in Merchant Center goes through it step by step.

The check: two questions per revenue driver

Instead of the fill rate, I recommend a sample weighted by money. For each high-revenue product, ask two questions: would this title fit the searches you want to pay for? Does every detail match the product page?

The effort stays manageable when the range is unevenly distributed. A worked example with round, assumed figures: a shop has 4,000 products, 200 of which bring in most of the revenue. At three minutes per product, checking these 200 takes around ten hours, the entire range 200 hours. Sort your Shopping products by contribution margin if you know it per product, otherwise by revenue.

This list has a blind spot: a product with a poor title may sell little through Shopping precisely because of that title, and never makes it to the top. So put a second list next to it: products that sell well in the shop overall but bring in hardly any revenue through Shopping. That is where you find the cases in which the feed presents the range as weaker than it is.

Who is responsible internally?

My assessment: the feed as a whole belongs to the head of e-commerce. IT is responsible for the pipeline, not the content. IT makes sure that price, stock and identifiers arrive correctly and on time, and treats deviations like an incident. E-commerce decides how products are named and grouped, because that is a product range decision which elsewhere in marketing would be handled as keyword strategy. Whether implementation sits in-house or with an external partner changes nothing here: the responsibility stays within the business.

The acid test is the IT queue. If every title change waits as a ticket for the next sprint, your keyword list only changes every few weeks, even ahead of high-revenue seasons. E-commerce therefore needs direct write access to wherever the titles come from, in the shop or in the PIM. I advise working there rather than overwriting afterwards in the feed, so that shop and ad do not drift apart. What works as a metric for management: the share of revenue drivers that pass both check questions, and the number of disapproved products. Both say more than a fill rate.

Small shops can skip the question of roles

With a small range looked after by one person, the question of roles disappears and the sample becomes a full check. If the master data in the shop is already clean, nothing needs to happen to the feed just because a provider recommends it. Whether the plugin is enough for this is covered in shop plugin or feed tool, and the bigger picture in the guide to Google Shopping and Merchant Center.

Nobody can seriously tell you in advance how much revenue a better title will bring, and that includes tool providers. A good feed ensures that a product qualifies for the right searches. Whether it is bought there and pays off depends on price, offer and margin, and no attribute fixes those. The attribute rules change constantly. These details are as of November 2026.