AI video for audiology clinics: caption everything
Your audience cannot rely on your audio track. Captioning and split-screen rules, an OTC-versus-fitted explainer structure, and claims to keep off camera.
Every other clinic can treat captions as a reach optimisation. An audiology practice cannot. Your prospective patient is defined by difficulty hearing, and a meaningful share of the people who watch your video will get nothing usable from the audio track no matter how well you mixed it. If the information only exists in the voiceover, you have made a video that excludes the audience it was made for.
That single fact reorganises the production. Captions stop being a post step and become the primary channel. The speaker's face stops being a stylistic choice and becomes an information source. And much of the standard social-video playbook — music-led pacing, cutting away from the talking head to b-roll — turns out to be counterproductive here.
Two rules that are not negotiable
Burn the captions in. A sidecar file is a setting the viewer has to know exists and remember to turn on, and on most social surfaces it is off by default or unavailable. Burned-in captions are part of the picture; they cannot be turned off or lost when someone re-uploads your clip. Keep the sidecar too — it is what search engines and platform indexes read — but the burned version does the work. Trade-offs in burned-in captions vs SRT and VTT sidecars.
Keep the speaker's face on screen the entire time. Many people with hearing loss lipread, consciously or not, and a lot of them do it well enough that it materially changes comprehension. The standard explainer move — cut from the presenter to a diagram, hold on the diagram for eight seconds — removes that channel exactly when the information is densest. The fix is not to lose the diagram. It is to run both at once: picture-in-picture with the speaker in a corner, or a true split screen with the presenter on one side and the visual on the other. Generated b-roll belongs underneath the speaker, not instead of them.
Caption craft matters more here than anywhere else. Two words per line with a new card every syllable works on entertainment content and fails on clinical explanation, because the viewer is reading for meaning rather than rhythm. Break lines at clause boundaries, time cards to the edit beat rather than raw speech-recognition word boundaries, and hold each card long enough to read comfortably — specifics in caption readability. High contrast, no thin fonts, and never place a card over a moving background.
The OTC-versus-fitted explainer
This is the highest-value asset an audiology practice can build, because it is the question that stops people from booking. Since the FDA created an over-the-counter hearing aid category, adults with perceived mild-to-moderate hearing loss can buy devices without a prescription, an exam, or a fitting. A lot of prospective patients now assume that means clinics are obsolete, and a lot of clinics respond by attacking OTC, which reads as self-interested and converts badly.
The structure that works is comparative rather than adversarial. Six beats, three to five minutes total, one speaker on camera throughout:
- Name both paths in the first ten seconds, on screen as text, not just spoken. "Over-the-counter" and "fitted by an audiologist" as two labelled columns that stay visible.
- Who OTC is designed for. Adults, self-perceived mild-to-moderate difficulty, comfortable self-fitting and self-adjusting. Say this plainly and without hedging. Credibility here is what buys attention for the rest.
- What the fitted path adds. A full audiologic evaluation, a device programmed to the individual's audiogram, verification that the device is actually delivering what it was programmed to, and scheduled follow-up adjustment. Show the equipment; the booth and the probe-tube setup are unfamiliar to most viewers and unfamiliar is interesting.
- The conditions that need a physician, not a device. Sudden change in hearing, hearing loss in one ear only, ear pain, drainage, dizziness. Frame it as routing, not alarm: these are the things a clinic is trained to send onward.
- What happens if OTC doesn't work out. Someone who gives up on a device often writes off the whole category rather than trying another route. Being the practice that says "bring it in and we'll tell you whether it's the device or the fit" is a genuine differentiator.
- Close on a booking action with the phone number and the practice name held on screen for a full five seconds — long enough for someone to photograph it.
Run this as your pillar asset, then cut each of the six beats out as a standalone short. The pillar sits on the website and the Google Business Profile; the shorts feed social. One production, seven deliverables.
Claims to keep off camera
Hearing devices are regulated medical devices and hearing-related advertising draws scrutiny from both device regulators and general truth-in-advertising enforcement. Beyond the legal exposure, the specific claims below are the ones that make an informed viewer close the tab.
- "Restore your hearing." Amplification is not restoration. Say what a device does: makes speech easier to follow in specific situations.
- "Cure tinnitus." Management, masking and habituation are defensible words. Cure is not.
- "Hear like you did at 20." An outcome promise dressed as a slogan.
- Unsubstantiated satisfaction or success percentages. If you cannot show the study and its method, do not put the number on screen.
- Comparative claims against a named competitor device or a named retail chain. High risk, low upside.
- Price claims without the conditions. "Starting at" without what it starts at, or a bundled price without what is bundled, is the complaint that actually gets filed.
- Real patient audiograms, ear photographs, or booth footage of an identifiable person. A practice that bills electronically for care is handling protected health information, and a patient's test result is exactly that. Use generated or schematic imagery for anything that looks like a chart or an ear canal.
Generated humans need one more line drawn around them. A generated person is fine as ambience — a figure in a waiting room, hands adjusting a device. A generated person who appears to be a satisfied patient describing their results is a fabricated testimonial, and it is treated as one regardless of how it was produced. See the fake review rule and AI testimonials for where that line sits, and label the synthetic imagery you do use.
What to generate, and what to film
Generated material carries the visual load. Filmed material carries the credibility.
| Asset | Source | Why |
|---|---|---|
| Ear anatomy and how sound travels | Generated | No patient involved, no PHI, endlessly reusable |
| Device close-ups and scale references | Generated or product shots | Avoid implying a brand relationship you don't have |
| Clinic exteriors, waiting area, parking, accessibility | Filmed empty | Real, specific, and answers a practical question |
| The audiologist speaking | Filmed | This is the trust asset; do not synthesise it |
| Booth and verification equipment | Filmed empty | Unfamiliar and genuinely interesting |
| Environments the video describes | Generated | Restaurant noise, a busy pavement, a family dinner |
That last row is where generated footage earns its place. Explaining why speech-in-noise is harder than speech in a quiet room is much easier with a shot of a crowded restaurant than with a slide about signal-to-noise ratio, and shooting a crowded restaurant is a permissions problem. The AI B-Roll Generator covers that class of shot, and the explainer video generator handles the anatomy sequences.
Still mix the audio properly
It would be easy to conclude audio doesn't matter here. It matters for a different reason: a large share of the people watching are family members. The adult child researching on behalf of a parent is often the one who books, and they watch with sound on.
Mix for them. Keep the voiceover forward, duck any music well under the speech rather than sitting it at a flat level, and deliver at around −14 LUFS integrated, the working target for social and YouTube playback. If you are generating audio, exclude reverb explicitly — it turns up uninvited and it is the worst thing for speech intelligibility. And if a music bed adds nothing, cut it.
FAQ
Do I need an audio description track as well as captions?
It is worth considering for the pillar asset. Captions solve the audio-to-text direction; audio description solves the reverse, for viewers with low vision, and hearing clinics see plenty of older patients with both. It is not the first thing to build, but if you are producing one long evergreen explainer a year, that is the asset that justifies it — the case is laid out in audio description for brand video.
Can I use an AI avatar instead of putting the audiologist on camera?
For the pillar explainer, no. The face on screen is the trust instrument, and viewers who lipread are watching that face closely enough to notice when the mouth shapes don't match the words. For utility content — opening hours, directions, a seasonal notice — a synthesised presenter with a clear disclosure is reasonable. The distinction is whether the video is asking for belief or delivering a fact.
How long should the captions stay on screen?
Long enough to read comfortably at a slower-than-average pace, which in practice means fewer, longer-held cards than a typical social edit uses. If a card is gone before you can read it twice, it is too fast for this audience. Timing captions to the edit beat rather than to raw word boundaries makes them easier to follow and easier to keep on brand across a library — see keeping captions on brand.
Should the OTC explainer live on social or on the website?
Both, as different cuts. The website and Google Business Profile get the full three-to-five-minute version, because someone who searched for this will sit through it. Social gets the six single-beat shorts, each ending on the same booking action. Do not put the long version on a feed and expect it to hold.