A portfolio that gets approved but never earns is usually failing at one of four points, and they are not interchangeable. Either buyers never see the files, or they see them and choose someone else's, or nobody is shopping for that subject at all, or the portfolio is still too small for agency search to have anything to rank.
The distinction matters because the fixes cost wildly different amounts. Re-keywording three thousand files takes days. Shooting a new subject takes weeks. Waiting out a portfolio that is simply young costs nothing. Contributors routinely do the expensive thing first because it feels like action, then discover the problem was somewhere else. The sequence below is ordered by how cheap the diagnosis is, not by how likely each cause feels.
Start by separating discovery from desirability
Most contributor dashboards report views or impressions alongside downloads. That one pair of numbers splits the problem in half before you change anything. If a file has accumulated hundreds of views and no downloads, buyers are finding it and rejecting it, which is a competition or quality problem. If it has almost no views at all, buyers never reached it, which is a metadata or demand problem. Those two situations look identical from the earnings report and have almost nothing in common underneath.
Agencies expose this differently and some barely expose it at all, so check what your specific dashboards offer before assuming the data is missing. Where impression counts are not available, the closest substitute is to search the agency yourself using the exact phrase a buyer would type, and see how deep you have to scroll before your file appears. If it is not in the first several pages for a query it genuinely answers, treat that as a discovery problem.
Failure mode one: the file never enters the results
This is the most common cause and the cheapest to fix, which is why it belongs first. Agency search engines do not look at your photograph the way a buyer does. They match the buyer's typed query against your title, description and keyword list. Whatever is not written down does not exist as far as the index is concerned.
So the question of whether keywords matter on stock photos has a blunt answer: they decide whether the image is ever shown, and the image only decides what happens after that. A technically excellent frame with eleven generic keywords is invisible. The usual failures are narrow rather than dramatic - a keyword set that describes the objects in the frame but not the concept a buyer is shopping for, titles written as file names, descriptions left empty, or the same block of keywords pasted across an entire shoot so every file competes with its own siblings.
The diagnostic is straightforward. Take ten files that have earned nothing, and for each one write down the phrase you believe a buyer would type to want that image. Then check whether that exact phrase appears in the title, the description or the keywords. If it does not, you have found the problem without needing any other explanation. If your agency reports which search terms led to your downloads, the terms that already drive your downloads are the most reliable guide to what the rest of the portfolio is missing.
Failure mode two: buyers see the file and choose another
High views with no downloads is a different and more uncomfortable diagnosis, because the metadata is working and the image is the thing being rejected. The honest test is to run the query yourself and look at your file sitting in the grid next to the ones around it, at thumbnail size, the way a buyer under deadline actually sees it.
The reasons a file loses at that stage tend to repeat. It has no copy space, so a designer who needs to place a headline cannot use it. It is composed for the wall rather than for a layout - beautiful, and unusable. It reads as dated through wardrobe, devices or interior styling. Or it is competent but interchangeable with two hundred near-identical frames, in which case nothing about it gives a buyer a reason to click it rather than the one above it.
There is also a market shift underneath this. As generated imagery has flooded generic categories, buyers in advertising have become noticeably more sensitive to work that looks synthetic or over-polished, because a brand that needs to look credible cannot afford imagery the audience distrusts. Files that read as anonymous and perfect now compete against an effectively unlimited supply of the same thing. Specificity is the defence.
Failure mode three: nobody is buying that subject
Some portfolios are keyworded well, shot well, and aimed at nothing. Stock buyers are overwhelmingly commercial - marketing teams, publishers, agencies, product and editorial work. What sells is what those people need to illustrate, which is why business, healthcare, technology and everyday lifestyle situations keep absorbing volume while sunsets, flowers and travel landscapes are permanently oversupplied. The question of what stock photos sell best is really a question about who is holding the budget.
You can test this without shooting anything. Search your own subject on the agency where you sell and look at the result count. A query returning hundreds of thousands of files tells you supply is enormous; it does not tell you demand is. Then invert the test and search for the thing you would need if you were writing an article in your own field - a specific profession, a specific piece of equipment, a specific process - and see how thin the results are. The gap between those two searches is where stock photo niches that sell usually live.
This is also the one failure mode that no metadata work can rescue. If the demand is not there, better keywords deliver your file faster to an audience that was never going to license it. Recognising that early is the difference between changing what you shoot and spending another quarter optimising files that had no buyer.
Failure mode four: the portfolio is too small to measure
Microstock earnings are a volume business, and the arithmetic is unforgiving at small scale. A per-download royalty measured in cents means a portfolio of a hundred files can be perfectly good and still produce a result indistinguishable from noise for months. Contributors read that silence as failure and quit, when the honest reading is that there was not enough sample to conclude anything.
A sensible stock photography portfolio size strategy accepts this and plans around it rather than fighting it. Judge the portfolio in batches of a few hundred files rather than image by image, keep uploading while you evaluate, and spread the same work across several agencies so that a weak month on one is offset elsewhere - the agencies worth submitting to differ enough in buyer base that identical files perform unevenly across them. Distribution is one of the few levers that costs you nothing in shooting time.
How long before silence actually means something
Agency search ranking is partly historical. A file with no downloads has no performance record to rank on, so it starts deep in the results and stays there until something surfaces it - a long-tail query nobody else answered, a seasonal spike, a buyer who scrolled further than most. That first sale is disproportionately important because it feeds the signal that produces the next one.
Practically, give a new file three to six months before you draw conclusions about it, and judge the batch rather than the frame. Two hundred files across several subjects producing nothing after six months is a real signal worth acting on. Two files producing nothing after six weeks is not information at all.
The order that saves the most time
Check the age and size of the portfolio first, because it is free and it stops you from acting on noise. Then check views against downloads, which takes minutes and splits the problem cleanly in two. If views are low, audit metadata on a sample of ten files before touching the whole library. If views are high, put your files next to the competition at thumbnail size and be honest about what you see. Only after all of that should you conclude the subject itself has no buyer, because that is the conclusion that costs the most to act on.
What automation fixes, and what it does not
Only the first failure mode is a workflow problem, and it happens to be the one that scales worst by hand. Rewriting titles, descriptions and keyword sets across an archive built over several years is the kind of task people postpone indefinitely, which is exactly why so many portfolios carry metadata written in a hurry years ago. If that describes yours, re-optimising an old portfolio is usually higher-yield than shooting anything new, because the images already passed review and already exist.
That is the job Rastock AI is built for: reading the files, writing agency-compliant titles, descriptions and keywords, and delivering the results to more than ten agencies over FTP or SFTP from one place. It applies each agency's banned and mandatory keyword rules rather than pushing one identical set everywhere, and changes can be rolled back if a batch turns out wrong. There is no revenue share and no ownership claim on your files, and metadata exports freely as IPTC, CSV or XML. You can see how the metadata and delivery pipeline works before committing to anything.
What no tool can do is create demand for a subject nobody is shopping for, or make an interchangeable frame worth choosing. Metadata gets the file in front of the buyer. Everything after that is the photograph. If discovery is the half you want to fix, there is a 14-day free trial with no card required; if the diagnosis pointed at demand or composition instead, spend the time behind the camera rather than in a spreadsheet.
Frequently asked questions
How long should I wait before deciding my stock photos are not selling?
Give a new file three to six months before judging it. Agency search ranks partly on download history, so a file with no sales has no record to rank on and sits deep in results until a long-tail query or a seasonal spike surfaces it. Judge batches rather than single frames: two hundred files across several subjects earning nothing in six months is a real signal. Two files in six weeks is not.
Do keywords matter on stock photos, or is the image all that counts?
Keywords decide whether the image is ever shown, and the image decides what happens afterwards. Agency search engines match a buyer's typed query against your title, description and keyword list rather than looking at the picture, so a strong file with thin or generic metadata never enters the result set at all. Both halves have to work, but metadata comes first in sequence.
How do I tell whether it is a keyword problem or an image problem?
Compare views against downloads in your contributor dashboard. Many views with no downloads means buyers are finding the file and choosing something else, which is a competition or composition problem. Almost no views means buyers never reached it, which is a metadata or demand problem. Where impressions are not reported, search the agency for the phrase a buyer would type and see how deep your file sits.
What kinds of stock photos actually sell?
Stock buyers are mostly commercial - marketing teams, publishers, agencies and editorial desks - so the work that keeps absorbing volume is business, healthcare, technology and everyday lifestyle situations that can carry a headline. Sunsets, flowers and generic travel landscapes are permanently oversupplied. Specific, unglamorous subjects from a field you actually know tend to face far less competition than broad aspirational themes.
Will better keywords fix a portfolio nobody is buying?
No. If there is no commercial demand for the subject, improved metadata only delivers the file faster to an audience that was never going to license it. Metadata fixes discovery problems, which is the most common cause and the cheapest to correct, but it cannot create a buyer. That is why demand should be the last conclusion you reach, after age, portfolio size and metadata have been ruled out.
Does uploading the same photos to more agencies help?
Usually yes, unless you have signed an exclusivity agreement. Agencies serve different buyer bases, so identical files often perform unevenly across them and a weak month on one is offset elsewhere. It costs no additional shooting time - only the delivery work, which is the part worth automating once the same batch has to reach several destinations.