Most advice about image keywords for SEO quietly assumes one destination. But a photographer publishing work in 2026 usually has two: their own website, where Google decides whether the image appears in search, and stock agencies, where an internal search engine decides whether buyers ever see it. These systems read different fields, reward different writing, and punish different mistakes.
Confusing them is the most common reason a well-keyworded portfolio underperforms in both places. This guide separates the two systems, shows exactly what each reads, and explains how to maintain one metadata set that serves both without compromise.
The two systems, side by side
Google Images does not read a keyword field. There is no meta keywords tag it honours, and there has not been for well over a decade. It infers the subject of an image from the alt attribute, the filename, the caption, the text immediately surrounding the image, the page title and headings, structured data on the page, and the visual content itself through computer vision. Everything is contextual.
Stock agency search works the opposite way. There is no surrounding page, no headings, no body copy. The only inputs are a title, a keyword list, a category, and in some cases location. Adobe Stock accepts up to 49 keywords per asset and states that the first 10 positions carry the most weight in ranking. Shutterstock requires a minimum of 7 keywords and caps the list at 50, with a title that must be a factual sentence describing the subject.
So the same photograph needs prose on your website and a ranked list on an agency. The good news is that both derive from the same underlying description of what is in the frame. The work is in the translation, not in doing the thinking twice.
What Google actually reads: the five fields that matter
Alt text
This is the highest-leverage field, mostly because it is so often missing or written badly. Alt text should describe the image the way you would describe it to someone who cannot see it - a short, natural phrase, typically five to fifteen words. "Senior woman flexing her muscles on a beach at sunrise" is good alt text. "stock photo woman beach fitness senior muscles sunrise summer health" is keyword stuffing, and it damages both accessibility and ranking.
Decorative images - dividers, background textures, spacer graphics - should carry an empty alt attribute rather than an invented description. Screen readers skip them, and Google does not benefit from noise.
Filename
Google's own documentation is explicit that short but descriptive filenames beat camera defaults. IMG_4471.JPG tells the crawler nothing. autumn-forest-path-fog.jpg tells it a great deal before a single pixel is analysed. Use hyphens between words, not underscores, and keep it to three or four words. Rename at export, not after upload, because changing a filename later means a new URL and a lost index entry.
Caption and surrounding text
A visible caption is read by both Google and human visitors, and it is the only image signal that also affects engagement. Text in the paragraph directly above or below an image counts as context. This is why a gallery page of forty images with no words around them ranks poorly: there is nothing to disambiguate one frame from the next.
Structured data and image sitemaps
Schema markup connects an image to an entity - a product, a recipe, an article, a person - and unlocks rich results that plain images cannot reach. For a site with hundreds or thousands of images, an image sitemap is the practical way to make sure every file is discovered rather than relying on crawl depth. Neither is a ranking trick; both are discovery infrastructure.
Format and file weight
WebP is the sensible default in 2026, with AVIF offering better compression where browser support allows. This is not a keyword issue, but page speed is a ranking factor and images are usually the heaviest thing on a photography page. A perfectly keyworded 6 MB JPEG still loses to a well-compressed competitor.
Embedded metadata: the field both systems can read
IPTC and XMP fields live inside the file itself - title, description, keywords, creator, copyright, location. Agencies read them directly on FTP ingestion, which is why embedding is the entire basis of bulk stock submission. Google does not use embedded keywords as a ranking signal, but it does surface IPTC rights and creator fields in the Image Credits panel in Google Images, which matters for attribution and licensing enquiries.
There is a practical trap here. Many CMS platforms and image optimisation plugins strip metadata during resizing to save bytes. If your licensing depends on embedded copyright fields surviving to the public file, check what your pipeline actually outputs rather than what you exported.
Our guide to writing metadata for stock photos covers the field-by-field mechanics of IPTC and XMP in more detail, including which fields each agency reads on ingestion.
Photo keywording best practices that hold in 2026
Order matters where it is ranked. Adobe Stock states plainly that the first ten keywords have the greatest influence on search ranking. That makes keyword order a real decision rather than an afterthought. Lead with the subject, the action and the primary concept; push demographic detail, camera angle and setting further down.
Mix literal and conceptual. A photograph of a person on a mountain is literally "hiker, summit, backpack". It is conceptually "achievement, solitude, perseverance". Buyers searching for a campaign visual use the second vocabulary far more often than the first, and purely literal keywording is the single most common weakness in AI-generated tag sets.
Use each keyword once. Repeating a term in singular and plural form, or across a title and keyword list, does not compound. Adobe explicitly lists duplicate keywords among the mistakes that lead to refusal.
Fewer, better keywords beat the maximum. The 49 and 50 limits are ceilings, not targets. A list padded to the cap with marginal terms dilutes relevance scoring and raises the chance of a metadata rejection. Twenty-five accurate keywords outperform fifty vague ones.
Keep brands out. Trademarks, brand names and product names in commercial metadata are a rejection reason at both Adobe Stock and Shutterstock, regardless of whether the mark is visible in the frame. The same applies to personal information and, on Adobe, to artist names used as style references.
One language per submission. Adobe requires titles and keywords to match the language set on your contributor account. Shutterstock requires English, with narrow exceptions for scientific Latin names, place names and foreign phrases that are standard in English. Mixed-language lists get refused.
Where the two systems conflict - and what to do
The genuine conflicts are few, but they are worth naming.
Agency titles are descriptive sentences optimised for an internal search index. Website page titles are optimised for click-through in a results page. "Woman working at laptop in sunlit home office" is a good agency title and a mediocre page title. Do not reuse one as the other.
Conceptual keywords that help on an agency - "freedom", "ambition", "new beginnings" - read as vague filler in alt text and provide no accessibility value. Keep them in the embedded keyword field and out of the alt attribute.
Filenames that help Google can leak information you would rather not publish, such as client names or shoot locations for private work. Rename before publishing to the web, not before delivering to a client.
A workflow that serves both without doubling the work
Write the description once, at export, as a plain factual sentence about what is in the frame. That sentence becomes the IPTC description, the agency title, and - lightly trimmed - the alt text on your own site. Derive the ranked keyword list from the same sentence by extracting subject, action, setting, concept and demographic terms, then ordering them by importance.
At small volumes this is entirely manual and takes a couple of minutes per image. Past a few hundred files it stops being viable, which is where automated metadata generation earns its place: Rastock AI writes the title, description and ranked keyword list, embeds them as IPTC and XMP, and delivers to agencies over FTP or SFTP, with banned and mandatory keyword rules so brand names and rejected terms never reach a submission.
Because output is exportable as CSV alongside embedded metadata, the same generated description can be pulled into a website CMS for alt text and captions rather than rewritten. Plans are tiered by monthly upload volume - see current pricing - with no revenue share and no ownership claim on files.
The mistakes that cost the most
Copying the same alt text across an entire gallery. Google sees forty identical descriptions and concludes it has one image, not forty.
Treating the keyword cap as a quota. Both agencies reject on metadata relevance, and irrelevant filler is exactly what that rejection targets.
Assuming AI-generated tags are submission-ready. They are usually accurate and usually literal, which means they describe the objects and miss the concept a buyer is actually searching for. Review the first ten, always.
Optimising images while ignoring the page they sit on. Image ranking is heavily influenced by the quality and relevance of the surrounding page, and no amount of alt text rescues a thin one. If you want to go deeper on the selection side, our piece on choosing the best keywords for stock photos covers how to build a term list from buyer demand rather than from what is visible in the frame.
Frequently asked questions
Does Google still use the meta keywords tag for images?
No. Google stopped using the meta keywords tag as a ranking signal many years ago and does not read an image keyword field at all. Image relevance is inferred from alt text, filename, caption, surrounding page text, page title, structured data and computer vision analysis of the image itself. Embedded IPTC keywords are read by stock agencies, not by Google ranking.
How many keywords should I use on a stock image?
Fewer than the cap. Adobe Stock allows up to 49 and Shutterstock requires between 7 and 50, but these are ceilings, not targets. Between twenty and thirty accurate, non-duplicated keywords usually outperforms a padded list, because irrelevant terms dilute relevance scoring and can trigger a metadata rejection at review.
Should alt text and stock agency keywords be the same?
No. Alt text is a short descriptive sentence written for people who cannot see the image, and Google reads it as natural language. Agency keywords are a ranked list read by an internal search index. Both should come from the same underlying description of the photo, but the formats serve different systems and should not be copied between them.
Does keyword order actually affect stock search ranking?
On Adobe Stock, yes - the platform states that the first ten keywords carry the greatest influence on search ranking, which makes ordering a real editorial decision. Lead with subject, action and primary concept, then follow with setting, viewpoint and demographic detail. Other agencies weight order less explicitly, but leading with the strongest terms is safe everywhere.
Can I put brand names in image keywords if the brand is visible?
Not for commercial licensing. Both Adobe Stock and Shutterstock treat trademarks, brand names and product names in commercial metadata as a rejection reason, and Adobe also refuses submissions that reference living artists' names in prompts, titles or keywords. Editorial content follows different rules, but commercial submissions should stay clear of marks entirely.
Do image optimisation plugins remove my embedded keywords?
Many do. Resizing and compression pipelines frequently strip IPTC and XMP data to reduce file size, which silently removes your copyright, creator and keyword fields from the published file. If attribution or licensing enquiries matter to you, download a live image from your own site and inspect its metadata rather than assuming the export survived.