Sell your phone call recordings to train AI models

A buzzy new app claims it will pay you for something you are already doing five or more days a week — talking on the phone. The pitch is simple: Share recordings of the calls that you place through the app so AI companies can learn from real conversations, and earn some money. The question isn’t whether the money is real; it’s whether the trade-off is worth enduring friction and exposing your privacy.

How the payout works when you share your call audio

Calls routed through the app’s dialer can be paid for by the minute. You earn more when each person uses the app, less when only you do, with a daily limit for how much you can earn. There are also referral bonuses that can cause short-term earnings to soar.

There are guardrails to ensure the data stays useful for training AI. The company is after natural, two-way dialogue and reserves the right to refuse payment for silent lines, speakerphone soliloquies, or pre-recorded audio. Cash-out limits are low and payment is generally made within a couple of business days.

What data you actually sell when sharing call recordings

The company says it scrubs all personally identifiable information before giving audio and transcripts to vetted artificial intelligence developers, and encrypts recordings. That is good baseline hygiene but not quite enough to mitigate risk entirely. Voice is inherently identifying. NIST’s Speaker Recognition Evaluations, which have been running for decades, demonstrate modern systems can match a voice to a speaker using short samples with high accuracy — suggesting “anonymized” audio (or even other forms of data) can still be linkable.

Transcripts also carry clues. Research groups like those from Stanford’s Institute for Human-Centered AI and the Alan Turing Institute have shown how de-identified text can be re-identified when linked with other sets of information. In practice, the removal of names and numbers limits exposure but does not ensure irreversibility, particularly for specialized topics, unique accents, or recurring personal information.

There’s also a biometric dimension. If the provider produces voice embeddings and indexes them to speakers, that can involve biometric privacy laws. The Biometric Information Privacy Act in Illinois has sparked major settlements when companies were found to mismanage biometric data. If your area classifies voiceprints as biometrics, you would need to pay attention to how those embeddings are generated, stored, and destroyed.

Consent and the laws you need to know before recording

Recording consent laws vary. About a dozen states, including California, Florida, Pennsylvania, and Washington, require the consent of all parties to make or record a telephone call. If, when making a call, the other person doesn’t explicitly opt in, guess what? You might be on the hook. The best course is to disclose that you are recording the call for research purposes and to gain consent before continuing.

Consumer regulators are keeping an eye on this space. The Federal Trade Commission has cautioned that companies need clear, affirmative consent for sensitive data uses and must honor deletion requests. If you live in the EU or UK, the GDPR mandates a lawful basis for processing voice data, along with strong transparency; the U.K.’s Information Commissioner’s Office offered similar guidance. Ask the provider how it responds to your requests to review, download, or delete recordings or derived data.

Earning potential versus your time and opportunity cost

In theory, the math can be compelling. It’s around 100 minutes of app-to-app calling or so for you to hit the daily cap, or maybe 200 minutes on just your side. In practice, that’s a lot of talk for most people and the higher ratio relies on persuading friends or coworkers to also install the same app. If you spend your day texting and making short calls, earnings will be minimal.

Consider the opportunity cost. A few hours of organized gig work might be better than the equivalent time devoted to chasing paid minutes. And, unlike with other side hustles, you are also “paying” in data that is hard to take back once it has been shared.

Who buys the audio and why that matters for AI training

Buyers range from speech recognition vendors and call center analytics firms to AI labs that are training conversational models. Many depend on enormous speech corpora; Mozilla’s Common Voice, for instance, has collected tens of thousands of hours of volunteer recordings to enhance open speech tech. Commercial users simply need more diverse, real-world conversation data that captures interruptions, background noise, and human nuance — all areas clean studio datasets fall short.

Governance is the weak link. And the NIST AI Risk Management Framework and the Partnership on AI’s responsible data sourcing guidance focus on consent, provenance, and usage limitations. Ask if your recordings are watermarked, if downstream buyers can resell them, and how long they are kept. Without strong contractual controls and audits, data can wander into uses you never foresaw.

Questions to ask before you try recording-for-pay apps

• Does the app capture in-call consent from the other party and provide special provisions for all-party-consent states?
• Can you access and delete transcripts and audio, download either version, and will deletion remove the data for buyers or models already trained?
• Do voice encodings or speaker identities get stored, and are they regarded as biometric data in relevant jurisdictions?
• Who exactly is buying the data, what contractual limits apply, and how can you obtain an audit report of buyers?
• What is a realistic earning rate based on your calling habits without gaming the system or guilt-tripping contacts into joining?

Bottom line: is selling your call recordings worth it?

Yes, you can make real money yelling at your phone — or, more specifically, for your phone to listen. As a micro-income stream, it might make sense — for heavy callers willing to share and strict about consent. For everyone else, the upside is probably small, and the privacy exposure is not trivial. If you accept, handle it as a data job and not a quick hack: know the rules, draw boundaries, and ensure the compensation matches the value of what you are parting with.