Artificial intelligence has brought speed and efficiency to keywording that would have been unimaginable a decade ago, but still we get AI keywording mistakes.
Modern systems can scan thousands of images in minutes, automatically tagging them with lists of relevant words. But while the technology is impressive, it isn’t foolproof. AI can get keywording wrong — and when it does, the consequences reach far beyond a few misplaced terms.
AI keywording works by recognising patterns. It analyses shapes, colours and textures, then predicts what the image probably contains. This is ideal for basic identification — “dog,” “car,” “beach.” But AI doesn’t truly understand context or meaning. It can’t tell whether that “car” is a taxi, a classic vehicle, or a police cruiser in a crime scene. Those distinctions matter to buyers and archivists, but not to algorithms.
That lack of nuance leads to predictable AI keywording mistakes. A photo of a journalist interviewing a politician may be tagged as “meeting, conversation, people.” Technically correct, but commercially useless. In a library search, those images will never surface for “press conference” or “election coverage” — the terms real clients use.
AI also tends to over-tag. Because it’s programmed to describe everything visible, it often generates long, cluttered lists such as “person, people, outdoors, nature, lifestyle, travel, day, holiday.” That might sound thorough, but it dilutes relevance and makes search results noisy and inconsistent.
Then there’s the risk of mislabelling. If AI mistakenly links a face with a known public figure or tags an image of a peaceful demonstration as a “riot,” the result can damage reputations — or worse, create legal exposure. AI doesn’t know when a word carries consequences; it simply outputs what seems statistically likely.
At Picsell Media, we see AI as a powerful assistant, not a replacement for professional judgement. Our AI Mechanic system produces fast, consistent baseline tagging, but every batch passes through our AI Professional editors for review. They refine order, remove redundant or inaccurate words, and ensure each keyword genuinely describes the image in line with our clients’ standards.
The result is accuracy with efficiency — automation guided by expertise. AI provides the speed, while humans ensure precision and responsibility. In keywording, that partnership isn’t optional; it’s essential.
Because when it comes to describing images, getting it nearly right isn’t good enough.