A fundraising video isn’t selling a product. It’s asking someone to believe in an outcome they can’t see yet. The voice has a specific job: make the problem understandable in thirty seconds, make the work credible in sixty, and make the next action obvious in the last fifteen.
AI voiceover lets a two-person team ship a video every week. It’s also the easiest place to lose trust: too smooth, too bright, too much like an ad. This guide is about making AI narration credible for fundraising — and about the lines that shouldn’t be crossed.
Three jobs, in a fixed order
| Job | Viewer’s question | What the voice does | Common failure |
|---|---|---|---|
| Understand | “What is this problem?” | Slow, concrete, opens with one person | Opens with mission statement and founding year |
| Believe | “Can they actually do this?” | Steady, slightly imperfect, no grand claims | Piles on adjectives, promises transformation |
| Act | “What do I do now?” | Specific, single, slowed down | Three donation tiers and two links at once |
Reversed order is everywhere: “We’re a foundation with twenty years of history” as the first line, and the viewer is gone by sentence two. Person first, number second, organization last.
The 60-second structure, second by second
| Seconds | Beat | Content | Voice direction |
|---|---|---|---|
| 0–5 | Hook | One concrete scene, or a line someone actually said | Even volume, no music; let the pause do the work |
| 5–20 | Problem | Who, where, what, how many | 95% rate, one fact per sentence |
| 20–35 | What you did | One verifiable action and its result | Start with verbs; avoid “committed to” |
| 35–50 | Impact | One specific change plus one number | 300 ms before the number |
| 50–60 | Ask | One CTA, one amount, one deadline | 90% rate, no music under the final line |
The 90-second version isn’t the 60 stretched. It adds a “why now” beat — season, policy, urgency. Don’t buy time by slowing down. That just sounds like stalling.
Four settings that build trust
Trust comes from consistency and restraint, not warmth.
| Parameter | Recommendation | Why |
|---|---|---|
| Rate | 130–150 wpm English, 190–230 characters/minute Chinese | Slower than an ad, slightly slower than news; room to absorb |
| Pauses | 400–600 ms between sentences, 800 ms between sections | Fundraising content needs processing time; a pause is respect |
| Pitch movement | Small, with a lift only at the CTA | Constant brightness reads as sales; flat reads as indifference |
| Emotional intensity | 7/10, not 10/10 | Maximum emotion triggers defensiveness and lowers giving |
One pattern that tests well: flat at the open, warming through the middle, level again at the ask. The curve is low–medium–steady, not a single ramp up.
Four ethical lines
These aren’t fine print. Getting one wrong costs a year of donations, not one video.
| Don’t | Why | Do instead |
|---|---|---|
| Clone a beneficiary’s, survivor’s, or volunteer’s voice | A voice is part of identity; synthesizing it is speaking for them | Narrate in third person, or use their own recording with written consent |
| Generate cases or quotes that don’t exist | Fundraising touches public interest; fabricated claims can be illegal | Use real cases, de-identified, with consent |
| Put words in a real person’s mouth via AI | Viewers hear it as that person’s own testimony | Separate narration from quotes, and label both in captions |
| Hide that the narration is synthetic | Advertising and fundraising ethics in many regions require disclosure | Note “narration is AI voice” in the description or end card |
Practical rule: AI narrates, humans testify. Organization statements, figures, and the ask can be synthetic. One line from a real person, or the ambient sound of a real site, stays real. Viewers can hear the difference — and that difference is the credibility.
A four-beat script template
[0-5s] (300 ms pause)
Last winter, Mrs. Chen ate a hot dinner at our center for the first time.
[5-20s] Her neighborhood has 180 people living alone.
A third of them speak to fewer than two people in a week.
(400 ms pause)
[20-35s] We deliver a meal every day, and sit down for ten minutes.
62,000 meals last year. 62,000 visits.
[35-50s] For her, it isn't a lunchbox.
It's someone noticing whether she's home today.
(600 ms pause)
[50-60s] Thirty dollars a month puts a knock on the door four days a week.
Give now, and this winter's meals don't stop.
Four things carry the weight:
- Concrete nouns beat adjectives. “62,000 meals” outperforms “a wide range of services.”
- One fact per sentence. Two numbers in one line is one number lost.
- Give the turn its own line. “It isn’t a lunchbox” needs its own space.
- One amount in the CTA. Multiple tiers create decision paralysis.
Multilingual versions: when they’re worth it
| Situation | Worth it? | Approach |
|---|---|---|
| Diaspora donor base | Yes | Separate English narration, not burned-in subtitles |
| Multilingual service area | Yes | Prioritize the languages you actually serve |
| International partner org | Depends on partner | Start with subtitled English, then decide on dubbing |
| Local small-dollar only | Usually not | Spend it on script quality and distribution instead |
The cost people miss: every language is another version to maintain. When annual figures change, every language regenerates. Keep one slug and identical paragraph splits, and regenerate only the changed sections.
Seven-step production flow
| Step | Action | Time |
|---|---|---|
| 1 | Pick one real case; secure consent and agree on de-identification | 1–3 days |
| 2 | Write the 60-second script; read it aloud until it flows | 2 hours |
| 3 | Mark pauses and number pronunciation | 30 minutes |
| 4 | Generate narration; audition two or three voices | 1 hour |
| 5 | Edit: picture, captions, music (fade out before the ask) | Half a day |
| 6 | Internal review: figures, consent scope, disclosure line | 1 day |
| 7 | Publish, distribute, check completion rate and conversion after a week | Ongoing |
Don’t skip step 6. An error in a fundraising video costs an order of magnitude more than in a marketing video.
Cost comparison for small organizations
Four 60-second refreshes a year (illustrative figures; actual cost depends on plan and volume):
| Approach | One-off cost | Cost per refresh | Notes |
|---|---|---|---|
| Hire a voice actor each time | Medium | Medium (scheduling, studio) | Most natural, least flexible |
| Retained voice actor + AI patch lines | Medium | Low | Good fit for annual small updates |
| AI with a cloned brand voice | Low | Very low | Consistent, needs a human pass on tone |
| Generic AI voice | Very low | Very low | Fastest, least recognizable |
The sweet spot for most small organizations is the third: clone the founder’s or a longtime volunteer’s voice and make it the organizational voice. Viewers hear someone familiar; marginal cost approaches zero. For consistency across channels, see voice cloning for brand consistency; for the multi-language side, the localization QA checklist lists pass criteria per item.
Common mistakes
- Music that never stops. Pull it out under the ask so the words land.
- Captions out of sync. Fundraising clips get screenshotted and forwarded; typos travel.
- Numbers without units or timeframe. “Helped 62,000 people” is weaker than “62,000 meals in a year.”
- Too many asks. Donate, subscribe, share, sign — four asks is zero asks.
- A mispronounced organization name. Listen to the whole cut, proper nouns especially.
FAQ
Can viewers tell it’s AI? On short, level, low-drama narration, most can’t. But one unnatural pause or one mispronounced name and they immediately can. Proofreading matters more than voice selection.
Should we disclose it? Yes, especially where donations and public interest are involved. One line in the description or end card costs nothing and protects you if challenged.
Can we recreate a deceased founder’s voice? Not advisable unless there’s explicit written authorization and family consent, and even then only for memorial use. Using it in a fundraising appeal carries high risk.
Can one voice cover both English and Chinese? Usually not — the phrasing differs. Cross-language consistency comes from rate, pauses, and the emotional curve, not from a single voice ID.
How long should the video be? 45–60 seconds for social, up to 90 for site and email. Beyond two minutes, completion rates for fundraising collapse unless the story is exceptional.
How often should we refresh it? Once a year for figures and annual results, plus short seasonal cuts. Keep paragraph splits stable so only changed sections regenerate.
Try it
To test a 60-second cut, generate narration and compare two or three voices with Soundwaver TTS. If you want a consistent organizational voice, voice cloning keeps every refresh sounding like the same person. For script structure, see voiceover scripts that convert; plans are on the pricing page.
Note: this is production practice, not legal or fundraising-compliance advice. Rules on solicitation disclosure, subject consent, and AI-content labeling vary by jurisdiction — review with local counsel and your organization’s policy before publishing.


