Rollback: How to Undo a Bad Metadata Batch

What a metadata rollback can actually reverse: which fields Adobe and Shutterstock let you edit after approval, and what a deletion really costs.

A bad metadata batch is reversible in three different places, and they do not reverse at the same speed. Your own files usually still hold a recoverable original. The agency's live record sometimes accepts a correction without review and sometimes refuses one outright. The delivery queue is the only layer where moving fast genuinely changes the outcome.

What rollback means once metadata has already shipped

Most contributors picture a rollback as one action, an undo button somewhere in a dashboard. In stock delivery it is three separate jobs, because a keyword set that went out last night now exists in three copies and each copy lives under different rules.

The order matters more than the speed. Fixing local files first and then re-sending is clean. Fixing the agency record first, then discovering your masters still carry the bad keywords, means doing the entire job twice.

The rollback clock, stage by stage

Six stages of a metadata rollback: stop the queue first, restore the backups ExifTool leaves on disk, edit approved records where the agency permits it, wait out Adobe's moderation lock, allow up to 72 hours for a Shutterstock deletion to clear the customer site, and accept that download history and search ranking do not transfer to a replacement file.

Restoring original metadata on your own files

If the bad batch was written with ExifTool, the undo is already on disk. The ExifTool documentation states that by default the original files are preserved with _original appended to their names, and tells you to verify that the new files are OK before erasing the originals. That default is the most useful safety net in a metadata pipeline, and it is also the one people routinely switch off to save disk space.

Two utility options finish the job. The restore_original option restores the specified files from their backup copies, and delete_original removes those backups once you are satisfied with the result. Run the restore on a copy of one folder first, confirm the keyword set is the one you actually wanted, and only then run it across the whole batch.

The option to be careful with is overwrite_original, which overwrites the original instead of preserving it by adding _original to the file name. The documentation's own caution is that it should only be used if you already have separate backup copies of your image files. If your batch script carries that flag, your rollback plan is whatever your backup system says it is, not what ExifTool can do for you. Our guide to IPTC, XMP and embedded keywords covers which fields these writes actually touch.

Editing the agency record after approval

Adobe Stock draws its line at moderation. Its contributor help on updating your content states that content can be edited before submission or after approval and publication, but not during moderation. In practice that gives you two windows with a locked gap in the middle: fix it before you send, or wait until it clears review and fix it then.

Timing on the Adobe side is split as well. Changes are reflected immediately in the My Portfolio section, but updated metadata may take time to appear in customer search results. A corrected keyword set that looks right in your portfolio is not yet a corrected keyword set in the search index buyers are using.

Shutterstock is more permissive with approved content than most contributors assume. Its help article on managing content in your portfolio states that to adjust the description, keywords or category of an approved image or clip you can edit one file at a time or select multiple images or clips at once, and that these corrections are not subject to review. No maximum selection size is given. That makes a metadata-only rollback on Shutterstock genuinely cheap, and worth doing properly rather than in a panic.

One exception sits inside the same article. Assets in the Data Catalog cannot be edited once approved, though they can still be deleted. If part of your portfolio sits there, the metadata question and the deletion question collapse into a single decision.

What a rollback cannot recover

Metadata and the file itself are different problems with very different price tags. Shutterstock's content resubmission guidelines state that approved content cannot be overwritten with a new version, and that to replace approved content you must first delete the image or video from your portfolio before submitting the new one. Three consequences follow, and all three are published rather than inferred.

This is the real reason a metadata rollback and a file rollback should never be filed under the same task. Fixing keywords on an approved asset costs you almost nothing. Replacing the asset costs you its entire sales history, and no tool on either side of the transaction can hand that back.

Stopping a batch that is still in flight

The cheapest fix is the one that never reaches moderation. If a batch is still uploading, or sitting delivered but unreviewed, stopping it keeps the entire problem inside the one layer you fully control. Once files enter review at Adobe you have lost edit access until they clear it, and once they are approved anywhere the replacement cost described above starts to apply.

Stopping mid-batch brings its own hazard, which is sending the same files twice when you resume. Resuming a failed batch without creating duplicates covers the ledger side of that in detail. The short version is that the record of what was actually delivered has to live outside the uploader, because an interrupted uploader is precisely the thing you cannot trust to tell you.

A five-step rollback run

The order a metadata rollback runs in: freeze the batch, scope the damage by file and by field, restore the local originals from their backup copies, then either edit the agency record in place where the agency allows it or re-deliver corrected files, and finally verify at the agency rather than inside your own tool.

Building batches you can actually reverse

None of this is exotic. It is the same discipline a bulk keyword edit across a large archive needs anyway, applied before the write rather than after it.

Where a workflow tool helps, and where it does not

A delivery tool can cover three of the four jobs above. Rastock AI keeps a reversible record of what each batch wrote, checks keyword sets against each agency's banned and mandatory terms before delivery rather than after a rejection, and exports IPTC, CSV and XML freely, so the state before a change is always yours to keep. Delivery runs over FTP and SFTP to 10+ agencies, and any other agency that accepts FTP or SFTP can be added manually. The feature list sets out what is included.

What no tool can do is reach into an agency and restore something that agency has already discarded. Nothing returns Shutterstock's download history to a replaced file, and nothing edits an Adobe asset while it sits in moderation. Those two limits belong to the agencies, and any tool that implies otherwise is overselling itself.

If you would rather test the reversible side on a folder of your own files before trusting it with a real batch, the 14-day free trial runs without a card.

Frequently asked questions

Can you undo a metadata batch after a stock agency has approved it?

Partly, and it depends which layer you mean. Metadata is often fixable: Shutterstock states that description, keywords and category can be edited on approved content, one file or several selected at once, and that those corrections are not subject to review. The file itself is a different matter, because approved content cannot be overwritten, so replacing it means deleting and resubmitting.

How do you restore original metadata after a bad ExifTool run?

ExifTool preserves the original by default, appending _original to the file name, so the previous version is usually still on disk. The restore_original option renames those backups back over the edited files, and delete_original clears them once you are satisfied. The exception is a batch run with overwrite_original, which skips the backup entirely and leaves you dependent on your own backups.

Can you edit keywords while an image is still in review?

Not on Adobe Stock. Its contributor help states that content can be edited before submission or after approval and publication, but not during moderation. That makes the review period a locked window: anything wrong that you notice after sending has to wait until the asset clears. Catching errors before delivery is the only way to avoid that wait entirely.

Does deleting and re-uploading a file lose anything?

Yes, and the loss is published rather than speculative. Shutterstock's resubmission guidelines state that download history and search ranking will not transfer to the new version, and that approved content cannot be overwritten. Deletion also takes effect slowly, with up to 72 hours for the asset to disappear from your portfolio on the customer website. Replace files only when the file itself is wrong.

What should you capture before running a large metadata batch?

Two things make a rollback possible: a CSV export of the titles, descriptions and keyword sets as they stand before the batch writes anything, and the untouched backup files the write leaves behind. Add the revision number of whatever keyword list generated the batch. With those three, a rollback is a precise diff. Without them it is guesswork and retyping.

Are the rollback rules different at each agency?

Materially, yes. Shutterstock allows multi-select metadata edits on approved content with no stated limit and no re-review. Adobe Stock allows edits before submission and after publication but locks them during moderation. Assets in Shutterstock's Data Catalog cannot be edited after approval at all, only deleted. Any rollback plan that assumes one rule across every destination will fail somewhere.