Bulk editing keywords across a large stock archive is permitted on the major agencies, and the real risk is not permission but reversibility. Adobe Stock and Shutterstock both let you change titles, keywords and categories on approved files, and neither lets you edit anything sitting in the review queue. What damages portfolios is changing thousands of records with no baseline, no copy of the old metadata and no way to tell afterwards whether sales moved because of the edit or in spite of it.
What the agencies actually allow
Adobe Stock states the rule plainly in its contributor documentation: changes can be made before content is submitted for moderation or after it has been approved, but not while it is pending. Editing published work is done through the Portfolio section of the contributor portal, where you open a file and edit the title, category or keywords directly. Keyword order can be rearranged, individual keywords removed, or the whole set erased and started again. Adobe's own help page walks through each of those actions.
Shutterstock works the same way with different names. Approved content is edited through Catalog Manager under Portfolio; pending content cannot be edited at all, only deleted from the queue. Two restrictions are specific to Shutterstock and worth knowing before you plan a large pass: rejected content cannot be amended and has to be re-uploaded even if the only change is a title, and commercial versus editorial status cannot be switched after approval. Shutterstock's contributor help sets both out.
There is one more rule that catches people running automated cleanups. Keywords an administrator previously removed from a file cannot be re-added, and restricted keywords cannot be introduced after approval. A script that reapplies a saved keyword set to every file in a category will happily try to do both. Shutterstock also audits its collection periodically, so a bulk edit that quietly reintroduces prohibited terms is a problem that surfaces later rather than never.
What losing performance actually means
The fear most contributors have when they hesitate over a large edit is that changing metadata resets a file's search history. Neither agency documents anything of the kind, and the more common outcome is more mundane: the edit removes a keyword that was quietly responsible for a slow trickle of sales, and nobody notices for three months because the file was only earning a few downloads a year in the first place.
That is why the safety of a bulk edit has almost nothing to do with the tool and almost everything to do with two habits: keeping the previous metadata somewhere you can read it back, and knowing which files were earning before you touched them. A tool that writes 8,000 keyword sets in an hour is not risky. Writing 8,000 keyword sets with no record of what was there before is.
Tools that handle bulk keyword edits
These solve different halves of the job. Some edit the metadata embedded in your local files; some edit the records the agency already holds. Confusing the two is the most common reason a cleanup does not take effect.
Agency-native editors: Adobe Portfolio and Shutterstock Catalog Manager
- Best for: edits that must apply to files already published, since this is the only route that changes the live record.
- Cost: included with your contributor account.
- Main limitation: they are per-file interfaces. Adobe lets you apply identical keywords to a multi-file selection during upload, but neither portal is built for rewriting several thousand published records, and neither gives you a copy of what you replaced.
Xpiks
- Best for: contributors who want direct, precise control over XMP, IPTC and EXIF fields in local files. Xpiks is a desktop application with batch editing, find and replace, batch keyword deletion, duplicate detection, CSV import and export, and FTP, FTPS and SFTP upload.
- Cost: a free tier plus a paid Pro tier; several of the features that matter most for large archives, including batch delete and find and replace, sit in Pro. Check the vendor's pricing page for current terms.
- Main limitation: it edits your local files. Changes reach an agency only when you re-upload, which does nothing for work already published unless you also edit the live record.
Adobe Bridge and Lightroom
- Best for: photographers already cataloguing in Adobe software, since keywords added there are preserved when files are uploaded to Adobe Stock.
- Main limitation: keyword order. Adobe's documentation notes that Lightroom keywords arrive in alphabetical order, and order matters to Adobe Stock's search. A Lightroom-driven bulk edit therefore needs a reordering pass afterwards, which is exactly the step people skip.
ExifTool and scripting
- Best for: anyone comfortable at a command line who needs an exact, repeatable transformation across tens of thousands of files, and a full dump of the existing metadata before touching anything.
- Cost: free and open source.
- Main limitation: it knows nothing about agency rules. It will write a trademarked term or a banned keyword into 20,000 files as cheerfully as a valid one, and it has no opinion about whether your new keyword set is any good.
Rastock AI
- Best for: regenerating metadata across a large archive when you want the new keywords checked against each agency's rules before they are written, and delivered to 10+ agencies over FTP or SFTP under your own contributor accounts.
- Cost: subscription, with a 14-day free trial and no card required. No revenue share, no ownership claim on your files, and IPTC, CSV and XML export stay open.
- Main limitation: the same one every external tool has. It writes and delivers metadata for files you send; it cannot reach into an agency's portal and rewrite records that are already published there. For that, you are back in Catalog Manager or the Adobe portfolio view.
When to use which
If the files are already published and the edit is small, use the agency's own editor. A few hundred files with one bad keyword is an afternoon in Catalog Manager, and nothing else touches the live record anyway.
If the edit is mechanical and you know exactly what the output should be, use ExifTool or Xpiks. Replacing one term with another across a category, stripping a keyword you now know is prohibited, or removing duplicates are all deterministic operations. You do not need a model to decide them, and a script leaves an audit trail a graphical tool does not.
If the edit is a judgement call, mechanical tools stop helping. Rewriting keyword sets from 2019 that were technically accurate but conceptually thin is not a find-and-replace problem. That is where an AI pass earns its place, and also where policy checking matters most, because a model that has not been told the agency rules will produce fluent keyword sets full of things Adobe explicitly prohibits.
If you are running several agencies, the answer is usually two tools rather than one: something that rewrites the local metadata and delivers fresh files, plus a pass through each agency's own editor for whatever is already live. Anyone selling you a single tool that does both halves is describing something that does not exist.
A procedure that survives a bad decision
The steps below are not tool-specific. They are what makes any of the options above recoverable.
- Export the current metadata for every file you intend to touch, to CSV or a full ExifTool dump. This is the backup. Without it, undo does not exist.
- Pull a sales baseline for the same set, by file, covering at least the last twelve months. Six months is not enough on a portfolio where individual files sell a handful of times a year.
- Split the set. Leave your best earners alone for now and edit the tail first. If something goes wrong, it goes wrong on files that were producing almost nothing.
- Run the edit on a sample of 50 to 100 files and read the output by hand. Every automated keywording pass has a characteristic failure mode, and one page of results is usually enough to find it.
- Check the new sets against agency rules before writing: no trademarks, nothing irrelevant, and a sensible count. Adobe's own guidance suggests roughly 15 to 25 keywords and warns that irrelevant terms cause rejection, with repeat offenders risking account blocks.
- Wait before doing the rest. Give the edited tail 60 to 90 days against its baseline. Then, and only then, decide whether the same treatment goes anywhere near your best files.
Related problems that are not this one
Three neighbouring questions get answered as if they were the same one, and they are not. If the issue is that your old keyword sets are simply bad, that is a keyword cleanup problem. If the issue is the same term appearing in six spellings across a library, that is deduplication and normalisation. And if the goal is getting an ageing portfolio seen again rather than fixing anything specific, that is re-optimising an old portfolio. This article is about the mechanics and the safety of the edit itself, whichever of those three sent you here.
The baseline step in particular is easier if you already track which keywords drive your downloads. Without that, step two of the procedure above is guesswork, and you will not be able to answer the only question that matters afterwards.
If the judgement-call half of the job is where you are stuck, that is what Rastock AI is built for: metadata generated with agency policy applied, banned and mandatory keyword handling, and rollback before anything is delivered. The features page covers how the metadata and delivery sides fit together, and there is a 14-day free trial with no card required if you want to run a sample of your archive through it before committing to anything.
Frequently asked questions
Can I edit keywords on files already approved by Adobe Stock and Shutterstock?
Yes on both. Adobe Stock allows metadata changes before submission or after approval, but not while a file is pending moderation, and edits are made in the Portfolio area of the contributor portal. Shutterstock allows the same through Catalog Manager under Portfolio, and also blocks edits on pending content. On Shutterstock, rejected files cannot be amended at all and must be re-uploaded.
Does editing keywords reset a file's search ranking?
Neither agency documents any reset tied to metadata edits, so the widely repeated claim is not something either company states. The realistic risk is different and more boring: a bulk edit quietly removes a keyword that was producing occasional sales, and on a low-volume file it can take months before anyone notices. That is a measurement problem, which is why a sales baseline matters more than the fear.
What is the safest order to run a large keyword edit in?
Export the existing metadata first, so an undo exists. Pull at least twelve months of per-file sales as a baseline. Edit your weakest files before your best ones. Hand-check a sample of 50 to 100 results before running the rest. Then wait 60 to 90 days and compare against the baseline before letting the same process touch anything that earns well.
Will editing metadata in Lightroom or Bridge update my published files?
No. Those tools change the metadata embedded in your local files, which only reaches an agency when you upload. Published records have to be edited in the agency's own portal. One extra detail matters for Adobe Stock: keywords carried over from Lightroom arrive in alphabetical order, and keyword order affects search there, so a reordering pass is needed after any Lightroom-driven edit.
Are there keywords I am not allowed to add during a bulk edit?
Yes, and automated passes hit this regularly. Shutterstock does not permit re-adding keywords an administrator previously removed, or introducing restricted keywords after approval, and it audits its collection periodically. Adobe prohibits trademarked terms such as brand and camera names, and states that irrelevant keywords cause rejection, with repeat offenders risking account blocks.
How many keywords should each file have after the edit?
Adobe Stock's own guidance suggests roughly 15 to 25, ordered with the main subject first, then other significant elements, then themes and concepts. Padding beyond that with weakly relevant terms works against you rather than for you, since irrelevant keywords are a documented rejection reason. Treat the count as a consequence of describing the image well, not as a target to fill.