Merchant Center AI feed checker
Drop a product feed (CSV, TSV or XML), say which titles, descriptions and images were made with generative AI, and see which rows follow what Google's Merchant Center help pages ask for: the structured_title and structured_description attributes with trained_algorithmic_media, and the IPTC DigitalSourceType trainedAlgorithmicMedia in the image files. It reads what your feed and files declare. It does not detect AI.
Published . Last reviewed .
Free · No signup · Your feed is read inside this browser tab and never uploaded. Image links you mark are fetched by your browser straight from their hosts, not through GotAlt. How to check that for yourself.
Not sure which rules apply to you? Find the AI disclosure rules that may apply. Want to read the markers in one image file? Use the AI image metadata checker.
1. Add your feed
Drop your feed file here
or press Enter to choose a file from your computer
CSV, TSV, a tab-separated .txt file, or an XML feed (RSS 2.0 or Atom), up to 25 MB and 20,000 products. Read in your browser, never uploaded.
A Google Sheets export or the file Merchant Center fetches from your store works. The first line must be the header row, and the checker looks for the columns id, title, description, structured_title, structured_description, image_link, additional_image_link and lifestyle_image_link. It reads files only, not a Merchant API feed.
What the feed says by itself
| In the feed | Rows |
|---|
2. Say what was made with AI
This checker cannot tell which of your content was made with generative AI, and it does not guess. You say which, and it checks the feed against Google's written rules for those rows. Leave a group at "None of them" to skip it.
Titles
Descriptions
Images
Pressing the button below makes your browser request the sampled image links, each from its own host, up to four at a time. Nothing else leaves this page. When the marked products have more links than the sample, the sample is spaced evenly through the feed. Hosts that do not let other sites read their images (CORS) cannot be read this way; you can add those image files yourself afterwards.
Results
What a match means
A match means the row is written the way Google's help pages describe, or that the image file declares the IPTC value they name. It is not a statement that Merchant Center will approve a product, that an image was made a certain way, or that any law is met. A file with no marker proves nothing about how its image was made.
| Content | Marked | Matches | Does not match | Worth a look | Other |
|---|
Findings
Rows
| Feed row | Product ID | Title | Description | Images |
|---|
Image links
| Feed row | Attribute | Image link | Result |
|---|
Choose the original image files. Each file is matched to a link in the feed by its file name, the last part of the link, and is read in this tab. A file you add shows what the original declares, which can differ from what your host delivers.
Then see what your live pages show today
This read the feed and the image links you gave it. The Site Image Check runs automated checks on up to 50 pages of your live site, once, for $49 with no subscription. Its AI disclosure section lists, for the images it opens, the AI provenance markers each file declares (a C2PA manifest, an IPTC DigitalSourceType value, XMP keywords such as contains-synthetic-performer) and whether the page shows AI-label text with the image: alt text, a figcaption or adjacent text. It does not read product feeds, and it never decides whether an image is AI-generated.
Check my site once from $49 See a sample fix sheet (CSV file)
Tagging a whole catalog? The AI Metadata Tag Pack writes the source type and keywords into many images in one batch, with a verify pass and a manifest. It is a paid pack.
What this checker can't tell you
- Which of your content was made with AI. You say which. A title nobody marked is checked only for a statement of its own that says trained_algorithmic_media.
- Whether Merchant Center will approve a product. It compares the file you give it with what Google's help pages say. A rule or another data source in your Merchant Center account can change product data after the file is read, and Google's own diagnostics in your account are the place to see what it decided.
- Whether an image is AI-generated, or whether a marker is true. It reads what a file declares about itself. A file made with AI can carry no marker, and a marker can be added to any file, so a result is never a finding about how an image was made.
- Which version of an image Google reads. Google's pages ask that the tag stays in the original image file, and the image link page says Google fetches the image from your link. This page reads the file your link delivers to a browser, which can differ from what Google's crawler is given, for example a resized copy or a file in another format. The AI image metadata checker compares the versions a host delivers.
- Whether Merchant Center's AI label setting is on. That setting is separate from feed attributes and file metadata. Google says using it does not mean you meet any regulation, and this page does not read it.
- Whether a law applies to you, or whether a feed attribute or tag is enough under one. That depends on facts this tool does not have, and nothing here is legal advice. The AI disclosure laws finder lists rules that may apply, each with its source.
How the attributes are written in a feed
From Google's title and description help pages, read on October 22, 2026. The same two sub-attributes, digital_source_type and content, are used for descriptions with structured_description.
| Case | How it is written |
|---|---|
| Text feed, title written with AI | trained_algorithmic_media:"Stride & Conquer: Trail Shoe" The type, a colon, then the text. The text is in double quotes here because it contains a colon. |
| Text feed, title not written with AI | :Trail Shoe, Blue An empty type means default. You can also use the plain title column. |
| XML feed, title written with AI | <g:structured_title><g:digital_source_type>trained_algorithmic_media</g:digital_source_type><g:content>Trail Shoe</g:content></g:structured_title> |
What Google's pages say
Each row is a Merchant Center help page, read on October 22, 2026. Google can change them, so check the current text for your account. None of this is legal advice.
| Page | What it says |
|---|---|
| AI-generated content | Product data created with generative AI must be identified as such. Images must carry the IPTC DigitalSourceType TrainedAlgorithmicMedia and the tag must not be removed; the page lists TrainedAlgorithmicMedia, CompositeSynthetic and AlgorithmicMedia as the IPTC values to preserve. Titles go in structured_title and descriptions in structured_description, each with digital_source_type trained_algorithmic_media and the text in content. The page shows no date. |
| Preserving metadata tags for AI-generated images | Dated February 2024. Asks you to keep, in the original image file, the metadata tags that mark an image as made with generative AI, and says all AI-generated images must contain the IPTC DigitalSourceType trainedAlgorithmicMedia tag. It applies to image_link, additional_image_link and lifestyle_image_link. |
| Title and structured title | The digital_source_type sub-attribute is optional and takes default or trained_algorithmic_media; an empty value means default. The content sub-attribute is required and takes 1 to 150 characters. In a text feed the two are written with a colon between them, and content that contains a colon goes in double quotes. If a product has both title and structured_title, Google uses only title. |
| Description and structured description | The same two sub-attributes, with content of 1 to 5,000 characters. The page says a wrongly formatted value can look like it has more sub-attributes than allowed and be disapproved with the error "Too many sub-attributes for the attribute". If both description and structured_description are given, Google uses only description. |
| Image link and additional image link | The image link page repeats the AI image metadata rule, with the IPTC tag given as an example, and says Google fetches the image from your link, so Googlebot and Googlebot-image must be allowed in robots.txt. Both pages list JPEG, WebP, PNG, GIF, BMP and TIFF as supported formats. The additional image link page allows up to 10 additional images, separated by commas in a text feed. This checker opens JPEG, PNG, WebP, GIF, AVIF and HEIC files, so a BMP or TIFF link is listed as not read. |
| AI content label settings and disclosures | A separate setting, in Merchant Center under Manage images, for labeling an image or video asset as created or edited with AI. Google says using it does not mean you meet any regulation and that a label can affect where the content appears. This checker does not read it. |
Questions
Is my feed uploaded?
No. A dropped or pasted feed is read in this browser tab by script files you can read, feed-checker-ui.js, feed-check.js, provenance.js and ai-metadata-view.js, and no server does the reading. When you press Check the feed, your browser requests the image links of the products you marked, up to the sample size you chose, straight from each image's host, so that host sees the request as it would if you opened the link. GotAlt does not. To check for yourself, open your browser's developer tools, switch to the Network tab and drop a file: apart from this page's own files nothing is requested until you press the button.
Can it tell which of my titles were written by AI?
No. It does not detect AI in text or in images. You mark what was made with generative AI, and it checks the feed against what Google's help pages ask for those rows. A row you did not mark is only flagged when the feed itself says trained_algorithmic_media, so you can check your marking.
What does "No trainedAlgorithmicMedia found" mean?
The image file this link delivered to your browser has no IPTC DigitalSourceType trainedAlgorithmicMedia in its XMP. Google's pages ask that AI-generated product images carry that tag. It says nothing about how the image was made: many tools write no marker, and many hosts and image services remove markers from resized copies. If you have the original file, add it under the results to see what the original declares, or use the AI image metadata writer to add the value without re-encoding the image.
Why does it say the tag is only in the WebP version?
Some hosts return a different file depending on what the browser asks for. Across 13 stores, GotAlt measured Shopify's image CDN on October 21, 2026: XMP appeared only in the WebP version of an image, in 12 of the 12 images that had XMP anywhere, and never in the JPEG or PNG version. When a Shopify link shows no tag, this checker asks once more for the WebP version and tells you if the tag is there. Google's pages do not say which version its crawler is given.
Why could my image links not be read?
The usual reason is that the host does not let other sites read its images (CORS), so a browser cannot fetch them from this page. The link may also be wrong, use http, point at a BMP or TIFF file, or be blocked by a content blocker. Download the images, then add the files below the results; each is matched to a link by file name.
Does a match mean Google will accept the product, or that a law is met?
No. A match means the row is written the way Google's pages describe, or that the image file declares the IPTC value they name. Whether a product is approved is Google's decision, and whether a law applies to you, and whether a feed attribute or tag is enough under it, depends on facts this tool does not have. See the platform rules for AI content and the AI disclosure law pages, which cite their sources and show the date each was checked. This is general information, not legal advice.
Which feed formats does it read?
Comma-, tab-, pipe- and semicolon-separated text with a header row, and XML feeds in RSS 2.0 or Atom form, where each product is an item, entry or product element. It reads up to 25 MB and 20,000 products. A header such as "Structured title [structured_title]" or "g:structured_title" is understood. A feed fetched through the Merchant API is not read; export a file instead.
Related guides
- AI image metadata checker: read what a file declares and what survives your store.
- AI image metadata writer: add a source type and keywords without re-encoding the image.
- AI Metadata Tag Pack: tag a whole catalog in one batch, with a verify pass, ZIP files and a manifest (paid).
- AI image metadata: what C2PA, IPTC and XMP keywords are and how they travel.
- AI disclosure rules by platform: Google, YouTube, TikTok, Meta, Amazon and Etsy, with sources and dates.
- Which AI disclosure rules apply?: rules that may apply to you, each with its source.
- Accessible AI label generator: wording and markup for a visible label.
A feed is one place; your pages are another
The free scan runs 16 common checks on up to three pages of your site in under a minute: missing alt text, form labels, headings, page language, link text and more.
Scan my site for freeFound this useful? Link to it
A link from your site, newsletter or resources page helps other people find this free tool. Copy it here: