Keywords for Stock Photographers

Spend any time in stock photographer forum worlds such as Reddit and Quora, or Shutterstock and Adobe Stock Contributor Facebook groups, and you'll find plenty of great advice and certainly some very firmly held opinions about keywords.

I know. I’ve gone looking for them myself.

One of the most commonly held beliefs is that an experienced photographer is automatically the best person to keyword their own images.

I’m not entirely convinced.

Not because photographers don’t understand their pictures.

Sometimes the problem is that we know too much about the picture.

We know where we were, why we were there, what happened five minutes before we pressed the shutter and what we were trying to achieve when we took the shot.

Take these photographs, all accepted into our Shutterstock, Adobe Stock and Pond5 portfolios. They were shot just after the Covid lockdown restrictions began to ease.

As the photographer, I remember that it was the first time we’d been able to travel out of our local area for months. People were still having to follow social distancing rules. Friends were finally allowed to be outside in groups of 6. I also remember waiting patiently for these groups to leave the frame because we could only model-release ourselves and didn’t want to be ‘caught’ in a shot breaking the rules by being too close to another group.

Looking at the files, I originally included covid-related keywords including social distance, safe, wellness, wanderlust and achievement. They reflected both the circumstances in which the photograph was taken and, frankly, the fact that I had already had some of my editorial social distancing photographs bought and published on BBC news webpages. I was keen to capitalise on that and document as much as I could.

So at that time, the pandemic was central to the purpose of the shoots.

Five years later, a buyer may not need to know the circumstances in which the shoot took place, they may simply need a photograph of an older woman in walking boots enjoying the countryside. Perhaps to convey the message that it’s never too late to start looking after your health.

That’s why metadata isn’t simply about noting the photographer’s message, intentions or the events being documented.

It’s about describing what is actually in the image and anticipating ways potential buyers may search for it.

That can be quite a mindset shift when you start contributing stock images.

And it leads to another long-standing common belief from the experienced: "If the photograph is good enough, it will sell itself". Hmmmm...will it really?

Metadata is the currency that buys visibility in a crowded marketplace.
A perfect photograph is only as valuable as its discoverability.
Keywords convert creative energy into searchable digital assets.
They ensure that artistic genius isn't lost in a sea of millions.

Search technology has become increasingly sophisticated and stock agencies can analyse visual content in ways that simply weren’t possible years ago. We’ve seen quite noticeable advances in image analysis even during the 18 months we’ve spent developing SpeedData Pro.

Contributor metadata (titles, descriptions and keywords) attached to that image gives search systems explicit information about what an asset contains. But it all comes down to this;

The buyer types words.

The search engines scan and filter trying to match the buyer’s words with relevant content.

So even the most beautiful photograph still has to be found, amongst millions of others, by a search engine matching WORDS to WORDS. The sheer beauty of the photograph and the skills of the photographer do not enable it to rise out of the library on merit alone.

KEYWORDS HAVE A SPECIFIC AND CRUCIAL ROLE IN THE PROCESS

So if you have a little shop full of beautiful photographs, and you need the right buyers to find you, even though you are surrounded by huge art warehouses, this is how it works.

Categories bring potential buyers into the neighbourhood where your shop is located.

Keywords lead the potential buyers to your shop front and encourage them to knock on the door because what they’ve searched for suggests you may have the images they need.

The title makes the introduction, draws them inside and starts the sales pitch, quickly confirming that they’ve found what they’re looking for.

The description fills in all the details and does its very best to close the deal.

Then and only then does the photograph stand on its own two feet screaming out from page 1 or page 2 of the search results saying "LOOK AT ME, look how amazing I am!”

That will be the first time your photo has a chance to sell itself!

It’s not a perfect analogy, but it demonstrates the different jobs each piece of metadata performs.

But as always - there is one catch.

Your keywords can also bring the wrong people to your door.

Fifty visitors looking for something you aren't selling aren’t much use.

They’ll glance through the window and walk straight past.

You want buyers who are genuinely looking for the sort of image you actually have.

Even when every keyword in a list is relevant, they aren’t necessarily equally relevant.

This photograph is a useful example.

When I took the photograph, I assumed the person may have been homeless. Indeed when I captioned that image several years ago I included homelessness in its keywords

But that was just that, an assumption made by me.

There is nothing in the photograph itself that actually confirms it.

And that distinction matters when we’re creating metadata.

When I put the image through SpeedData Pro, it didn’t make that leap.

Its ten highest-priority keywords were:

The high-priority terms are remarkably literal.

They describe what can actually be seen rather than trying to invent a story about the person in the photograph.

The next layer begins to add more detail and context.

We move from the principal subject and obvious visual identifiers into surroundings, physical details and, cautiously, some concepts that the photograph might reasonably illustrate.

Then come the lower-priority terms, where the hierarchy becomes particularly interesting.

Retail is there.

There is clearly a retail environment in the photograph. There is a shopfront, window and signage behind the person, so the word has a legitimate connection to what we can see.

But SpeedData hasn’t expanded that into:

shopping · shopper · shopping experience · retail therapy · consumer · consumerism · shopaholic

Why would it?

That’s not what the photograph is fundamentally about.

And this is the difference between recognising something that exists within an image and allowing it to define the image.

A useful keyword hierarchy starts with the strongest evidence.

What or who can I actually see?

Then:

What are they doing?

Where are they?

What else is genuinely present?

Only then should we start moving into context and broader concepts that the photograph might reasonably illustrate.

SpeedData’s result demonstrates that progression rather neatly.

At the top we have person, red, coat, sitting, pavement.

Further down we reach hardship, vulnerability, solitude and isolation.

And lower again we find broader interpretations such as precarity, marginalisation and inequality.

Those later words may have value, but they shouldn’t be allowed to push the fundamental subject matter out of the way.

And there’s another important lesson hiding in what SpeedData didn’t say.

Homeless person isn’t there.

Homelessness isn’t there.

Rough sleeping isn’t there.

I may believe that’s what I photographed. I may even remember circumstances surrounding the photograph that confirm it.

But the image itself doesn’t establish it.

AI shouldn’t turn my assumption into somebody else’s fact.

If I have reliable contextual information that isn’t visible in the pixels, particularly with editorial photography, I can actually provide confirmed factual information that AI can’t verify in vision in SpeedData Pro’s Additional Information box which is then submitted before the metadata is generated.

That gives SpeedData information the photograph alone cannot provide and allows it to use that confirmed context when creating the metadata.

That is precisely why the photographer remains part of the process.

The software can analyse the photograph. The photographer can provide the story behind it

Both matter.

Then there is 24 hours

Now in the red-coat photograph, retail is part of the environment. It belongs, but it doesn't define the photograph.

But then there is 24 hours.

Literally, those are simply words on the sign behind the person. But to a buyer, they could open another conceptual route into the image.

24 hours can change everything. A day in the life. 24 hours is a long time. One day out of a lifetime.

Suddenly, something that began as a small visual detail can potentially become part of the photograph’s meaning.

That’s the interesting thing about conceptual keywords. Their importance isn’t necessarily determined by how much space something occupies in the frame.

It's determined by what the photograph can genuinely communicate to a buyer.

So far, we’ve established three things.

A photograph needs to be discoverable.

Different buyers may search for the same photograph in very different ways.

And more keywords are useful only for as long as those keywords remain relevant.

Now comes the part that used to consume an extraordinary amount of our time.

Actually producing them.

Every photograph needs analysing.

Every image in a batch is slightly different.

Ideally, every keyword list should reflect those differences.

Every term needs considering.

And if you submit to several agencies, as we do, different requirements, preferences and workflows have to be taken into account.

That is where shortcuts creep in.

We knew we were cutting corners.

Copy and paste became a coping strategy.

A keyword list that was excellent for the first photograph gradually became good enough for the fifth.

And perhaps rather questionable by the fifteenth.

One of the main reasons we built SpeedData Pro was that pain point.

We were fed up.

Filling in metadata for thousands of stock assets had become overwhelming, exhausting and, frankly, tiresome.

And keywords were probably the worst part.

Not necessarily difficult.

Just relentlessly repetitive.

Our first attempt at escaping that involved using general AI chats, including ChatGPT, to help create metadata for individual images.

It worked.

But we’d essentially swapped one labour-intensive process for another.

Easier in the sense that we were editing the output rather than creating it from scratch.

But it was still relentlessly repetitive.

The light bulb flickered.

What followed became an 18-month exercise in working out how to take the genuinely useful capabilities of this technology and combine them with what we’d learned through years of actually submitting stock.

Our question became: Could we automate the repetitive work without automating the photographer out of the decision?

That became the most challenging 18 months of our lives.

Rank First

We are constantly improving our system, especially when we get feedback from other stock contributors. For keywords, we've developed a system where SpeedData Pro produces a full set of relevant keywords for the individual asset and ranks them in order of importance.

The strongest identifiers come first.

Supporting context follows.

Broader conceptual and commercial terms come later.

That matters because a list of 50 equally weighted words isn’t particularly useful.

Then Review

This part matters to me.

SpeedData doesn’t generate the metadata and quietly send it off for submission.

It stops.

It shows you the results on screen.

Those results are presented as a colour-coded ranked list, so you can examine the keywords before anything is exported.

Nothing needs to disappear into a black box.

You can question a word.

Look at number 37 and think: Nope. That’s pushing it.

Or look at number 48 and realise: Actually, yes. I hadn’t thought of that.

That review matters because however sophisticated AI becomes, you still know things about your photograph that a system may not.

You know whether a location is correct.

Whether an interpretation is justified.

Whether a concept genuinely fits.

Whether something that looks like one thing is actually another.

AI can do the heavy lifting.

The photographer still gets the final say.

Then Export

Once you’ve reviewed the metadata, you export it in the format required for the stock agency , or agencies , you’re submitting to.

If there is something you want to change, this is where you still have control. The exported CSV can be edited before submission, allowing you to remove a keyword, change wording or add information that only you know.

And that is an important final distinction.

Creating good keywords and delivering them appropriately to different agencies are related problems, but they aren’t exactly the same problem.

Different platforms have different requirements and workflows.

The way that metadata needs to be organised, prioritised or exported may change according to its destination.

That’s why our aim with SpeedData has never simply been:

Let AI write some keywords.

It has been to take the repetitive part of a real stock contributor’s workflow, automate what can sensibly be automated, and leave the photographer with the part that actually benefits from human judgement.

That review screen is still one of my favourite parts of the software because it reinforces something I feel quite strongly about:

And it raises another question which I think is worth investigating further in a future edition of Metadata Matters.

Five years later, this photograph is still perfectly usable. But if I were uploading it today to my portfolios would I keyword it the same way? Absolutely not.

Where an agency allows existing metadata to be edited, there may be real value in revisiting older images, particularly those uploaded when you were less experienced, those tied to trends that have since disappeared, or simply good photographs that have never performed as well as you expected.

The same photograph. Four agency-ready outputs. SpeedData Pro re-generates platform-specific results for an image created during the pandemic, potentially revitalising the image's commercial potential for 2026.