A curator pushed back on me last month. Her museum had spent three years rewriting every label in the building. Plain language. Multiple voices. A diversity of perspectives baked in. Tested with focus groups. The label program won an award. Her question, fair and direct: "Why on earth would I spend another forty thousand on an audio guide when the interpretation is already this good?"
The honest answer is that audio guides are still worth it, but not for the reasons she was imagining. Audio doesn't replace excellent labels. It does things labels structurally cannot do, no matter how well they're written. The two solve different problems. If you already have great labels, you've solved one problem completely. You haven't touched the others.
What labels are actually good for
Strong wall labels do four things well, and it's worth being precise about them before talking about anything else.
They reward the reader who's already standing there. Someone who has chosen to stop, lean in, and spend ninety seconds with an object gets a payoff: dates, materials, a context paragraph, maybe a provocative question. That's a beautiful interaction. No audio guide replicates the intimacy of reading a well-written label inches from the thing it describes.
They scaffold the visit at a glance. A visitor walking through a gallery scans labels the way you scan a bookshelf. The first sentence tells them whether to commit. Good labels are written so that the first six words do the work. That speed matters. Audio can't be skimmed.
They anchor the room. A label is physically attached to its object. There's no risk of the visitor listening to the wrong story while staring at the wrong vitrine. The pairing is forced by geography, which is a quiet kind of design genius.
They hold up across centuries. A printed label installed in 2019 still works in 2026. No batteries, no updates, no dependencies. Curators who came up through the label tradition are right to defend it. It's one of the most durable interpretive tools the field has ever developed.
What labels can never do
Now the asymmetric part. There are visitor needs that no label, no matter how brilliantly written, can meet. Not "doesn't meet well." Cannot meet at all.
Labels assume a sighted reader who is fluent in the language they're written in, has the visual stamina to read forty of them in a session, and is willing to break eye contact with the object to do it. Every one of those assumptions excludes a real share of your audience.
A blind or low-vision visitor cannot read a label. A visitor with severe dyslexia processes a 120-word label as a wall of texture, not text. A visitor whose first language isn't represented on your walls is functionally label-blind even with twenty-twenty vision. A seven-year-old can read but won't. A visitor in hour three of a long tour has reading fatigue and is glancing at labels rather than reading them — even great ones.
These aren't fringe cases. International visitors at a major European museum routinely run thirty to sixty percent of attendance. Roughly one in five adults has some form of literacy challenge. About four percent of any audience has a vision impairment significant enough to affect reading. Add up the overlapping groups and a non-trivial fraction of your visitors are getting nothing from your labels, regardless of how good those labels are.
There's a second category labels can't handle: depth. A label has, at most, 150 words to spend on an object. That's a hard constraint of the medium. Stand in front of any object in your collection that you could happily talk about for ten minutes and you'll feel the squeeze. Labels are extraordinary at compression. They are physically incapable of expansion.
And there's a third: response. A label can't answer a question. If a visitor wonders why a Madonna's face looks asymmetrical, why a particular bronze is green in some places and brown in others, why this vase has a hole in the bottom — the label either anticipated the question or it didn't. Most labels didn't.
Where audio fills the gap
Audio guides aren't a parallel translation of your labels. The good ones do specific things labels are bad at.
Eyes on the object, not on the text. This is the single biggest behavioral shift audio creates. A visitor with a label reads, looks up, reads again, looks up. A visitor with audio looks at the object the entire time the narration runs. Dwell time in front of works rises, often by fifty to a hundred percent. Visual attention is the whole point of a museum visit. Audio buys you more of it.
Length on demand. A good audio system lets the visitor choose: a forty-second snippet for the casual pass, a four-minute deep dive when something grabs them, a fifteen-minute interview with the artist when they've fallen in love. Labels can't tier like this. They have to pick a single length and serve every audience badly except one.
Languages without reprinting. This one is concrete and worth dwelling on. A 200-object label program translated into six languages requires roughly 1,200 individual label fabrications. Every time a curatorial team tweaks a label — and they will — every translation has to be redone, refabricated, and reinstalled. We've worked with museums whose translation refresh cycle ran 18 months and cost north of $60,000 per pass. A multilingual audio guide covers thirty to forty languages from one source of truth and propagates updates the same day. If your international share is meaningful, this alone justifies the system.
Audio that isn't narration. This is where curators with strong label programs often light up. Labels are text. Audio can be an oral history clip from the descendants of the people whose objects you're showing. It can be the voice of the artist describing their own work. It can be a fragment of music the painting was made to be seen with. It can be the actual sound of the thing — a temple bell, a war horn, an automaton mechanism. None of this fits on a label. All of it fits in audio. For object-rich collections with cultural depth, this is where audio earns its keep regardless of how good your labels are.
Access for visitors who can't read your walls. A well-designed audio program is, among other things, a working set of audio descriptions for blind and low-vision visitors. It's also the most realistic way to serve visitors with low literacy, dyslexia, cognitive disabilities, or simple reading fatigue. Labels treat reading as a default condition. A meaningful share of your audience doesn't share that default.
Answering the visitor's actual question. Modern AI-driven systems — see AI museum guides for the longer version — let a visitor ask the specific thing they're wondering about and get a real answer in real time. Why does this look like that. What was she holding originally. Who painted over the background. A label has to guess what the visitor will ask, and it has to guess once, in 1992 or 2019 or whenever it was last revised. An AI guide responds to the visitor in front of it. This isn't a small upgrade. It's a different category of interpretation.
When excellent labels make audio less urgent
Time to take a position the curator I was talking to actually agreed with. There are museums where adding an audio guide on top of a great label program is a marginal investment, not a strong one. Be honest about whether you're one of them.
If your audience is overwhelmingly local, reads English (or your primary language) fluently, skews adult, has low rates of vision impairment in your visitor data, and your labels are already producing the dwell times you want — the lift from adding audio is real but small. The visitors who'd most benefit from audio aren't the ones walking through your doors. The marginal cost might genuinely belong somewhere else: a conservation campaign, an evening program series, a new label refresh, a community partnership.
The same goes for highly specialized museums whose audience already arrives knowing the field. A research-focused decorative arts museum drawing scholars and serious collectors gets less from audio than a general-audience museum drawing first-time visitors. The labels are already pitched at the audience. Audio aimed at the same audience is redundant.
A small museum with a 30-minute self-guided route and great labels probably doesn't need audio. Audio shines in places where visitors would otherwise feel lost, overwhelmed, or excluded. A tight, intimate museum with strong text doesn't generate those conditions. Adding audio because it's the modern thing to do is a bad reason.
I'd rather a museum skip audio entirely than ship a mediocre audio guide because they felt obligated. A bad audio program actively damages the visit. Empty pedestals are better than ones holding broken sculpture. The same logic applies here.
How to think about the budget overlap
The framing that gets curators stuck is treating audio and labels as competing line items. They're not. They're different layers of the same interpretive stack, and they should be budgeted as complements.
A useful mental model: labels are your synchronous, in-place, opt-out medium. Audio is your asynchronous, on-demand, opt-in medium. The first reaches everyone who walks past an object. The second reaches everyone who chooses to engage more deeply, or who needs an alternate channel to engage at all.
If your annual interpretation budget is, say, $80,000 and you're currently spending all of it on labels and translations, the question isn't "should we cut labels to fund audio." It's "is the right split closer to 60/40 than 100/0." For most museums above a certain visitor count and above a certain international share, the answer is yes. The label program stays excellent. The audio layer adds the things labels can't do. Both get funded, neither gets gutted.
A practical sequencing point: don't try to translate your label corpus into audio one-for-one. That produces the worst kind of audio guide — flat, redundant, and competing with the labels visitors are already reading. Build audio around what audio is good at: longer narratives, oral history, multi-voice perspectives, interactive Q&A, languages your labels don't cover. Let the two media complement rather than overlap.
A second practical point: pilot it. Two galleries. Eight weeks. Real data on dwell time, language uptake, accessibility usage, visitor satisfaction. If the lift isn't there in your specific museum with your specific audience, you've spent ten percent of a full rollout to learn that. If it is there — and in our experience it usually is, even at museums with award-winning label programs — you have the evidence to fund the rest.
A short word to the curator who started this
If you've spent years getting your labels right, an audio guide is not a vote of no confidence in that work. The labels did what they were supposed to do. They reached the readers, they anchored the rooms, they earned the awards. None of that gets undone by adding a layer that reaches the people the labels never could.
The museums doing the most interesting interpretive work right now aren't the ones with the best labels or the best audio. They're the ones who stopped framing it as a choice. If you're weighing whether to add an AI audio guide to a strong label program, the framing worth holding onto is this: your labels solved one problem completely. The audio is for the other problems. They're real, they're measurable, and you can't reach them with text alone.
If you want to talk through what that overlap looks like for a specific collection, we'd be happy to walk through it with you. Platforms like Musa price the audio layer on a per-interaction or revenue-share basis, which means it doesn't compete with your label budget the way a capex procurement did — there's no upfront bill to reallocate away from fabrication, and the guide only costs the museum when a visitor actually uses it. In practice that makes the 60/40 conversation above a margin question rather than a budget fight.