Controlled Vocabulary Isn't Dead.
It Got a Promotion.
AI took over the tagging. So why do you still need a list of approved words? Because the list isn't for your team anymore — it's for your AI.
With AI offering to take over all the tedious tasks — the big ones and the small ones that have eaten our time for years and kept us from the creative work we actually enjoy — is it finally time to throw away the controlled vocabulary list?
Quick refresher on why it existed in the first place. The controlled vocabulary was built to make search consistent — so the same asset didn’t get filed under five different names by five different people. Picture one product shot of a knit top: one person tags it “sweater,” someone across the Atlantic tags it “jumper,” a third goes with “pullover,” and the freelancer who never opened the style guide calls it a “knit top.” Four tags, one item — and now every search for it misses three times out of four. Multiply that across hundreds or thousands of assets and “close enough” stops being good enough. You know something applicable didn’t surface, so you go spelunking to find it anyway. More time, more tedium. The exact thing the system was supposed to save you from.
Then came synonym search. Fuzzy search. Semantic search. And now the AI does the tagging for you.
So you’re good, right?
Wrong.
Give the robots their due
Let me hand the machines the win they earned, because they earned it. The original, number-one job of the controlled vocabulary — gray vs. grey, color vs. colour, t-shirt vs. tee vs. tshirt, the typos, the plurals — that job is basically done. Semantic and fuzzy search genuinely solved it, at least for the words that describe what’s in the asset. Search “jumper” and a good engine serves up the sweaters and pullovers too. If the only reason you kept a controlled vocabulary was to wrangle spelling and label the obvious, go ahead — let it go. You won’t miss it.
But that was never the hard part.
The job didn’t disappear. It changed.
It used to discipline your team. Now it disciplines your AI.
Think about what you just did when you turned on auto-tagging. You handed an automated system the keys and asked it to label thousands of assets — fast, at scale, with total confidence and zero knowledge of your business. That’s not a reason to throw away your guardrails. That’s the reason you need them more than ever.
Three things the machines still can't do for you:
AI tags what it sees. Not what you mean.
Point an AI at a photo and it'll nail the obvious — a woman, a beach, a blue sky, a coffee cup. Genuinely useful. What it will never know is that this asset is part of "Project Atlas," shot for the spring campaign, approved for paid social but not organic, and featuring a product you discontinued last quarter. That meaning lives in your business, not in the model's training data. The descriptive layer, AI's got. The layer that actually runs your operation — campaign names, product lines, approval states, brand terms — it can't guess, and it won't.
Inconsistency you can catch. A confident wrong answer you trust.
A teammate who forgets to tag is a gap you can see. An AI that hallucinates a clean, plausible, completely wrong tag is a landmine — because nobody goes back to check the machine. It looks authoritative, so you believe it. Ungoverned AI doesn't just tag unevenly; on anything ambiguous or industry-specific, it invents. A controlled vocabulary is how you tell it: these are the words that exist here, and these are the ones that don't.
Semantic search still needs something to be semantic about.
And this is the part the "taxonomy is dead" crowd keeps missing. Semantic search is brilliant at fuzzy. That's the gift — and it's the limit. When you want everything in the neighborhood of "beach," fuzzy is exactly right. But the moment you need an exact, defensible answer — every asset cleared for paid social, every shot from the spring campaign, every image of the product you discontinued last quarter and need pulled today — fuzzy is the last thing you want. You want a clean filter on a known term, every time, no guessing. That's not a job you hand to vibes and proximity. It's a list. The controlled vocabulary is that list — the thing that turns "we probably caught them all" into "we caught them all."
I’ve watched this play out
I’m not theorizing here. As an early employee at one of the first cloud-based DAMs, I spent years onboarding teams — the big ones and the small ones. The ones who skipped the vocabulary work didn’t get freedom. They got chaos with a better search bar. The tooling has gotten incredibly good since then. The principle hasn’t moved an inch.
So, do you need it? Will you use it?
Yes — and yes. Just stop thinking of it as the rulebook you force on your team. Keep it lean and let it do its new job: a short, sharp list of your non-negotiables — brand terms, campaign and project names, product lines, approval states — the words your business can’t afford the AI to get wrong. Let the AI handle “beach” and “coffee cup.” You handle “on-brand.”
You don't need a controlled vocabulary to discipline your team anymore. You need it to discipline your AI.
That tension — loving what AI can do and still keeping it disciplined — is exactly what we're building around at Vanday.
Stop losing the work you paid for.
Vanday auto-tags, crops and serves your library — free for 14 days, no credit card.