Adobe publishes its keywording rules in plain sight, and contributors still lose files to them every week. The gap is not secrecy. It is that some of these rules are enforced by a human moderator reading your submission, while others are enforced silently by the search engine after your file is already live, and the two failure modes look nothing alike.
What follows is a working reference for the Adobe Stock keywording rules that carry real consequences: the hard limits, the ordering behaviour that decides which searches your file appears in, and the metadata mistakes that land a submission in the refusal queue rather than simply low in the results.
The 49-keyword limit is a ceiling, not a target
Adobe's contributor documentation caps keywords at 49 per submission. That number gets misread constantly. Contributors treat it as a quota to fill, on the theory that more terms mean more surfaces to be found on. Adobe's own guidance points the other way: keyword order is described as essential and expected to reflect the content's importance, which only makes sense if the terms at the top carry more weight than the ones at the bottom.
Look at the examples Adobe publishes alongside that rule. The sample photos in its titles-and-keywords reference carry roughly ten to eleven keywords each, not forty-nine. Adobe is showing you what a well-tagged file looks like, and it looks restrained. Every term you add past the point of genuine relevance buys almost no ranking benefit while adding one more chance to trip the irrelevant-metadata rule described further down. The cap exists to stop abuse, not to tell you how many words a good submission needs.
The first ten positions decide what your file competes with
Adobe states directly that the first ten keyword positions have the greatest influence on search ranking. This is the highest-leverage rule in the entire metadata workflow, and it is the one most contributors never act on, because most keywording tools emit an alphabetical or confidence-ranked list and the contributor pastes it in as-is. A confidence score describes how sure a vision model is that an object is present. It says nothing about what a buyer would type.
Adobe maintains a separate help page for nothing but changing the order of your keywords. Platforms do not document a feature that does not matter.
A reliable ordering pattern, drawn from the structure of Adobe's own examples: lead with the literal subject, then the action, then the setting, then the number of people, then demographic and viewpoint details, and only then the conceptual and emotional terms. A buyer searching for a picture types the subject first. Your keyword list should agree with them.
What "irrelevant keywords and metadata" actually means
Adobe's refusal documentation lists submitting inappropriate or irrelevant keywords and metadata as a non-compliance reason. It sits in the same list as watermarked content and hateful material, which tells you how the moderation team classifies it. This is not a ranking penalty applied quietly after the fact. It is a rejection, and it is decided by a person looking at your image next to your keyword list.
Three patterns reliably fall under it.
- Popular but unrelated terms. Tagging a quiet landscape with business, technology and teamwork because those words sell is the textbook case, and the easiest one to catch, because the moderator has the image open in front of them.
- Names of government agencies, and any wording that implies the image depicts a newsworthy event. Both are named refusal reasons in Adobe's own list, and both are usually accidents rather than gaming attempts.
- Artist names whose work is copyrighted. This rule is broader than most contributors realise: Adobe applies it to the prompt, the title and the keywords, all three.
That last rule matters more every quarter, because it is where keyword policy and generative AI policy meet. Writing a style reference into your keyword field is a compliance failure rather than a stylistic choice, and a term that would sink an image in the prompt sinks it just as reliably sitting in a description you never edited.
The keyword rules that live outside the keyword field
Adobe's metadata guidance treats the title with equal weight, and title problems get files refused just as reliably as keyword problems. Five rules are stated plainly enough that there is no room to argue with a moderator about them.
- Keep the title short. Adobe recommends staying under 70 characters and describing what is visually most important, not the gear or the technique used to capture it.
- Keep intellectual property, artist names and real people out of the title entirely. The same prohibition that governs keywords governs the title.
- Include the words from your title in your keyword list. Adobe asks for this explicitly. It costs nothing and it is one of the few places where redundancy is the instruction.
- Match the language of your titles and keywords to the language set in your Contributor Account. This is a compliance requirement, not a preference, and it catches contributors who translate metadata mid-portfolio.
- Use the local spelling your buyers use. Adobe names colour against color as its own example of the problem.
Generative AI turned keyword accuracy into a labelling question
Adobe refuses content in both directions here. AI-generated work submitted without the generative AI checkbox is refused, and work that is not AI-generated but is labelled as if it were is refused too. The label is metadata, and it is checked. If you produce both kinds of file in the same session, the real risk is not that you break a rule on purpose. It is that a batch process carries the wrong flag across a folder boundary and you never look again.
The keyword-level version of the same rule is the artist-name ban above: anything that reads as an attempt to attach your file to a living artist's market is treated as non-compliance whether the image was generated or photographed. The wider picture across agencies is in our guide to metadata for generative AI images, since Adobe's approach is strict but not unique.
The rules moderators never mention, because search enforces them
Nothing in this next group will get a file refused. All of it decides whether an accepted file ever earns anything, which is why it goes unnoticed for months at a time. Adobe's documentation lays out the categories a strong keyword set covers, and the pattern is worth copying literally.
- Separate your descriptive elements instead of compounding them. Colour, texture and age each get their own term rather than being fused into one phrase nobody searches for.
- Pair general and specific terms for the same subject, so a file matches both the broad query and the narrow one. Animal and mammal and the actual species, not just one of the three.
- State the number of people, the setting, the time of day and the weather. Indoors, outdoors, night, one person, three people. These behave like filter values on the buyer's side, not like decoration.
- Name the viewpoint. Aerial view, high-angle view, portrait. Buyers search by framing far more often than contributors expect them to.
- Add conceptual terms last. Solitude, childhood, milestone, growth. This is how a designer briefs a search when they do not yet know what the picture looks like, and it is the group most contributors skip entirely.
A pre-submission check that takes under a minute
Run this on a sample from every batch rather than on every file, and it will catch the errors that repeat, which are the only ones worth catching at scale.
- Read your first ten keywords on their own, without the image. Do they describe the picture to someone who cannot see it?
- Scan the rest of the list for any term you could not defend to a moderator with the file open. Delete it rather than argue with it later.
- Confirm the words in your title appear somewhere in the keyword list.
- Confirm the metadata language matches your Contributor Account setting.
- Confirm the generative AI flag matches reality for every file in the batch, not just the first one you opened.
Where automation actually helps
Most AI keywording tools solve the wrong half of this problem. Producing forty-nine plausible words for an image is the easy part, and vision models have been able to do it for years. The part that decides whether a file survives moderation and then sells is everything above: order, restraint, policy compliance, and consistency across a batch of two thousand rather than a batch of one.
Rastock AI was built around that second half. Metadata is generated against per-agency rules instead of one generic output, with banned and mandatory keyword lists you control, ordering you can lock, and a rollback if a batch goes out wrong. Delivery to 10+ agencies over FTP and status tracking sit in the same place, so a file and its metadata never drift apart. There is no revenue share and no ownership claim on your work, and IPTC and CSV export stay open, so you can leave whenever you want. The feature overview covers the workflow, and pricing includes a 14-day free trial with no card required.
Where to go from here
The rules above are the enforced subset. For the full field-by-field picture, including everything the keyword field does not cover, read the Adobe Stock metadata requirements reference. For the refusal side, our breakdown of Adobe Stock rejection reasons maps each stated reason to a fix. And if the underlying problem is that your keyword sets were never structured consistently in the first place, start with how tagged keywords should be organised, because ordering rules only pay off once there is something coherent to order.
None of this is exotic. It is written down, it is short, and it changes rarely. The contributors who lose files to it are almost never the ones who read it and disagreed. They are the ones who assumed the rules were about quantity.
Frequently asked questions
How many keywords should I use on Adobe Stock?
Adobe allows up to 49 per submission, but that is a ceiling rather than a recommendation. The examples in Adobe's own titles-and-keywords documentation use roughly ten to eleven keywords per image. Since Adobe states that the first ten positions carry the greatest influence on ranking, terms added far down the list contribute very little while increasing the chance one of them reads as irrelevant to a moderator. Accuracy beats volume here.
Does keyword order really matter on Adobe Stock?
Yes, and Adobe says so directly: the first ten keyword positions have the greatest influence on search ranking, and keyword order should reflect the importance of the content. Adobe also publishes a dedicated help page on changing the order of your keywords, which is not something a platform documents for a feature that does nothing. Lead with the literal subject, then action, setting, people count, and only then conceptual terms.
Can Adobe Stock reject a file just for its keywords?
It can. Adobe's refusal documentation lists submitting inappropriate or irrelevant keywords and metadata as a non-compliance reason, alongside items like watermarked content. Keyword spam is therefore a rejection risk, not only a ranking penalty. The same section also bans keywords containing the names of government agencies, wording that implies a newsworthy event, and the names of artists whose work is copyrighted.
Can I use an artist's name as a keyword if the image only resembles their style?
No. Adobe prohibits references to the names of artists whose work is copyrighted, and applies the rule to the prompt, the title and the keywords alike. It does not matter whether the image was photographed or generated, and it does not matter that the resemblance is loose. If the term is there to attach your file to a named artist's market, it is treated as non-compliance and the submission is refused.
What language should my Adobe Stock titles and keywords be in?
The language set in your Contributor Account settings. Adobe states that titles and keywords must match that setting, so it is a compliance requirement rather than a style choice. Adobe also advises using local spelling conventions where they differ, naming colour against color as the example, so match the spelling your buyers are most likely to type rather than the one your keyboard defaults to.
Do AI keywording tools break Adobe Stock's rules?
The tool is not the problem; unreviewed output is. Most generators return an alphabetical or confidence-ranked list, which ignores Adobe's ordering rule entirely, and many will happily fill all 49 slots with terms that get progressively less relevant. Both behaviours work against you. A generator is safe when it applies per-agency rules, respects banned and mandatory keyword lists, and lets you fix ordering before delivery rather than after.