Recruiters send the same handful of messages hundreds of times a week: the first outreach, the "here's what the interview looks like" note, the offer follow-up. Plain-text templates get ignored, and asking a recruiter to record a personal video for every candidate does not scale. If you build recruiting software, avatar video lets each recruiter record once and then send personal-feeling video messages from your product without sitting in front of a camera every day.
This post covers where avatar video fits in a recruiting workflow, how to model it in your backend, and the consent and disclosure questions you need to answer before you ship.
Where avatar video fits in the hiring funnel
Not every touchpoint benefits from video. The ones that do share a pattern: the message is mostly the same for everyone, but a few details (the candidate's name, the role, the next step) change.
- First outreach. A 30–45 second message from the recruiter that names the candidate, the role and why their background caught attention.
- Interview preparation. "Here's who you'll meet on Thursday and what we'll cover." Useful for reducing no-shows and nerves.
- Job description summaries. A hiring manager explains the role in their own words, generated once per job posting rather than per candidate.
- Status updates and rejections. A respectful, human-sounding "we went another direction" is better than silence. Keep these short and never fake warmth you would not offer in text.
- Offer and pre-boarding. A welcome from the future manager between offer acceptance and day one.
The first two are per-candidate and benefit most from personalization. The job-description video is per-job, so you generate it once and cache it.
Whose face, and who owns the avatar
In recruiting, the avatar is usually the recruiter or hiring manager, an employee of your customer. That raises practical questions your product should answer explicitly:
- Consent. The person in the avatar has to agree to it. PitchTwin's Terms require that every person in an avatar has agreed, and your product should capture that agreement in its own UI too, with a timestamp and the person's account ID. For background, see our post on consent and likeness rights for AI avatars.
- Offboarding. Recruiters change jobs. When an employee leaves, your customer will want their avatar gone. Call
DELETE /avatars/:id, which also deletes the cloned voice, and stop any queued sends that reference it. - Scope. Decide whether an avatar belongs to the person or to the workspace. Tying it to the person, with workspace admins able to revoke, matches how most teams think about it.
Because PitchTwin builds the avatar from real video of the person and keeps that footage, the result looks 1:1 like the recruiter, with their own skin, hair, lighting and expressions. That matters here: candidates who later meet the recruiter on a live call should recognize the same person.
Data model
A minimal schema for a recruiting product looks like this:
| Table | Key fields |
|---|---|
avatars | id, pitchtwin_avatar_id, owner_user_id, consent_recorded_at, status |
video_templates | id, stage (outreach, prep, offer), script_template, avatar_id |
video_messages | id, candidate_id, template_id, generation_id, status, stored_video_key |
Store the PitchTwin generation ID on video_messages so your webhook handler can find the row. Store your own copy of the MP4 in stored_video_key: the video_url PitchTwin returns is a signed link that expires after 24 hours, so copy the file to your storage as soon as the generation succeeds.
Generating a candidate outreach video
Recruiter scripts work best as a script array with short pauses, which gives the delivery a natural rhythm. Our guide to writing scripts for AI avatars covers pacing in more detail.
const res = await fetch("https://pitchtwin.ai/api/v1/generations", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.PITCHTWIN_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
avatar_id: recruiter.pitchtwinAvatarId,
script: [
{ text: `Hi ${candidate.firstName}, I'm ${recruiter.firstName} from ${company.name}.`, pause_after_ms: 400 },
{ text: `I came across your work on ${candidate.highlight} and thought of our ${job.title} role.`, pause_after_ms: 500 },
{ text: "If you're open to a short call, the link is right below this video." },
],
voice: { speed: 1.0, language: "en" },
webhook_url: "https://app.example.com/webhooks/pitchtwin",
}),
});
const { id, status } = await res.json(); // 202, status: "queued"
await db.videoMessages.update(messageId, { generation_id: id, status });
The call returns 202 with a queued generation. When it finishes, PitchTwin sends a generation.succeeded or generation.failed event to your webhook, signed with the PitchTwin-Signature header. Verify the signature, look up the message by generation ID, copy the MP4 to your storage, and only then mark it ready to send. Webhooks are delivered once, so run a periodic job that polls GET /generations/:id for anything still pending. The webhooks section of the docs has the signature format.
Validate inputs before you generate
Candidate data comes from resumes, LinkedIn imports and ATS syncs, so it is messy. Before you call the API:
- Strip or reject fields that are empty, all caps, or contain URLs.
- Keep the highlight field short and factual. Let the recruiter edit it rather than generating it from the resume without review.
- Show the recruiter a text preview of the final script. If your product offers a "send to 200 candidates" button, require a preview of at least a sample.
The response includes words with word-level timings. Use them to burn in captions or ship a WebVTT track, since many candidates watch on mute and captions help accessibility.
Disclosure and hiring rules
This is not legal advice, and rules vary by jurisdiction. A few points worth knowing as you design the feature:
- Tell candidates the video is AI-generated. Beyond being the honest default, it may be required. The EU AI Act requires deployers who use AI to generate or manipulate video resembling real people to disclose that the content is artificially generated (Article 50 of Regulation (EU) 2024/1689); those transparency obligations apply from 2 August 2026. A short line in the email ("This video was created with AI from a recording of Sam") and a small on-screen label are easy to add as product defaults.
- Know which laws are about evaluating candidates. Some hiring laws target AI that analyzes candidates rather than AI that generates messages. Illinois's Artificial Intelligence Video Interview Act (Public Act 101-0260), for example, covers employers who use AI to analyze applicants' recorded video interviews and requires notice, an explanation and consent. A recruiter's avatar outreach is not that, but if your product also scores candidate videos, those obligations apply to that part.
- Keep it accurate. An avatar message should never promise anything the recruiter would not put in writing: salary, timelines, or interview outcomes.
What it costs
PitchTwin charges prepaid tokens that never expire: 1 token per second of video when you bring your own audio, 1.5 tokens per second when PitchTwin voices the script in the recruiter's cloned voice. A 40-second outreach video in text mode costs 60 tokens. On the Growth pack ($39 for 3,000 tokens), that is 50 videos, or about $0.78 each. Failed generations are refunded automatically. Full numbers are on the pricing page.
That makes it practical to include a monthly video allowance in your recruiting plans and sell extra volume on top. There are no seats, so adding recruiters does not change your cost per video.
Try it
Sign up at /signup to get 60 free tokens with no card, upload a recruiter clip in the dashboard, and generate a test outreach video in the Playground. When you are ready to wire it into your ATS, the quickstart walks through the first API call.