A product has an invalid GTIN. Google shows it with limited performance. Amazon suppresses the listing. And ChatGPT does not recommend it, because it is missing from Google’s shopping results. A single data error costs you revenue in three channels at once.
Most articles on product data quality come from PIM vendors. They explain definitions and end with their own tool. This guide is vendor-neutral. It is written for managing directors and heads of e-commerce at online stores and D2C brands in Europe, with a focus on the DACH region. You will learn what poor product data actually costs, how errors affect each channel and how to close the biggest gaps in 30 days.
The short answer
What is product data quality? Product data quality describes how complete, accurate, consistent and up to date the data about your products is. It covers every channel: your store, Google, marketplaces and AI assistants.
What does poor product data cost? It costs in three places. First, returns: according to KPMG and EHI, 60.4 percent of online shoppers in Germany name an inaccurate description or images that differ from the product as a reason for returns (December 2025). Second, visibility: faulty items are limited on Google and suppressed on marketplaces. Third, internal time spent on corrections.
What matters for AI search? Above all a clean Google Shopping feed and visible, complete text on the product page. Structured data and llms.txt, on the other hand, do little to get you mentioned in AI answers.
What is product data quality?
Product data is all the information that describes a product. That includes master data such as GTIN, brand, price and weight. It also includes attributes such as material, size and color, plus copy, images, mandatory information and variants.
Good product data meets six requirements. You will find the first four in almost every guide. The last two have become important because of AI search.
| Dimension | Question | Typical error |
|---|---|---|
| Completeness | Are all mandatory and buying attributes filled for each channel? | material missing, no GTIN, no size |
| Accuracy | Does the data match the real product? | wrong dimension, wrong color, outdated image |
| Consistency | Do store, feed and marketplace say the same thing? | feed says “in stock”, store shows “sold out” |
| Timeliness | Are price, stock and delivery time current every day? | a price change only reaches the feed the next day |
| Richness | Does the data answer buyers' questions? | no use case, no compatibility, no care instructions |
| AI readability | Can machines read the information unambiguously? | dimensions only in the image, specs only in a PDF, variants as plain text |
A study by Aalen University in Germany shows how retailers weigh these: 94 percent consider accuracy important, 88 percent completeness and 81 percent timeliness. But only 30 percent had fully defined how product data is maintained. The study dates from 2017. There are no newer figures on this for the DACH region.
What poor product data costs: the figures
Many figures on poor product data circulate online. Many of them have no traceable source. The table below only shows figures that can be traced back to the original study.
| Figure | Source and year | Sample | Vendor study |
|---|---|---|---|
| 60.4% name an inaccurate description or differing images as a reason for returns | KPMG with EHI, December 2025 | 500 online shoppers, Germany | no |
| 36.6% stop ordering from a retailer after a bad return experience | KPMG with EHI, December 2025 | 500 online shoppers, Germany | no |
| around 560 million return parcels in Germany, almost one in four parcels | University of Bamberg (Asdecker), forecast 2026 | projection | no |
| more than a quarter of retailers do not know their own cost per return | EHI Returns Management 2026 | retailer survey | no |
| 31% research products with AI chatbots, 35% of them more often than with search engines | ECC Club Cologne (IFH), October 2025 | 1,000 respondents, Germany | no |
| 44% abandoned a purchase due to insufficient product information | Syndigo, June 2025 | over 8,500 respondents in 6 countries incl. Germany | yes |
| 65% of Germans are less likely to consider a brand with incomplete product content | Syndigo, June 2025 | German subsample | yes |
| 54% abandoned a purchase due to contradictory product information | Salsify, 2025 | 1,910 respondents, US and UK | yes |
The most important figure for the German market is the first one. It comes from an independent survey and directly affects your costs. Inaccurate descriptions and differing images rank fourth among return reasons. Only quality defects, wrong size and damage rank higher. The wrong size is often a data problem too: the size chart is missing or does not fit the product.
These figures circulate but are not backed up
Some figures appear in many articles. Do not use them for your business case:
- “Poor data costs companies $12.9 million a year” (Gartner): The figure refers to enterprise data in general, not product data. It is based on self-reported data from 2020. Versions with $9.7 million and $14.2 million also circulate.
- “GTINs bring up to 40 percent more clicks” (Google): The claim dates from around 2016. It no longer appears on Google’s current help pages.
- “88 percent abandon”, “conversion up 40 percent”: Values like these often appear in guides without any source.
- “Enhanced content increases conversion by up to 90 percent”: Vendor marketing claims without a published test design.
One data core, three shelves
Your product data now ends up on three “shelves”: on Google, on marketplaces and in AI assistants. All three essentially read the same fields: GTIN, brand, title, variants, price, stock, shipping, returns and attributes. An error in these fields therefore almost always affects several channels at once.
| Data error | Google Shopping | Marketplaces | AI search |
|---|---|---|---|
| GTIN missing or invalid | disapproval or limited performance | listing rejected or suppressed | product missing from Google's results and usually from ChatGPT too |
| Mandatory attribute missing (e.g. size, color, material) | limited performance, mandatory for apparel | listing invalid or suppressed | product does not match the user's question and is not mentioned |
| Price or stock differs from the product page | items disapproved as a precaution, account suspension if repeated | cancellations, worse account health | wrong recommendation, loss of trust |
| Image too small, placeholder or promotional overlay | disapproval; from 31 Jan 2027 at least 500 × 500 pixels | rejected at validation | weaker presentation |
| GPSR information missing | notice in Merchant Center, legal risk on the product page | listing blocked (Amazon, OTTO) | no direct effect, but the risk remains |
| Variants grouped incorrectly | duplicates, wrong variant shown | wrong catalog match | wrong size or color recommended |
Google: disapproval, “limited performance”, account suspension
Google checks every item in Merchant Center. If mandatory information is missing, the item is disapproved. If attributes such as brand or GTIN are missing, Google often still shows the product, but with “limited performance”. That means fewer impressions and clicks. If price or availability in the feed differ from the product page, Google disapproves items as a precaution. Account-level violations usually lead to suspension after a 28-day warning period.
Two dates matter now:
- Since 18 August 2026 the old Content API has been retired. Store plugins and feed tools that have not moved to the new Merchant API lose their connection. More on this in the Google Merchant Center feed guide.
- From 31 January 2027 all product images must be at least 500 × 500 pixels. Merchant Center already warns about images that are too small. Google does not want upscaled images.
Marketplaces: the GTIN is the key
On almost every marketplace, the GTIN (in Europe usually the EAN) matches your offer to the right product in the catalog. If it is invalid, not registered with GS1 or used twice, the listing is rejected or suppressed.
- Amazon checks GTINs against the GS1 database. Invalid numbers can lead to suppressed listings. Missing mandatory attributes, a non-compliant main image or overly long titles do too. Amazon keeps introducing new mandatory fields. Existing listings can then be suppressed after the fact.
- Kaufland does not display products with missing mandatory attributes. Which attributes are mandatory depends on the category. Kaufland itself states that better product pages can reduce returns and complaints.
- OTTO Market requires mandatory attributes per category and the GPSR manufacturer information for every single item.
- Zalando requires a valid 13-digit EAN for every item and maps sizes to its own size system.
- Galaxus ignores records without a GTIN. Without certain mandatory attributes per product type, a product does not go live.
AI search: your Google feed is your ChatGPT feed
There is one key finding for AI search from 2026. Peec AI and the agency Precis analyzed more than one million shopping queries in ChatGPT. The result: every product that ChatGPT Shopping showed was also among the top 40 organic Google Shopping results. Bing explained only 11 percent. ChatGPT re-ranks the products, but the selection effectively comes from Google.
The conclusion: if you are missing or weak on Google Shopping, you are usually missing from ChatGPT too. A clean Merchant Center feed is therefore the most important lever for AI search. How to supply ChatGPT Shopping directly as well is covered in the article on ChatGPT Shopping for online retailers.
The study shows two more things:
- Products rated below three stars on reviewed attributes practically never appear.
- ChatGPT adds words such as “best”, “review”, “2026” or “vs” to queries. If the searched terms are missing from your feed and product page, your chances of being found drop.
The shelf in AI search is also much smaller. According to an analysis by the vendor Productrise, Google AI Mode shows products for only 23 percent of shopping queries. When it does, it shows 4.3 listings on average instead of 22.5 in classic search. If you are not among the few products mentioned, you are not seen.
What really matters for AI search and what is overrated
Many agencies promise quick wins in AI search through structured data or an llms.txt file. The evidence says otherwise.
| Measure | Effect | Evidence |
|---|---|---|
| Complete Merchant Center feed | high: basis for Google AI Mode, Gemini and, according to the study, ChatGPT | Google AI guide 07/2026, Peec AI and Precis 06/2026 |
| Visible, complete text on the product page | high: AI mainly reads visible content | searchVIU 10/2025: no AI system read a price that was only in the code |
| Same data in feed, product page and markup | high: contradictions lead to disapprovals and wrong answers | Google Merchant Center policies |
| New Merchant Center attributes (Q&A, documents, related products) | likely high: introduced specifically for AI answers | Google Marketing Live 05/2026 |
| Ratings for individual attributes | medium to high: poorly rated products drop out | Peec AI and Precis 06/2026 |
| Adding structured data (schema.org) | low for AI citations, important for rich results | Ahrefs 2026: no measurable effect on AI citations |
| llms.txt | minimal: Google does not use the file | Google AI guide 07/2026 |
Google states clearly in its AI optimization guide from July 2026: structured data is not required for AI features, and there is no special schema for them. Google does, however, recommend Merchant Center feeds so that products appear in AI answers. In 2026, Ahrefs compared almost 1,900 pages with newly added structured data against 4,000 control pages. The result: AI citations did not rise measurably.
Structured data is still worth having. It adds stars, prices and delivery information to regular search results. It is just not a miracle cure for AI search.
For context: According to ECC Cologne, 31 percent of Germans already use AI chatbots for product research. But they hardly want AI to buy for them: only 9 percent would hand over a purchase including payment. Checkout directly in Google AI Mode and Gemini is so far only available in the US. For European retailers in 2026, the goal is visibility during research, not sales inside the chat.
Product data as a legal requirement: 2026 and 2027
Product data quality is no longer just a revenue question. Several EU and German rules prescribe which information must appear on the product page. If it is missing, retailers in Germany face cease-and-desist letters (“Abmahnungen”), and marketplaces block listings.
| Rule | What must be in the product data | Applies since or from | Risk |
|---|---|---|---|
| General Product Safety Regulation (GPSR) and new German Product Safety Act | manufacturer with address and email, responsible person in the EU, product identifier, warnings and safety information | 13 Dec 2024; new ProdSG since 19 Feb 2026 | cease-and-desist letters, blocked listings on Amazon and OTTO |
| Ban on generic environmental claims (EmpCo, implemented in German unfair competition law) | no terms like “eco-friendly” or “climate neutral” without proof, labels only with certification | 27 Sep 2026, also for existing products | cease-and-desist letters |
| German Price Indication Ordinance (unit price) | unit price visible at a glance together with the total price | in force | one of the most common reasons for cease-and-desist letters |
| EU textile labelling | fiber composition before purchase, only official fiber names | in force | cease-and-desist letters |
| Energy label | label or efficiency arrow close to the price, link to the product information sheet | in force | cease-and-desist letters |
| European Accessibility Act (German BFSG) | accessible product information in B2C stores | 28 Jun 2025 | fines up to €100,000 |
| Digital Product Passport (ESPR) | product passport with data on materials, repair and recycling | battery passport from 18 Feb 2027, textiles expected from around 2029 | no market access |
Two points are often confused. The EU’s planned Green Claims Directive is on hold. The ban on generic environmental claims has nevertheless applied since 27 September 2026. It comes from a different directive (EmpCo) and also covers copy in feeds and on marketplaces.
What does your product data cost? The calculator
The calculator estimates three cost blocks per year:
- Returns caused by data errors: Revenue divided by average order value gives the number of orders. From that we take the return rate and the share of returns due to description or image. Multiplied by the cost per return, this gives the return costs.
- Lost revenue on feed channels: Revenue via Google, marketplaces and AI search multiplied by the share of disapproved or limited items.
- Internal time: Hours per week spent on corrections times 46 working weeks times the hourly rate.
The default values are an example for a store with €5 million in revenue. Enter your own values. You will find the share of returns due to description or image in your return reason report. The share of faulty items is shown in the Merchant Center diagnostics and in marketplace error reports.
Sum of return costs, lost revenue and internal time
- Return costs caused by data errors
- –
- Lost revenue on feed channels
- –
- Internal time
- –
- Returns caused by data errors
- –
- Share of revenue
- –
Your inputs and the result are copied into the contact form.
Rough estimate. Not included: abandoned purchases due to missing information, legal risks and loss in value of returned goods. Lost revenue is revenue, not profit.
The calculator is deliberately conservative. It works with unit counts, not revenue, and leaves out three items: purchases abandoned because of missing information, the loss in value of returned goods and legal costs. The real costs are therefore likely to be higher.
More on calculating return costs in the article on the e-commerce return rate.
Where poor product data comes from in mid-sized companies
In projects with retailers between €1 million and €50 million in revenue, I see the same causes again and again:
- The ERP is the source. ERP systems are built for purchasing, warehousing and accounting. Fields for buying attributes, copy and images are often missing. Whatever is not in the ERP is then missing in every channel.
- Supplier data is taken over unchecked. It arrives as Excel, PDF or BMEcat, each supplier in a different format.
- Excel is the hidden PIM. Several versions circulate, nobody checks the values.
- Nobody is responsible. Purchasing maintains master data, the e-commerce team the copy, marketing the images. Nobody feels responsible for overall quality. According to a study by the German Society for Information and Data Quality, almost one in three companies has no one responsible for data quality.
- Variants are set up incorrectly. Size and color are set up as separate products, or a variant has no EAN.
- Error reports reach nobody. The marketplace connector reports errors, but nobody reads them.
- New mandatory fields arrive faster than new processes. GPSR, energy labels and, from 2027, the image size on Google.
Measuring product data quality: score and KPIs
What you don’t measure doesn’t improve. A simple score from 0 to 100 is enough to start with. Rate your 50 to 100 best-selling products on the six dimensions.
| Dimension | Fashion and shoes | Tech and electronics | Home, garden, furniture | Checks |
|---|---|---|---|---|
| Completeness | 25% | 25% | 25% | mandatory attributes per channel filled, GTIN per variant |
| Accuracy | 20% | 20% | 20% | spot check against the goods or manufacturer data |
| Consistency | 15% | 15% | 15% | price, stock, title identical in store, feed and marketplace |
| Timeliness | 10% | 15% | 10% | price and stock updated at least daily |
| Richness | 20% | 10% | 20% | size chart, material, care, use case, dimensions |
| AI readability | 10% | 15% | 10% | specs as text, highlights, questions and answers |
Add five KPIs that you should review every week:
- Completeness per channel: share of products with all mandatory attributes of the channel filled.
- Feed error rate: share of disapproved or limited items in Merchant Center.
- Suppressed listings per marketplace: for example from Amazon’s listing quality section or Kaufland’s “invalid items” report.
- Time to listing: days from goods receipt to a complete listing in all channels.
- Return rate “not as described”: ideally per category.
Improving product data: the 30-day plan
You don’t have to improve every product at once. Start where the revenue is. For most retailers, 10 to 20 percent of products generate most of the revenue.
Week 1: Take stock
- select the top 100 products by revenue over the last twelve months
- export Merchant Center diagnostics
- export marketplace error reports
- analyze return reasons per product
Week 2: Mandatory first
- fix invalid or missing GTINs
- resolve disapproved items on Google and suppressed listings on marketplaces
- check GPSR information, unit prices and other mandatory information
- search product copy for generic environmental claims
Week 3: Consistent and complete
- align price and stock between store, feed and marketplaces
- add missing mandatory attributes per category
- replace images below 500 × 500 pixels
- improve size charts and dimensions for products with high return rates
Week 4: Ready for AI search and clean for good
- maintain product highlights and product details in Merchant Center
- add the ten most common customer questions per top product as Q&A
- assign owners per data group
- introduce a check before every publication
- report the five KPIs weekly
ERP, feed tool or PIM: what do you need?
Not every retailer needs a PIM system. The decision depends on assortment, channels and languages.
| Situation | Usually enough | Why |
|---|---|---|
| One store, Google, a few hundred products, one language | store system plus Merchant Center rules | little complexity, maintenance in the store is manageable |
| Store plus 2 to 4 marketplaces, up to a few thousand products | store or ERP plus feed tool | the feed tool translates the data per channel and shows errors |
| Many channels, several languages, many suppliers or thousands of variants | PIM system as the central source | without a central source, channels drift apart |
More important than the tool is the rule of which system owns which data. Prices and stock usually come from the ERP. Copy, attributes and images belong in the store or PIM. When a PIM pays off is covered in detail in the article on PIM for Shopify. A short definition is in the glossary under PIM.
Checklist: how good is your product data?
FAQ
What is product data quality?
Product data quality describes how complete, accurate, consistent and up to date the information about your products is. Richness and AI readability come on top. Good product data is identical in store, Google feed, marketplaces and AI assistants and answers buyers’ questions.
What does poor product data cost?
Poor product data costs in three places: through returns, through lower visibility on Google and marketplaces, and through internal correction work. According to KPMG and EHI, 60.4 percent of German online shoppers name an inaccurate description or differing images as a reason for returns. The calculator in this article estimates the amount for your store.
How do I measure the quality of my product data?
Rate your best-selling products on six dimensions: completeness, accuracy, consistency, timeliness, richness and AI readability. Add five KPIs: completeness per channel, feed error rate, suppressed listings, time to listing and returns due to wrong descriptions.
How do I get my products into ChatGPT?
The most important step is a clean Google Shopping feed. According to a 2026 study by Peec AI and Precis, practically all products in ChatGPT Shopping come from the top 40 Google Shopping results. You can also supply a product feed directly to OpenAI. Shopify merchants are connected automatically via Agentic Storefronts.
Does structured data help with AI search?
Hardly. Google states that structured data is not required for AI features. A 2026 study by Ahrefs found no measurable effect on AI citations. Structured data remains important for stars, prices and delivery information in regular search results.
Why is my Amazon listing suppressed?
Common reasons are missing mandatory attributes, a main image that breaks the rules, overly long titles, an invalid GTIN or missing GPSR information. Amazon also regularly introduces new mandatory fields. You will find the reason in Seller Central under inventory and listing quality.
Do I need a PIM system?
Not always. With one store, a few hundred products and one language, the store system and feed rules are usually enough. With several marketplaces, a feed tool helps. A PIM pays off with many channels, several languages, many suppliers or thousands of variants.
Which product data is legally required in the EU?
For many retailers this includes GPSR information on manufacturer and safety, the unit price in Germany, fiber composition for textiles and the energy label for electrical appliances. Since 27 September 2026, generic environmental claims without proof are also banned. If in doubt, have a lawyer review it.
How often should I check my product data?
Price and stock should be synchronized across all channels at least daily. Review error reports from Google and marketplaces weekly. A full assessment with a score is worthwhile once a quarter and before every season.
Conclusion: product data is revenue, not master data maintenance
Poor product data now costs you in three places at once: on Google, on marketplaces and in AI search. All three channels essentially read the same data. An error therefore almost always has multiple effects.
The good news: the biggest levers are well known. Valid GTINs, complete mandatory attributes, identical data in all channels and a clean Google feed. For AI search, add visible copy and answered customer questions. Start with your best-selling products. Define who is responsible for which data. Then results often show within a few weeks in fewer returns and more visibility.
Sources
- KPMG with EHI: Returns as an opportunity, press release (11 Dec 2025, German)
- Logistik Heute: 2026 reaches a record 560 million returns (30 Sep 2026, German)
- Retail News: EHI study Returns Management 2026 (German)
- IFH Cologne / ECC Club: The next stage of e-commerce (10/2025, PDF, German)
- Syndigo: State of Product Experience 2025
- Salsify: Consumer Research Report 2025
- SDZeCOM and Aalen University: ProDa study 2017 (PDF, German)
- CIO.de: Better data quality is often lip service (DGIQ study, German)
- Google Merchant Center: Policies and enforcement
- Google Merchant Center: Unique product identifiers
- Google Merchant Center: 2026 product data specification update
- Google: Content API for Shopping sunset
- Google Search Central: Optimizing your website for generative AI features
- Google: Shopping updates from Google Marketing Live 2026
- Productsup: Six conversational attributes in Merchant Center
- Precis: How ChatGPT shops (study with Peec AI, 06/2026)
- PPC Land: AI Mode cuts Google Shopping listings by 95% (Productrise)
- Ahrefs: Does schema help AI citations?
- OpenAI: Agentic Commerce, key concepts
- Shopify: Agentic commerce momentum (24 Mar 2026)
- Kaufland Global Marketplace: Seller University, product data
- Zalando Partner: Products onboarding
- Händlerbund: GPSR on OTTO (German)
- IHK: Misuse of GS1 number systems (PDF, German)
- EUR-Lex: Regulation (EU) 2023/988 (GPSR)
- Gleiss Lutz: Implementation of the EmpCo Directive in German law
- IHK Munich: Green claims (German)
- WBS Legal: Federal Court of Justice on unit prices (German)
- IT-Recht Kanzlei: Textile labelling regulation (German)
- German law: BFSG § 11
- European Commission: Digital Product Passport
