A Campaign Manager’s Disclosure Gap. Thousands of Calls Could Look Like Fraud.

Autonomous calls can be mistaken for fraud when the agent hides who is calling, why the call is automated, or how the recipient can stop future contact. Identity disclosure, consent and opt-out handling need to work inside the conversation before a campaign scales.

In 1938, Orson Welles and the Mercury Theatre broadcast a radio adaptation of The War of the Worlds from CBS studios in New York. The programme identified itself as fiction, but its news-bulletin format sounded convincing to some listeners, especially those who tuned in after the opening announcement. The scale of the public panic was later exaggerated, yet contemporary confusion was real enough to become part of the broadcast’s history.

Hadley Cantril examined the response in his 1940 study, The Invasion from Mars. One lesson survives the debate over how many people were fooled: a disclosure cannot protect someone who does not hear it, understand it or believe it applies to what follows.

A voice campaign faces the same design problem. A disclosure placed in a dashboard, campaign brief or opening line does little for a person who answers midway through a transfer, misses the first sentence or switches between Twi and English. Trust must remain legible throughout the call.

Identity has to be clear before persuasion begins

The first spoken turn should identify the organisation and explain that the caller is an automated voice agent. It should also state the purpose of the call in language the recipient can understand.

That sequence matters. Starting with a persuasive script before disclosing automation creates the impression that the system is trying to pass as human. A familiar Ghanaian name, natural Twi pronunciation or local conversational rhythm may increase that risk because the voice can feel recognisable before its role is clear.

A useful opening would follow this order:

• Name the organisation responsible for the call.
• Say that the person is speaking with an automated voice agent.
• Give the specific reason for calling.
• Ask whether the recipient agrees to continue.
• Explain how to decline or stop future calls.

The wording should survive realistic conditions: a noisy roadside in Accra, a brief answer from a shop floor in Kumasi, or a caller who responds in English and then changes to Twi. If the agent cannot establish identity and consent, it should not continue into the campaign script.

Consent must become an observable call state

A friendly “okay” does not always prove informed consent. The response may acknowledge that audio is working, confirm the recipient’s name or simply invite the caller to repeat the first sentence.

The agent needs a distinct consent step tied to what was disclosed. The system should record the prompt presented, the language used, the recipient’s response and the resulting decision. Ambiguous replies should trigger clarification rather than automatic progression.

This is also why an account match or completed call status cannot prove understanding. Bilingual voice-agent testing should test whether the person understood who called, why automation was involved and what choices were available.

Asenda Talk includes consent, opt-out and audit trails for calls, alongside a telephony lifecycle webhook pipeline designed to preserve call truth. Those controls provide the structure for review. They do not remove the need to evaluate the actual words, language transitions and state changes produced during a call.

“Stop calling” must change what happens next

An opt-out is an instruction, not a sentiment score.

The agent should recognise direct requests such as “stop calling,” their Twi equivalents and common conversational variations. Once detected, the call should move into a defined opt-out state, confirm the request without adding another sales attempt, end appropriately and prevent another campaign from calling the same recipient.

The audit trail should make that sequence reviewable:

• What did the recipient say?
• Which language was used?
• How did the agent classify the request?
• Was the number added to the suppression record?
• Did any later call attempt occur?

A green completion status answers none of those questions by itself. A stronger review checks the recording or transcript against lifecycle events and the suppression outcome. A call review after “stop calling” should establish that the instruction changed system behaviour, not merely that the webhook arrived.

Scale only after the controls survive failure

The 1938 broadcast shows why a single disclosure at the beginning can fail even when it technically exists. People enter conversations with incomplete context. They miss words. They interpret familiar formats using prior experience.

Before thousands of autonomous calls, run a small, controlled evaluation across Twi, English and code-switched conversations. Test interrupted openings, ambiguous consent, immediate refusals, mid-call opt-outs and repeat-call prevention. Review each result against the audit trail rather than relying on aggregate completion counts.

Asenda Talk is in active early access. Native Twi speech recognition and synthesis are built and fine-tuned in-house, while Vapi orchestrates the assistant runtime. Outbound calling remains behind an operator-controlled real-money gate, and the live telephony-provider decision has not yet been made. That gate should remain closed until identity disclosure, consent capture and opt-out enforcement produce evidence a reviewer can follow from the first spoken sentence to the final call state.

Orson Welles’s opening disclosure did not reach every listener in the way its creators intended. A campaign manager should assume the same possibility, then design the call so trust does not depend on one sentence being heard once.