Every product photo you take carries a hidden passenger: a block of metadata recording the camera model, the exposure settings, the date, often the GPS coordinates of the spot where you pressed the shutter, and a small integer that decides whether your photo appears upright or lying on its side.
Most sellers never think about it. Then a listing shows a sideways photo that looked fine on the desktop, a Google Merchant Center feed gets disapproved for an unlabelled AI image, or a file that should be 300KB arrives at 480KB for no visible reason. All three are metadata problems, and all three are avoidable.
What is actually in a product photo file
A JPEG is not just pixels. It is pixels plus several metadata containers, and they do different jobs:
- EXIF - written automatically by the camera or phone. Make and model, focal length, aperture, ISO, shutter speed, date and time, orientation flag, and with location services on, GPS latitude and longitude. Frequently an embedded thumbnail preview too.
- IPTC - the publishing-industry standard for descriptive and rights information. Creator, copyright notice, web statement of rights, licensor URL, and since generative AI arrived, the digital source type field.
- XMP - Adobe's XML container, used by Lightroom and Photoshop for edit history and keywords, and increasingly as the carrier for IPTC fields.
- ICC colour profile - technically metadata, functionally essential. It tells browsers how to interpret your colour values. Strip it from an Adobe RGB file and the photo renders visibly duller.
The rule that matters commercially: some of this is dead weight, some is a liability, and some is load-bearing. Treating it all the same way is how sellers get into trouble.
Problem one: the photo that will not stay upright
Phone camera sensors have a fixed physical orientation, usually landscape. When you shoot in portrait, the sensor still records a landscape frame. Rather than rotating the pixel data, the phone writes an orientation flag into EXIF - a number from 1 to 8 that tells the viewer how to rotate and flip the image before displaying it. The pixels are sideways; the tag says "turn me". Everything depends on whether the software reading the file honours that tag.
Here is what most advice online gets wrong in 2026: browsers do respect it now. The CSS image-orientation property defaults to from-image, and that behaviour has been baseline across Chrome, Safari, Firefox and Edge since April 2020. A plain img element rotates the photo correctly with no CSS at all. The old "the web ignores EXIF orientation" advice is five years stale.
So why do photos still flip? Because the failure has moved upstream, into processing:
- A tool strips metadata without baking in the rotation first. The classic disaster. The pixels stay sideways, the instruction to rotate them is deleted, and now nothing can display the photo correctly. Cheap bulk compressors do this routinely.
- A resizer or cropper reads raw pixels. Many resizing libraries and canvas-based croppers work on the decoded pixel buffer and ignore the orientation flag. Your 1200 x 1600 portrait shot comes back as 1600 x 1200 landscape, and your square-crop logic then crops the wrong axis.
- Marketplace ingestion pipelines. A platform that re-encodes your upload may or may not preserve the flag, which is why the same file can look right on Shopify and wrong on a marketplace importer.
The fix
Bake the rotation into the pixels, then reset the flag to 1 (normal). Any competent resizing tool does this automatically: apply the EXIF rotation to the actual pixel data, write out an upright image, and set orientation to neutral so nothing downstream double-rotates it. The photo is then upright everywhere, regardless of what later tools do with the metadata.
To test your workflow, shoot a portrait photo on your phone, run it through your resize step, and open the output. If it comes out sideways, your tool strips before rotating and you should not be using it on product photos.
Problem two: the location data you did not mean to publish
If you photograph stock at home, and most small sellers do, your phone may be stamping your home address into every file as GPS coordinates accurate to a few metres.
Marketplaces generally strip metadata when they re-encode uploads, so exposure on a live Shopee or Amazon listing is usually low. The risk sits everywhere else your original files travel: images emailed to wholesale buyers or dropship partners, files shared in supplier chat groups, photos uploaded through an app or theme path that bypasses re-encoding, and anything sent as a file attachment rather than an in-app upload.
The answer is not to disable location on your camera, which is useful for other things, but to make metadata stripping a standard export step, the same way resizing is.
Problem three: kilobytes you are paying for and not using
Metadata is small relative to a full-resolution photo but it is not free, and it scales with your catalogue.
| Metadata component | Typical size | Keep for e-commerce? |
|---|---|---|
| Core EXIF (camera, exposure, date) | 2-8KB | No |
| Embedded JPEG thumbnail preview | 10-120KB | No |
| XMP edit history from Lightroom or Photoshop | 5-40KB | No |
| GPS coordinates | Under 1KB | No |
| IPTC rights and creator fields | 1-3KB | Situational - see below |
| ICC colour profile (sRGB) | 3-8KB | Yes, always |
The embedded thumbnail is the one worth hunting. Cameras and phones write a small preview into the file so the gallery app can render quickly. It serves no purpose on a web listing, and it can account for a large share of the total weight once the main image is already resized down to 1200px. A 500-SKU catalogue carrying 60KB of dead thumbnail per image is 30MB of pure waste, loaded by every shopper who scrolls a collection page.
This is why stripping and resizing belong in the same step rather than as two separate passes. PixelPrep handles the resize to each marketplace's dimensions while clearing out the metadata your listing does not need and applying the correct orientation, so you are not choosing between a small file and an upright one.
What you should deliberately keep
Blanket "strip everything" advice is where sellers now get burned, because two categories of metadata have become commercially significant.
The ICC colour profile
Never strip this. Convert to sRGB and embed the sRGB profile. If your file is Adobe RGB or Display P3 and the profile is deleted, browsers assume sRGB and render the colours flat and desaturated. Customers see one colour, receive another, and return the item. Saving 6KB is not worth a returns spike.
The AI-generated content tag
This is the one that catches sellers out. Since February 2024, Google Merchant Center has required AI-generated product imagery to carry the IPTC DigitalSourceType property, applied to image_link, additional_image_link and lifestyle_image_link. Two values matter:
- trainedAlgorithmicMedia - the image was generated entirely by a model
- compositeSynthetic - a real photograph with AI-generated elements, such as a genuine product shot dropped onto a generated background
Google's guidance is explicit that you must not remove these embedded tags. Generative tools, including Google's own Product Studio, write the tag automatically. If your optimisation step then strips all metadata, you have silently deleted a required disclosure and your listings can be disapproved - and the seller usually has no idea why, because nothing in the image looks any different. If any part of your catalogue uses generated imagery or generated backgrounds, your metadata policy needs to be selective rather than total. This sits alongside the broader questions in our guide to product image copyright and licensing.
IPTC rights fields, if your photography is itself an asset
Optional for most product sellers. If your images are a commercial asset in their own right, the IPTC Web Statement of Rights field is the minimum needed for Google Images to show its Licensable badge, with Licensor URL and Creator adding attribution and a route back to you. Note that Shopify's CDN compresses uploads and strips much of this on the way through, so verify it survives on the live URL rather than assuming.
How to check what is in your files
You cannot manage what you cannot see. Three approaches, in order of effort:
- File properties. On Windows, right-click, Properties, Details - and use the "Remove Properties and Personal Information" link for one-offs. On macOS, open in Preview and press Command-I. Both show camera data and GPS if present.
- ExifTool. The reference command-line utility. exiftool -a -G1 photo.jpg dumps every field grouped by container. exiftool -all= -tagsfromfile @ -icc_profile:all -orientation= folder/ strips everything while preserving the colour profile.
- Spot-check a live listing. Download your own published image from the storefront and inspect that. It shows what actually survives your platform's processing, which is the only figure that matters.
A metadata policy that works
Set it once, apply it to everything, and stop thinking about it:
- Archive originals untouched. Keep masters with full EXIF in a backup folder - it is evidence of authorship and a record of which settings worked. Strip on export, never on the master.
- Bake orientation, then reset the flag. Rotate the pixels, set orientation to 1, test with a portrait phone photo.
- Convert to sRGB and keep the profile. Non-negotiable for colour accuracy.
- Strip GPS, camera identifiers, serial numbers, edit history and embedded thumbnails. None of it helps a shopper, and some of it exposes you.
- Preserve DigitalSourceType on any AI-generated or AI-composited image. Check this before it costs you a Merchant Center feed.
- Do not rely on file metadata for SEO. Search engines read alt text, file names, surrounding copy and structured data. EXIF keywords are not a ranking factor.
- Verify after processing, not before. Run one file end to end through your actual pipeline every time you change tools.
The takeaway
Metadata is the least glamorous part of product imagery and one of the few things that can break a listing without changing a single visible pixel. The failure modes are quiet: a photo that flips only on mobile, a home address travelling with a wholesale enquiry, 60KB of invisible thumbnail multiplied across a catalogue, a disapproval notice with no obvious cause.
The fix is one decision made once. Strip the noise, bake the rotation, keep the colour profile, protect the AI disclosure tag - then let your resize step enforce it automatically on every image you export, so the discipline does not depend on you remembering it at 11pm before a launch.