Travel content has a peculiar shape: extreme seasonality, itineraries that change constantly, and almost every product needing several languages. A tour operator running 30 itineraries across 5 languages, with prices and flight times shifting every season, cannot keep up with a studio-booking model.
This guide covers the three formats that matter — itinerary promos, on-site audio guides, and short-form ads — how each should sound, how to get place names right, how to plan languages, how to scale for peak season, and a pre-publish checklist.
Three formats, three voice strategies
The most common mistake in travel content is using one “tour promo” voice for everything. An audio guide needs to be slow; an ad needs drive. Those are opposite settings.
| Format | Length | Viewer state | Voice job | Sound |
|---|---|---|---|---|
| Itinerary promo | 60–120s | Comparing whether to book | Show what one day on this trip feels like | Warm, slightly brisk, cinematic |
| Attraction / trail guide | 60–120s per stop | Standing on site | Add the history and detail they can’t see | Slow, spacious pauses, steady level |
| Short-form ad | 15–30s | Scrolling, doesn’t know you | One visual plus one price or promise | Fast, hook first, lifted energy |
The shorthand: ads sell “I want to go”, promos sell “it’s worth it”, guides sell “now I understand”.
Voice settings by format
One voice model can produce all three textures with parameter changes. These are practical starting points, not standards.
| Parameter | Promo | Guide | Short ad |
|---|---|---|---|
| Pace (English) | 145–160 wpm | 115–130 wpm | 160–180 wpm |
| Pause between sentences | 200–300ms | 400–600ms | 150–200ms |
| Intensity (1–10) | 5–6 | 3–4 | 7–8 |
| Pitch | Steady, slight lift | Conversational, flat | Slightly higher, more movement |
| Music bed | Present, around −20 dB | Very light or none | Present, rhythmic |
| Ambience | Optional (markets, surf) | Optional, low | Usually none |
The fastest way to ruin a guide is to read it like an ad. The listener is walking, wearing one earbud, standing in front of the thing you’re describing. A pushy delivery gets switched off. The target is a knowledgeable friend talking next to them.
Place names: where AI voiceover exposes itself
Almost every failure in travel narration is a proper noun. Chinese scripts mixing English place names, English scripts mixing local-language names — one mispronunciation and the credibility is gone.
| Written | Common failure | Fix |
|---|---|---|
| Bali | Read as “BAY-lee” | English: /ˈbɑːli/; keep Chinese scripts in Chinese characters |
| Reykjavík | Letter-by-letter | Respell as “RAY-kyah-veek” for the English render |
| Machu Picchu | Stress on the wrong syllable | “MA-choo PEE-choo” |
| Angkor Wat | “ANG-ker” | “ANG-kor waht” |
| Hallstatt | Anglicized vowels | Keep the local spelling and test; consider a phonetic respelling |
| Kinkakuji | Half-Chinese, half-English | One language per script — no mixing |
| Mount Fuji | “Fuji Mountain” | “Mount Fuji” is the only idiomatic form |
| Dubai | “doo-BAY” | “doo-BYE” |
Three working rules:
- Never mix languages in one script. Write place names in the script’s own language, or romanise fully. Mixing forces the model to guess.
- Listen to the place-name lines first. After generating, replay only those sentences. Highest return per minute of QA in the whole workflow.
- Respell when precision matters. Rewriting “Reykjavík” as “Rey-kya-veek” fixes most engines; SSML techniques are covered in the SSML pauses and pronunciation guide.
Itinerary promos: a 90-second timecode template
A promo is not a list of attractions read aloud. It lets someone live one day in 90 seconds.
| Seconds | Section | Content | Voice note |
|---|---|---|---|
| 0–5 | Hook | One concrete image: “Six in the morning, and there’s nobody on the lake yet.” | Slow, leave air; don’t bury it in music |
| 5–20 | What this is | Days, region, who it suits | Three facts maximum |
| 20–50 | What a day looks like | One scene each for morning, afternoon, evening | Verb-first, visual |
| 50–70 | Differentiation | Why not a standard coach tour (small group, local guide, off-peak timing) | Specific and checkable |
| 70–85 | Practical info | Departure months, duration, what’s included | Slow down, item by item |
| 85–90 | CTA | “Full itinerary and pricing below” | One action |
The critical part is seconds 20–50 having the shape of a day, not “Day one… day two… day three…”. People remember the lake at dawn and the bowl of noodles in a side street, not the schedule.
The 30-second short-form cut
| Seconds | Content |
|---|---|
| 0–3 | Strongest visual + one line (price or seasonal window) |
| 3–12 | Three fast cuts, one sentence per location |
| 12–22 | One differentiation line + one concrete number |
| 22–27 | Who it suits (“for people who don’t want to rush”) |
| 27–30 | CTA |
Burn in captions on anything 15–30 seconds. Most first views are silent.
Audio guides: one stop, one file, 60–120 seconds
Guide structure is nothing like ad structure. What it needs is orientation — telling someone standing there where to look.
| Section | Seconds | Content | Example |
|---|---|---|---|
| Orientation | 0–10 | What you’re looking at | “The three-tiered tower in front of you is…” |
| What to notice | 10–35 | Three concrete details, directed by eye | “Look at the second tier of the eaves…” |
| Story | 35–80 | One person, one turn, one number | Who built it, why, how long it took |
| Practical | 80–105 | Hours, route, photo spots | Skip ticket prices — they change |
| Hand-off | 105–120 | One line to take away + next stop number | “Two hundred meters ahead is…” |
Implementation details that matter:
- One file per stop. Don’t cut a 20-minute route file; people need to skip and replay.
- Lead with the number. “Stop three” maps to a menu; “next we come to” doesn’t.
- Say critical directions twice. There’s wind, traffic, and other tourists.
- Never read ticket prices or timetables. Point to the official page, or record a swappable sentence.
Languages: choose a strategy before a count
Selling into five markets doesn’t require five voices. Pick the strategy first.
| Strategy | Approach | Cost | Fits |
|---|---|---|---|
| One voice, many languages | Same model across locales | Lowest | Brand tone over local flavour |
| One voice per language | Native voice per locale | Medium | The default for most operators |
| One voice per market | Split within a language (en-US/en-GB, zh-TW/zh-CN) | Highest | High-ticket, detail-sensitive products |
Three things to watch:
- Length inflates. Japanese and German often run 20–40% longer than Chinese for the same content. Leave slack in visuals and captions.
- Don’t translate literally. “Deep travel” means nothing in English; “small-group trips with local guides” does. See the multilingual product video workflow.
- Localize numbers and dates. 2026/3/15 is 3/15/2026 in US English, and it’s read differently too.
Language priority
| Priority | Language | Reason |
|---|---|---|
| 1 | Main source market | Usually 70%+ of revenue |
| 2 | English | Universal, and the search entry point |
| 3 | Second source market (e.g. Japanese, Korean) | Often higher spend per booking |
| 4 | Others | Add on real demand, not speculation |
Ship the top two properly rather than five badly.
Peak-season capacity: why travel fits AI voiceover
| Item | Studio voiceover | AI voiceover |
|---|---|---|
| Turnaround for one promo | 2–5 business days | Same day |
| Changing a departure date | Re-record, re-sync | Edit and regenerate that section |
| Adding a language | Find and audition a voice actor | Generate from the same script |
| Ten short videos in a day | Can’t book the studio | Batch generate |
| A tour added last minute | No time to produce video | Live within two hours |
The real bottleneck in travel isn’t the price of one video. It’s “eight more assets are needed the week before peak season.” That demand shape matches AI voiceover exactly.
A practical split: brand films and guides that need a genuine local accent and performance go to humans; itinerary updates, price changes, language versions, and short-form variants go to AI. Most mid-sized operators get a three-to-five-fold capacity increase from that split without dropping brand-film quality.
Lock the voice with voice cloning, batch generate with TTS, and size costs on the pricing page.
Accuracy and trust: four things travel content must not get wrong
| Item | Risk | Practice |
|---|---|---|
| Price | Expires; guests arrive to a mismatch | Don’t narrate prices, or say “see the official site” |
| Weather and season | “You will see snow” invites complaints | Use probability language plus a caveat |
| Visas and entry rules | Change often; errors cause real harm | Never narrate details; link to official sources |
| Safety info | Closed trails, cancelled ferries | Keep a swappable segment and show an updated date |
Two more trust notes:
- Don’t pretend to be local. If the narrator isn’t a native of the place, don’t write “I’ve lived here twenty years.” Write “your local guide takes you…”.
- Disclosing AI narration isn’t a penalty. A line in the description or about page — “narration generated with AI voice” — prevents confusion and reads as honest.
Music, ambience, and loudness
| Item | Guidance | Why |
|---|---|---|
| Narration loudness | Around −16 LUFS | Common social and streaming baseline |
| Music | 18–22 dB below narration | Phone speakers lose the voice first |
| Ambience | Between sections, not under sentences | Especially noticeable in guides |
| True peak | Under −1 dBTP | Avoids codec distortion |
| Licensing | Commercial licence, keep the receipt | Travel content gets reposted widely |
One guide-specific point: listeners are outdoors, in wind, on one earbud. Keep the midrange clear. Heavy low end for “immersion” just makes speech unintelligible.
Nine pre-publish checks
| # | Check | Pass condition |
|---|---|---|
| 1 | Place names correct | Replay every proper noun |
| 2 | Numbers and dates localized | Matches that locale’s convention |
| 3 | Understandable silent | Watch once with sound off |
| 4 | Captions match narration word for word | Spot-check five sentences |
| 5 | Prices and times removed or caveated | Search the script for digits |
| 6 | Language versions aligned | Within 10%, or visuals adjusted |
| 7 | Loudness consistent | All around −16 LUFS |
| 8 | Guides: one file per stop, numbering correct | Listen in route order |
| 9 | Updated date shown | In filename and on page |
Five common mistakes
| Mistake | Consequence | Fix |
|---|---|---|
| Reading guides in ad voice | On-site listeners switch off | 115–130 wpm, intensity 3–4 |
| Mixing languages in one script | Model spells names out letter by letter | One language per script, or phonetic respelling |
| Launching five languages at once | All five half-done | Ship the top two markets first |
| Narrating exact ticket prices | Out of date, complaints | Point to the official site |
| One 20-minute route file | Can’t skip or replay | One stop, one file, 60–120s |
FAQ
Do guides have to be human-recorded?
No. For history, orientation, and route guidance, AI narration is hard to distinguish and far cheaper to update. Reserve human recording for guides that depend on strong personality or an authentic local accent.
Can AI pronounce local place names?
Usually, with a test. Respell the name phonetically in the target language (e.g. “Rey-kya-veek”), regenerate, and listen. Don’t ship the raw spelling unchecked.
How many languages does one itinerary need?
Look at revenue. The top two source languages usually cover over 70% of demand; add the rest when there’s real enquiry volume rather than diluting quality upfront.
Should the video state prices?
Preferably not, or not definitively. Travel pricing changes constantly; “full pricing and departure dates on the site” costs the least to maintain.
Can one video serve both short-form and the website?
Yes, but recut. Short-form needs the hook earlier, a shorter runtime, and burned-in captions. Reusing the website cut usually loses viewers in the first three seconds.
What if peak season arrives before production?
Build the templates first: a 30-second hook, a 90-second itinerary skeleton, and a fixed guide structure per stop. With those in place, swapping visuals and script produces same-day output — which is exactly the point of AI narration in peak season.
Travel voiceover isn’t really about a beautiful voice. It’s about whether the voice helps someone imagine being there. A promo gives a day its shape, a guide gives the site its orientation, an ad gives a reason to tap. Doing those three separately and well beats one video trying to do all of it.


