Adobe updated its official guidance on titles and keywords on 18 August 2026. Not a blog post, not a forum answer from a moderator: the help page contributors are graded against.
Most of it confirms what good contributors already do. One rule is stated more plainly than before, and it is the one that decides whether your work gets found: the first ten keywords carry the most search weight.
Here is the whole page, rule by rule, and what each one changes about the text you type.
Short version: Titles under 70 characters, no gear talk, no brand or artist names. Up to 49 keywords, ordered by importance, because the first 10 positions have “the greatest influence on search ranking”. Each keyword once. One language, and it must match your account language. Relevance decides whether the asset is accepted at all.
The title rules
Adobe gives four instructions for titles:
- Keep it short, ideally under 70 characters.
- Focus on what is most visually important in the content.
- Avoid overly technical or gear-heavy terms.
- Do not refer to anything involving IP, artist names, or real people.
The two example titles Adobe publishes are plain sentences: “Gay couple hugging in the park” and “Women in a laboratory with face masks and gloves”. No lens, no ISO, no “shot on 85mm”. No “in the style of” anybody.
That third rule kills a habit photographers bring from portfolio sites. Buyers do not search for equipment. A title that spends 20 characters on gear has 20 fewer characters describing what the buyer typed.
The fourth rule is a rejection trigger. Artist names, character names and brand names in a title are an IP problem, and Adobe refuses the asset rather than editing it for you.
The rule that matters most: your first ten keywords
Adobe’s wording is direct: “Place the most important and relevant keywords in the first 10 positions as they have the greatest influence on search ranking.”
Keywords are not a bag. They are a ranked list, and the top of that list is scored harder than the bottom.
Look at what Adobe puts in those first ten slots in its own examples:
| Image | First ten keywords |
|---|---|
| Senior woman flexing her muscles on beach | Woman, back, muscular, flexing, muscles, beach, Caucasian, senior adult, adult, one person |
| Living with sign language | Sign language, family, meeting, deaf, three people, communication, smiling, sitting, table, indoors |
Read the first list again. Subject, body part, attribute, action, object, location, demographic, age, adult, count. Ten slots and not one wasted word. Nothing decorative. No “beautiful”, no “professional”.
This is where most uploads lose. A contributor types 40 accurate keywords in the order they happened to think of them, or lets a tool dump them alphabetically, and the ten strongest terms land in positions 23 to 40 where they carry the least weight. The keywords are not wrong. They are in the wrong seats.
Getting the ten strongest terms into the first ten seats is the whole job, on every file. AutoKeyWorder (ad, own product) is a Chrome extension that reads each image with AI vision and fills titles, keywords and categories directly inside the Adobe Stock upload form, with no CSV export and no re-import. AutoKeyWorder runs at roughly 5 seconds per image and covers 7 platforms: Adobe Stock, Shutterstock, Freepik, Displate, TeePublic, Zedge and Redbubble. Install is free with 50 credits included, one credit per image, no credit card. Install the Chrome extension
The keyword rules
Adobe allows up to 49 keywords per submission and asks for relevant nouns, adjectives and concepts specific to the content. Two more instructions sit alongside the count: use local spelling where applicable (color versus colour), and think like a buyer, meaning ask what they would type to find this exact asset.
The page then lists the kinds of keyword worth a slot, with Adobe’s own examples:
| Keyword type | Adobe’s examples |
|---|---|
| Separate descriptive elements | White, fluffy, young animal, pup |
| General and specific | Animal, mammal, carnivore |
| Location-based | London, England or London, Ontario |
| Conceptual | Solitude, childhood, milestone, cold |
| Number of people | One person, three people, alone |
| Setting | Indoors, outdoors, day, night, sunny, cloudy |
| Viewpoint or camera angle | High-angle view, aerial view, portrait |
| Model information | Demographics: Black woman, Latinx teen, senior man. Gender: female, male, non-binary |
Notice what “camera angle” means here. Aerial view is a keyword because buyers filter by it. Aperture is not, because nobody searches for it. The gear ban in the title section and the angle allowance here are the same rule seen twice: keep what buyers type, drop what photographers discuss.
Location is worth a second look. Adobe writes it as “London, England or London, Ontario”, which is a warning about ambiguity. If you cannot name the place with confidence, leave it out rather than guess.
The language rule people miss
“Titles and keywords must match the language set in your Contributor Account.”
And in the mistakes list: “Submit content in one language only.”
So a German contributor typing English keywords under a German account setting is fighting the search engine instead of feeding it. Check the language in your account settings once, then keep every submission in that language. Mixed-language metadata on one asset is an explicit mistake, not bonus reach.
The four mistakes Adobe names
The page lists what gets content refused:
- Irrelevant keywords. Adobe ties this directly to refusal, not just to poor ranking. Padding a list to hit 49 is a risk, not a strategy.
- Trademarks, brand names and personal information. Keep them out of the submission.
- More than one language in a single submission.
- Repeating a keyword. Use each one once. A duplicate wastes a slot and buys nothing.
Point one deserves emphasis because it inverts a common instinct. Contributors treat the 49 cap as a target and fill it. Adobe treats relevance as a gate. Thirty accurate keywords beat 49 with a padded tail.
Adobe’s own suggested keywords still need you
The page confirms Adobe generates suggested keywords when you upload to the Contributor Portal, and says you should “review and adjust them to improve accuracy”.
Take that literally. Suggestions come out of an image model that has never seen your buyer. It does not know which of the terms it produced is the one people actually type, and ordering is precisely what Adobe says decides ranking. Accepting a generated list unedited puts your ranking in the hands of whatever order the suggestions happened to arrive in.
What AutoKeyWorder already does about all of this
AutoKeyWorder was built against these rules, so most of this page is not new work for anybody already using it. The honest mapping:
| Adobe’s rule | What the extension does |
|---|---|
| Titles ideally under 70 characters | Every title is fitted to a 70-character ceiling before it reaches the field. The trim drops weak adjectives first, then trailing secondary objects, never the subject or the action verb. Across 45 live outputs: titles over 70 characters fell from 93.3% to 0%, average length 81 to 66 |
| First 10 keywords decide ranking | Keyword one is the dominant noun of the top buyer query. Two to five complete that query, six to ten cover the next one or two. Style, colour and format words are demoted out of those ten slots |
| No gear-heavy terms | Camera and format tokens are scrubbed: 4K, HD, 1080p, DSLR, mirrorless, high resolution, plus Canon, Sony, DJI, GoPro |
| Think like a buyer | Before any keyword is written, the model drafts the five to seven phrases a buyer would type for that image, then builds the list to win them |
| Up to 49, relevant only | Output is 15 to 35 keywords, set by how much is genuinely in the frame. The count is a result, never a quota, so there is no padded tail to refuse |
| Use each keyword once | A dedupe pass removes repeats, singular and plural twins, and near-synonyms chasing one search |
| No trademarks or brand names | The brand name filter is off by default for Adobe Stock. Nike becomes athletic shoes, iPhone becomes smartphone |
| Number of people, setting, viewpoint, concepts | Each is an explicit keyword tier, so one person, outdoors, aerial view and the concept layer come out as standard |
One divergence, stated plainly instead of glossed. Adobe lists model demographics as keyword-worthy, and its own example puts Caucasian in slot seven. AutoKeyWorder strips ethnicity tokens on every platform, because the best-selling listings we measured contain none of them. Gender and age stay, since buyers search those. If you want ethnicity terms on a specific asset, add them by hand before you submit.
Two things stay with you regardless of tooling. Set your Contributor Account language to English, because that is the language the extension writes in and Adobe wants the two to match. And tick the AI-generated toggle yourself when a batch is generative, because it is not auto-detected.
What to change on your next upload
- Reorder before you submit. Ask which ten terms a buyer would type, and put those first.
- Cut the tail instead of filling it. Relevance is a refusal gate.
- Read your title at 70 characters and ask whether it still names the subject.
- Check your account language once, then never mix languages again.
- Never paste a brand name, an artist name or a real person’s name into either field.
For the full metadata strategy behind these rules, see the Adobe Stock keywords guide. For the extension’s Adobe Stock workflow, batch by batch, see AutoKeyWorder on Adobe Stock. If assets are being refused for reasons metadata cannot fix, the rejection reasons guide covers the technical side.