Typing the word "cleaner" into the search box of online photo library Alamy produces a list of more than 9000 images - ranging from vacuum cleaners and bottles of Windowlene to tropical fish. Alamy has more than 20 million images and includes the work of specialist stock shooters (hence the tropical fish: Cleaner Wrasse, Labroides Dimidiatus), outtakes from photojournalists' assignments and an extraordinary number of holiday snaps of churches, castles and donkeys (15,717 at the last count). Without careful use of the search feature, finding what you want - if it's there - can take a very long time.

Getty Images has even more on sale, although fewer donkeys (4723). Most have some sort of news angle, but that's still a lot of donkeys. My own library is somewhat smaller (around 10,000 images), but is much more focused. It can offer only four donkey pictures, one of which is of Donkey Beach, with not a quadruped in sight.

Photo libraries have grown exponentially as a result of their migration online over the past 10 years. The big players - principally Getty and Corbis - have swallowed up many smaller collections, both general and specialist.

And sites such as Alamy and Flickr have made a world, once the exclusive preserve of professional photographers, accessible to anyone with a digital camera and an internet connection.

It has become much easier for both professionals and amateurs to get their pictures into a library, but lost in a vast ocean of images, getting them seen is another matter. John Harris, who runs Reportdigital, says: "Picture researchers can get quite disgruntled with the process at very large picture libraries, and often feel drowned in an ocean of mediocrity. It is not uncommon to hear picture researchers claiming that it is as difficult to find a decent picture now as it was 15 years ago."

So, what can photographers do to cheer up these unhappy potential clients? However strong an image, it is not its visual qualities that determine whether it gets found. For now, search engines cannot see pictures - only the words attached to them. Good captioning and keywording are crucial, but that's not the only strategy to pursue.

I can't tell who is searching for what in my library, but I can tell how they are going about it, and surprisingly few bother to use the search facility. Given the amount of time I spend captioning and keywording, this might seem rather disappointing.

Actually, I think it's a good sign. Most users come to the library because of its specialist focus on social issues. A list on the home page provides direct links to galleries covering easily understood subject areas. Each gallery displays images in date order, with the most recent first, and often it is only worth doing a word search if you're looking for something specific - a particular housing estate or a named individual. The rest of the time a visual inspection is quicker.

The Reportdigital website is structured in a similar way and covers similar subject areas. It represents a number of photographers and is significantly larger - currently hosting around 80,000 images - but still has the advantages of niche specialisation and allows for the narrowing down of the parameters of a search by topic and world region.

A search for "cleaner" produces 503 images, which include neither fish nor donkeys. The library does have donkeys (40), but the majority are from a series on nomads in Sudan, a subject that falls comfortably within its well-defined remit.

The big attraction of smaller specialist libraries, for both photographers and picture researchers, is that they are much easier to navigate. But captions and keywords are still essential. Without them, work uploaded to the huge generalist libraries would be invisible. For photographers shooting pictures intended for storage in these systems, entering data into the IPTC fields is one of the most time-consuming elements of the digital workflow.

Keywording strategies vary from photographer to photographer and from library to library. Some use as few as three or four, selected from a predefined set; others add synonyms and singulars, as well as plurals, "concepts", and even common misspellings, to produce lists of 50 or more.

So why are researchers disgruntled? John Harris suggests that some clients expect picture searches to work like Google, without recognising that Google interrogates a massively bigger set of text native documents.

Some searches fail because the keywording just isn't good enough, while others do because they are too specific and don't consider the likely intentions of the keyworder. Or they don't recognise that searching for an image has nothing to do with its visual content - it is entirely dependent on the words that have been used to describe it.

Some of these failings can be overcome with a little thought by the photographer. Here's a failing of my own: a search by an unknown user on my site for "fair trade in Caribbean" produced no results. Why? Because I had captioned the relevant photos using "Fairtrade" (all one word), the correct name of the scheme I had photographed, but had not also included the more commonly used "fair trade" as a keyword. However, it's worth noting that a less specific search would have worked: "trade in Caribbean" delivers 15 images, of which six are the Fairtrade photos.

Another example demonstrates that thought is required on both sides: "photo in 1981 Mozart Estate" produced nothing (it's a photo library, so why would I caption every photo "photo in"?), but "1981 Mozart Estate" does. And in a small collection, or with an unusual name, even that is more than is necessary. Obviously, I don't have any images of the composer, so "Mozart" on its own only brings up eight images of the estate, which are quick and easy to sort through.

What about the 9000 Alamy cleaners I began with? If you refine your search to images of "bank cleaners", Alamy offers you 26 (no fish), Getty 24 and I can offer 22. However, although a search for a singular "bank cleaner" produces the same results from my library, from Alamy you get 49, including the tropical fish and several images of no obvious relevance - probably the consequence of inaccurate or excessive keywording.

On Alamy, photographers do their own IPTC data entry, each using whatever method they choose. Some other libraries, particularly smaller ones, edit or control keywording and offer a greater degree of consistency.

For photographers there is a play-off. The smaller collections offer subject focus and ease of navigation, but a niche customer base, while the big ones have a mass audience, but the prospect of being a small fish in a very, very large pond.

Whichever you choose, writing captions and keywords - and carrying out a search to match them - requires a mixture of common sense and intuition on behalf of both the photographer and researcher. One thing technology has not changed: picture research is an art, not a science.

www.philipwolmuth.com