Usually not as the face of your brand. An AI avatar works for narrow jobs such as FAQ explainers, aftercare or onboarding videos, and translated versions of a real person's message, if clearly labelled. If customers buy from you because they trust a person, as they do in health and advice work, keep a real person on camera.
The choice is really three separate options: a stock avatar (a paid actor's likeness, shared with every other customer of the platform), a digital twin of you or a staff member (built from your own footage with recorded consent), or you on your phone. Each one fails in a different way, and the digital twin raises a question most owners forget to ask until it is awkward: what happens to that face when the person leaves?
Stock avatar, digital twin or you on camera
| Stock avatar | Digital twin of a real person | You, filmed on a phone | |
|---|---|---|---|
| Setup | Minutes | An hour or two of recording plus processing | None |
| Cost per month | Roughly $0-$90 on common plans | Same plans; premium studio versions cost far more | Nothing extra |
| Updating a video | Edit the script, re-render | Edit the script, re-render | Re-film |
| Trust with customers | Lowest: it's nobody they'll meet | Medium: a real person, but not really them | Highest |
| Main risk | Same face appears in other firms' videos | Consent, leavers and impersonation | Time and nerves |
| Best for | Internal training, generic how-tos | Repeated explainers, multilingual versions | Anything that builds a relationship |
There is also a quieter fourth option that many owners overlook: a cloned voice with no face. You record a few minutes of your own voice, the platform learns it, and you narrate new scripts over real footage of your premises or products without re-recording each time. Viewers see your real place and hear your real voice, and the only synthetic part is the delivery. For an owner who dislikes being on camera but whose voice customers know from the phone, it keeps most of the trust at a fraction of the effort, provided you label it and check every script as carefully as you would a video.
The comparison shows why "spokesperson" is the wrong frame. A spokesperson is the person customers associate with you. An avatar is better thought of as a presenter for content that changes often or needs translating, where re-filming would be the real cost.
What the main avatar tools cost and ask of you
The two platforms most small businesses compare, with figures from their pricing pages in September 2026:
- HeyGen. A free plan makes three videos a month of up to a minute each, watermarked. Creator is $29 a month ($24 billed annually) with 600 credits a month, watermark removal and a voice clone. Pro is $49 a month. Business is $149 a month plus $20 per seat.
- Synthesia. The free plan gives 10 video minutes a month with a small set of stock avatars. Starter is $29 monthly or $18 a month billed annually, with 10 minutes a month and up to three personal avatars. Creator is $89 monthly or $64 a month annually, with 30 minutes a month. A studio-quality custom avatar costs $1,000 a year on annual plans.
Both platforms require a recorded consent statement before they build an avatar of a real person, and the consent has to come from the person whose face it is. HeyGen asks that person to record a short consent video, which they can do from their own phone, and Synthesia's personal avatars need a live consent recording that matches the footage. You cannot make an avatar of an employee without their active participation, which is a good thing and also the first sign of the consent questions below.
A quick sum on minutes: a typical FAQ or aftercare video runs 60 to 90 seconds, so 10 minutes a month covers six or seven finished videos, fewer if you re-render after edits. For a small library that you update a few times a year, the entry plans are enough.
Four questions that settle it for your business
Score each question 0, 1 or 2 and add them up.
- Is a person's credibility part of what you sell? Clinical, financial or advice work scores 0. A shop selling products where the person doesn't matter much scores 2.
- Will the content change often? Price lists, opening times, booking steps or policies that change several times a year score 2. A one-off welcome video scores 0.
- Do you need it in more than one language? Yes, regularly: 2. Occasionally: 1. Never: 0.
- Is the video instructional rather than persuasive? "How to" content that people rewatch scores 2. Anything asking for trust or a purchase scores 0.
Six to eight points: an avatar is worth trialling for that content. Three to five: use one only for specific videos, with a real person elsewhere. Zero to two: film yourself. Most health businesses land low on question 1 and high on questions 2 and 4, which is exactly the pattern in the example below: an avatar for some explainers, never for the relationship.
A hearing-aid shop weighs up an avatar
An illustration. A hearing-aid shop with two audiologists wants nine short videos: a welcome video explaining the first appointment, and eight aftercare videos covering charging, cleaning domes, changing wax guards, pairing with a phone and so on. Many customers are over 70, and some prefer to watch in a second language.
The owner's first idea was a digital twin of the senior audiologist presenting all nine. Running the scorecard changed the plan:
- The aftercare videos are instructional and rewatched (question 4 scores 2), but most of them need real hands changing a real wax guard on a real device. An avatar can talk about the steps; it can't show them. So those eight became close-up phone footage of the audiologist's hands, with her voiceover recorded separately and edited with AI captions.
- The welcome video is about trust (question 1 scores 0), but the shop wanted it in three languages (question 3 scores 2). The compromise: the audiologist filmed the main-language version herself, and a labelled digital twin delivers the two translated versions, checked by fluent speakers among the staff and their families.
The numbers, roughly: filming the eight aftercare clips took one afternoon. The twin needed about an hour of recording and a consent video, then HeyGen Creator at $24 a month on annual billing. Each translated welcome took 20 minutes to generate and check. When the shop changes its appointment process, it edits two scripts and re-renders in under an hour instead of arranging a re-shoot. That is a sensible, narrow use of an avatar, and the audiologist remains the face customers meet.
Sorting a real content list: a veterinary practice's twelve ideas
The scorecard is easier to trust once you've run it over a real list. An illustrative mixed practice brainstormed twelve video ideas for the year. Sorted, they looked like this:
- Avatar suitable (4): how to book online; what to bring to a first vaccination; how our pet health plan payments work; out-of-hours arrangements. All four change when prices or processes change, none needs hands-on demonstration, and nobody chooses a vet because of who explains the booking page.
- Real person only (5): meet the new vet; what happens on the day of your pet's operation; saying goodbye (euthanasia and aftercare options); a nurse clinic tour; the practice's annual open day. These are about reassurance or grief, where a synthetic presenter would feel cold or worse.
- Real footage with a cloned or recorded voice (3): how to give a cat a tablet; checking your dog's teeth; trimming claws safely. They need real animals and real hands.
Only a third of the list suited an avatar, and none of those four would ever be described as the practice's "spokesperson". That is the typical outcome: the avatar becomes the presenter for admin and process content, while the practice's people front everything customers feel strongly about.
The leaver problem: consent wording for a staff member's twin
The platform's consent recording covers the platform. It doesn't cover your relationship with the employee. Before you build a twin of anyone other than yourself, put a short written agreement in place. A starting draft to adapt with your employment adviser:
AI PRESENTER AGREEMENT (draft for review)
1. [Name] agrees to [Business] creating an AI avatar and voice
based on recordings made on [date], using [platform].
2. Use is limited to: aftercare and information videos on our
website and social channels. Not for advertising, pricing
announcements or anything [Name] has not approved.
3. [Name] sees and approves every script before it is published.
4. [Name] may withdraw consent at any time by telling [manager]
in writing. We will stop making new videos within 7 days.
5. If [Name] leaves, we stop making new videos immediately,
remove existing ones within 60 days, and delete the avatar
and voice from the platform.
6. Any payment for this use: [amount / none, by agreement].Point 2 matters more than it looks. If the twin can be used for "anything", a future manager could have it announce a price rise or reply to a complaint in the employee's face and voice. Scope it narrowly now. Point 5 answers the leaver question before it becomes a dispute.
Disclosure wording that doesn't sink the video
Owners often worry that a label will make people stop watching. A clear, early label is less damaging than a viewer working it out halfway through. Compare:
- Before: small grey text in the last frame reading "Created with AI technology."
- After: an on-screen line in the first two seconds reading "AI presenter made from Maya's recordings, with her permission. Maya checks every script," plus the same sentence in the caption.
The second version tells viewers three useful things: it's synthetic, it's based on a real staff member, and a real person stands behind the content. It turns the label into a reason to trust the video. Apply the platform's own AI label as well when you upload; the reasoning behind being open is covered in whether to tell customers you use AI.
Testing a script on an avatar before you commit
Avatars handle some scripts much better than others. Short sentences, one idea each, with no rhetorical questions, come out best. A prompt to prepare a script:
Rewrite this script for an AI video presenter. Rules: sentences under
15 words, one instruction per sentence, no questions to the viewer,
no jokes, no exclamation marks. Keep every medical or product term
exactly as written. Mark any word an avatar might mispronounce with
a phonetic spelling in brackets. Target length: 70 seconds.
[paste script]An illustrative output for a pharmacy's "taking your antibiotics" video came back with sensible short sentences and flagged the medicine name with a phonetic spelling. What the pharmacist had to fix: the rewrite had softened "finish the whole course" into "try to finish the course", which changes the advice. That is the general pattern. The avatar will read whatever it's given, confidently, so a professional must approve the final wording word for word, and the phonetic spellings need testing by ear because each platform's voices handle them differently.
Where avatar spokespeople go wrong
The shared face. Stock avatars are licensed to every customer of the platform. The presenter introducing your osteopathy clinic may also be introducing a crypto course and a competitor's clinic, and press reports in 2023 found stock avatars fronting propaganda videos. A customer who recognises the face won't think well of either business.
The in-person mismatch. An illustrative dental practice uses a polished avatar receptionist in its new-patient video. Patients arrive expecting that person at the desk. The gap between the video and the real experience is small, but it's the first impression and it's slightly false.
The comment section. On social platforms, the first comment on an unlabelled avatar video is often "Is this AI?" Once that comment is there, the conversation is about the avatar, not your message. Labelling up front takes the question away.
Impersonation. Once a convincing twin of you exists, your account on the avatar platform is worth protecting like your bank login: two-factor authentication, no shared passwords, and a note on your website that you never ask for payment or personal details by video. Criminals already fake owners' faces and voices without any help; the patterns to watch for are in how to spot deepfake voice and video scams aimed at your business.
Easy to update means easy to forget: keep a register
The main advantage of an avatar, that you can change a script and re-render in minutes, creates a maintenance job. Filmed videos date visibly; avatar videos look just as fresh with an old price in them. An illustrative podiatrist's avatar video kept quoting last year's nail-surgery fee for four months after a price rise, because nobody remembered it existed. A simple register prevents that:
| Video | Where it's used | Facts it states | Owner | Review by |
|---|---|---|---|---|
| Booking a first appointment | Website, confirmation email | Booking link, cancellation window | Practice manager | Every price or policy change, or 1 March |
| Payment plan explainer | Website pricing page | Monthly fees, what's included | Owner | Every price change |
| Translated welcome (two languages) | Website | Opening hours, parking | Receptionist plus fluent checker | Every 6 months |
Tie the review column to events, not just dates. When you change a price, the register tells you which videos to re-render that afternoon.
A four-week trial that doesn't bet the brand
- Week 1: pick one instructional video that scored six or more. Make two versions: avatar and real person (or real person's voice over footage). Show both to three customers you know well and ask which they would rather rewatch.
- Week 2: publish the avatar version somewhere low-stakes, such as an aftercare page linked from a follow-up email, clearly labelled.
- Weeks 3-4: track watch time, the replies or calls asking questions the video should have answered, and any comments about the presenter.
- Decide: if the avatar version is watched as fully and doesn't generate more follow-up questions, extend it to similar videos. If people ask "who is that?" or call anyway, go back to the real person.
Most small businesses finish this trial with a mixed answer: an avatar for a handful of explainers and translations, a real person for everything else. If you want the broader picture of what generated video handles well, the job-by-job verdict in whether AI video is good enough for small business marketing covers the rest of your video plan, and making marketing videos without being on camera sets out the faceless formats that need no presenter at all.
Avatar spokesperson questions
Do I have to tell viewers the presenter is an AI avatar?
In practice, yes. YouTube, Meta and TikTok all ask you to label realistic synthetic people, and if you sell to customers in the EU the AI Act's transparency duties for deepfake content have applied since 2 August 2026. Beyond the rules, viewers who discover an unlabelled avatar tend to assume you were hiding it, which costs more trust than a short label ever would.
Can the avatar speak other languages in my own voice?
Yes. HeyGen and Synthesia both offer voice cloning and translation on paid plans, so a digital twin can deliver the same script in several languages. Have a fluent speaker check every translated version before it goes live, because translation errors in instructions or prices are easy to miss when you don't speak the language yourself.
Can I keep using a former employee's avatar after they leave?
Only if your written agreement with them clearly allows it, and even then it is risky. Viewers and the former employee may see new scripts as putting words in their mouth. The safer default is to stop producing new videos with the twin when the person leaves, retire existing ones within an agreed period, and delete the avatar. Take employment advice on the wording.
Would an avatar work as the welcome video on my homepage?
It can, but the homepage is where visitors decide whether to trust you, so it is the riskiest place for a synthetic presenter. A short phone-filmed welcome from the owner usually does more. If you do use an avatar there, label it, keep it under a minute, and test it against a real-person version.
Further reads
- How to Keep AI-Made Videos On-Brand: Captions, Voice and Intros — Keep avatar videos consistent with the rest of your content.
- How Much Does an AI Marketing Video Cost Compared With a Shoot? — Compare avatar costs with filming properly for a day.
- How to Write an AI Disclosure Statement for Your Website — Write the site-wide statement your avatar labels can point to.
- How to Write an AI Content Policy for Your Marketing — Set rules on when AI presenters may and may not be used.
- Can You Legally Use AI-Generated Images in Your Marketing? — The legal questions around synthetic likenesses and images.
- How to Build a Brand Voice Guide That AI Can Follow — Make avatar scripts sound like your business, not a template.
- How to Make TikToks and Reels for a Small Business With AI — A five-stage short-video routine: AI for ideas, scripts, captions and editing, your phone for honest footage, plus a four-week batch plan and what to measure.
- Should You Tell Customers When You Use AI in Your Marketing? — Which marketing assets need an AI label, what Meta, TikTok, YouTube and Google already require, and wording that sounds honest rather than apologetic.
- Can You Use AI to Write Testimonials? Where the Legal Line Is — Where the legal line sits for AI and testimonials: an allowed-grey-never table, what the rules prohibit, an approved-edit workflow and a consent email.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: HeyGen pricing page and help pages (consent video for digital twins), Synthesia pricing page and ethics page (consent for personal avatars), YouTube Help on altered or synthetic content, EU AI Act transparency provisions.