Keywording for stock photography that actually sells

For a stock photographer, keywords are not metadata — they are distribution. An unkeyworded image on a stock agency is an image that does not exist, because the only way a buyer finds it is by typing words that match your keywords. Get them right and you sell. Get them wrong, or leave them thin, and a technically better photo than yours sells instead.

This is the part of the job nobody enjoys and everybody underestimates. Here is how to do it well, and how to do it fast.

How stock search actually works

A buyer types a phrase. The agency matches it against your keywords, weighs relevance, freshness, past sales and a dozen other signals, and ranks the results. You are competing with everyone else who shot a similar frame.

Two things follow immediately:

• A keyword you did not add is a search you cannot win. If you did not tag “candid”, you will never appear for “candid”.
• A keyword nobody searches is dead weight, and enough of them can actively hurt you — several agencies penalise obvious keyword padding, reading it as spam.

So keywording is a balancing act: cover every search a real buyer might run, and not one word more.

How many keywords?

The honest answer is “enough, and then stop”, but a working range helps: most successful stock keyword sets land somewhere between 15 and 40 terms. Below fifteen you are almost certainly missing real searches. Above forty you are almost certainly adding synonyms and stretches that dilute your relevance and risk a padding flag.

The instinct to add fifty keywords “to be safe” is exactly backwards. A tight, accurate set of twenty-five beats a bloated set of fifty, because relevance is a ratio: every irrelevant keyword lowers the average.

Literal, then conceptual — buyers search both

The keywords that matter fall into two layers, and beginners only do the first.

Literal — what is visibly in the frame. woman, laptop, coffee, office, window. Easy, necessary, and low-value on its own, because everyone shooting a similar scene has the same literal tags.

Conceptual — what the image is about and what it could sell. remote work, work-life balance, entrepreneurship, burnout, monday morning, solitude. This is where sales come from, because this is what art buyers and marketers actually search. A laptop-on-a-desk photo is a commodity; the same photo tagged “freelance lifestyle” and “digital nomad” is findable by the person with a budget.

The literal layer is exactly what an AI classifier is good at, and exactly what bores a human to tag. The conceptual layer is where your judgement earns its keep — and it is invisible to a general vision model, which sees an empty office, not “remote work”.

Ordering matters (on some agencies)

On several agencies the order of your keywords carries weight — the first ones are treated as most relevant. So lead with your strongest, most specific terms and let the generic ones trail. “Golden retriever puppy” comes before “dog”, “animal”, “pet”. Check your primary agency’s rules; some ignore order, some lean on it heavily.

Model and property releases change what you can tag

If you tagged “commercial” or added conceptual lifestyle keywords to an image of a recognisable person with no model release, you have created a licensing problem, not a sale. Keyword honestly about what the image can actually be licensed for. This is not a search-optimisation point, it is a “don’t get the image pulled” point.

Getting through a batch of 500 without burning out

A single shoot can produce hundreds of keepers, and keywording them one blank field at a time is how photographers come to hate the business side of their own work. The workflow that survives contact with a real backlog:

1. Auto-generate the literal layer. Run the batch through an AI pass to get the “what is in the frame” keywords for every image at once. PicsTag does this in your browser — drop in the folder, or feed a CSV of URLs — and exports the results as CSV or JSON. Your images never leave your machine, which matters when the shoot is under embargo or you simply do not want it on someone’s server.
2. Curate, don’t accept blindly. The model will be confidently wrong sometimes (it will call a husky a wolf). Use the confidence score: accept the sure ones in bulk, glance at the rest. The tool exports only the tags you accepted.
3. Add the conceptual layer yourself. This is the ten minutes per shoot where you, the human who knows what the image is for, add the terms that actually sell it.
4. Trim. Before you upload, cut anything that is a stretch. Twenty-five accurate keywords beats forty hopeful ones.

The mistake that costs sales

The biggest keywording error in stock is not too few keywords — it is keywords that describe the photo instead of predicting the buyer. “Woman, cup, table, sitting” is a description. “Morning routine, mindfulness, slow living, self-care” is a set of searches a buyer with a licensing budget will actually run. The camera captured the first list. Your job is the second, and it is the only part of keywording a machine cannot do for you.

For the underlying method — controlled vocabulary, the four layers, clearing a backlog — see keywording a photo library people can actually search.