Among the arguments for moving an audio program off the traditional studio-and-handset model, the one that gets the least airtime in vendor decks but the most pull in curator meetings is update cadence. Cost gets the spreadsheet. Multilingual gets the board slide. But the thing that actually changes the daily experience of running an audio program is whether a typo, a corrected attribution, or a reframed introduction is a Tuesday morning or a quarterly project.
I'm Eric Duffy. I run Convo, a phone-based platform in this category. This piece is about what AI changes about content cadence, what the operational discipline that comes with same-day updates actually looks like, and where over-frequent updates do real harm. If you're sizing up the category, this is one of the dimensions worth understanding before you sit through another demo.
Why was the traditional audio guide effectively frozen?
Because every correction was a production cycle, not an edit. A traditional audio tour is the product of a long sequential chain — drafting, casting, studio booking, recording, edit, sound design, mastering, multilingual versioning, and pressing to the device fleet. Any change to the artifact after launch has to re-enter that chain at the appropriate step, and most of those steps depend on a specific person being available on a specific day.
Re-recording a single stop usually means re-booking the original voice actor (often impossible six months later, when they're on another project or have moved cities), re-engineering the new audio to match the original room tone and mix, re-translating across each language version, and re-deploying to the handset fleet. Production cycles in the legacy model run on the order of months, and re-recording cycles run the same way. The cost of fixing one line is approximately the cost of producing one line.
The result is the open secret of the legacy model: most museums I've talked to have at least one tour stop they know is wrong but cannot justify fixing. A misattribution. A deaccessioned object the tour still describes. A wall label that was updated three years ago that the audio still contradicts. The tour and the gallery slowly disagree, and the institution has to choose which one tells the truth.
What does "same-day update" actually mean on an AI platform?
It means the bottleneck moves from production scheduling to curator review. On a modern AI-narrated platform, the production steps that took months in the studio model take minutes in software. A script edit produces new audio on demand. A re-translation across every language ships on the same render. The handsets are the visitor's own phones, so there's no fleet to update; the next QR scan loads the new file.
What's left is the work that doesn't compress. A curator still has to write or approve the new copy. A factual change still has to be checked against the source materials. A reframing still has to pass whatever editorial standard the institution has set for itself. Those steps are the same steps a museum has always had over its interpretive copy — they don't get faster, and they shouldn't.
So "same-day" is honest when the curator has the authority and the source materials to make the change today, and the platform doesn't add hours of latency on top of that. On modern platforms the platform side is seconds. The whole cycle is bounded by editorial review, which is exactly where it should be bounded.
What operational discipline does same-day editing require?
Four things, in order of how badly they hurt to live without: approval gates, version history, an audit log, and a clear rule about who can publish. Without those, the speed becomes a liability. With them, it's the headline operational improvement of the category.
Approval gates. Same-day doesn't mean same-person. A curator should still be the author of record; an editor or interpretation lead should still be the second pair of eyes; the institution should still have a defined publish step rather than a save-and-ship default. The best platforms make this configurable — draft, review, approve, publish — so the institution can decide which roles trigger which step.
Version history. Every published version of every script should be retrievable. When a visitor on a guided group hears one version and a visitor an hour later hears a different one, the institution needs to be able to answer the inevitable question: what did we say, and when did we change it.
An audit log. Who edited which line, when, and why. This is the artifact a director or board member needs when an external party asks how a corrected attribution made it onto a live tour. It's also the artifact that keeps curators honest with each other.
A clear publish rule. Whose name goes on the change. Who can override a previous curator's decision. What happens when two curators disagree about the same line. These are not platform questions; they are institutional policy questions that the speed of the platform forces a museum to actually answer.
For a related discipline question — keeping cadence sane while a temporary show is on the road — see the spoke on audio guides for traveling exhibitions.
What gets unlocked by being able to update in a day?
Three things that the legacy model effectively forbade. First, ongoing factual maintenance. An object gets re-attributed; the audio updates the same week. A date is corrected by a visiting scholar; the audio reflects it before the next tour cycle. The tour and the catalog can finally tell the same story.
Second, responsive interpretation. A current-events context — a new acquisition's provenance question, a political moment that changes what a piece means in the room — can be addressed in the audio rather than only in a hastily printed wall card. Curators get to use the audio layer as a place to think out loud about a collection, not just as a one-time recording that has to outlast its own context.
Third, multilingual parity. When a correction ships in English, the same render updates the Spanish, French, Mandarin, and other language tracks. The legacy model often left non-English tracks stale because re-recording them was a separate budget; the AI model treats them as the same artifact. See the related piece on revoicing tours across languages for the language side of the same operational story.
A reasonable proxy for whether your current vendor delivers this: how often does your audio tour get updated today, in any language? If the honest answer is "rarely," the cadence dimension is one of the biggest reasons to look at the category at all.
Where does same-day updating go wrong?
When the institution updates too often, with too little discipline, or for the wrong reasons. The risks are real and the vendor decks don't talk about them.
Loss of the canonical tour. A repeat visitor who took the tour last spring and again this fall should have a recognizable experience. If half the stops have been rewritten in between, the institution has effectively asked them to learn a new tour. Most museums want a canonical version that drifts slowly, not a feed.
Visitor consistency inside a single day. If a curator edits a stop at 11am, the visitor who scanned at 10:30 and the visitor who scanned at 11:15 hear different things. That's usually fine for a correction; it's confusing for a substantive rewrite. The rule worth setting: small fixes ship immediately, larger reframes ship after gallery close.
Curator burnout. The technical ability to keep tweaking a tour invites a kind of perfectionism that the legacy model's production cycle naturally prevented. A curator who is editing copy at 9pm because the platform lets them is using the platform wrong. The headline benefit of fast cadence is moving a correction off the queue, not adding a new continuous-editing job to a curator's week.
Substituting cadence for substance. A tour that updates monthly is not, by itself, a better tour than one that updates yearly. The cadence dimension is a means; the substance is the curator's. A platform that lets you ship a thin update every week is not delivering more value than one that lets you ship a careful update twice a year.
The honest framing: the goal of same-day cadence is to remove a constraint, not to set a metric. "How often did we update the tour?" is the wrong KPI. "Is the tour currently accurate?" is the right one.
What's the realistic update cadence for most museums?
Weekly drift, daily fixes. The pattern I see across institutions that use the cadence dimension well is roughly the same: substantive reframes and new stops ship on a weekly or biweekly cycle, gated by the institution's normal editorial review. Factual fixes — typos, mispronounced names, an attribution change a visitor flagged — ship the same day they're caught, often the same hour.
That cadence keeps the canonical tour stable enough for repeat visitors and casual returning audiences to recognize, while ensuring the tour is never demonstrably wrong for more than a day. It also matches how most interpretation teams actually work: an editorial lead reviews a queue weekly; the curator on duty handles urgent fixes as they surface.
For most institutions, the cadence question is best answered backward from the institutional review process they already have, not forward from the platform's technical capability. The platform should support whatever cadence the institution has decided it wants; it shouldn't dictate one.
Convo's changelog shows roughly how our own product cadence runs — small fixes ship daily, larger features weekly, with a clear audit trail. That's a useful model for what an editorial cadence inside a museum can look like too.
FAQ
Where this fits in the pillar
For the broader operations picture — how tours actually get made, kept current, and managed across temporary exhibitions, traveling shows, and multi-site institutions — see the operations pillar hub. For the related cadence question on temporary shows specifically, see audio guides for traveling exhibitions. For how cadence interacts with language coverage, the multilingual side of the same operational story is in revoicing tours across languages.
About the author
Eric Duffy is the founder of Convo, a platform that lets museums and cultural institutions publish multilingual audio tours their visitors can have a conversation with. He writes about the operational side of museum interpretation from inside the category — drawing on conversations with curators, interpretation leads, and directors at small and mid-size US museums about what actually changes when content can move at the speed of curatorial judgment rather than the speed of the studio. Reach him at eric@convo.app or on LinkedIn.