AI Calling Compliance in India: TRAI Rules and Best Practices

If you plan to run AI calling in India at any reasonable volume, you will run into TRAI's telecom commercial communication rules sooner or later. Better sooner. This is a plain-English summary of the rules that matter most for AI voice agents, and the practical do's and don'ts that keep your campaigns alive.

None of what follows is legal advice. It is operational guidance from running real campaigns. For specific legal interpretation, talk to a lawyer.

The frame: TCCCPR 2018

The relevant TRAI regulation is the Telecom Commercial Communications Customer Preference Regulations, 2018 (TCCCPR), with updates added in 2024 and 2025. It governs commercial voice and SMS in India and rests on three core ideas:

1. Customer preference. Subscribers can register on the DND (Do Not Disturb) framework for fully blocked, partially blocked or category-specific blocking.
2. Sender registration. Entities sending commercial communication must register and assign each campaign to a registered header and template.
3. Time and content discipline. Commercial calls are restricted in timing (broadly 09:00 to 21:00 for most categories) and in what can be said.

The 2024 and 2025 amendments tightened these in three ways: clearer rules around AI/auto-dialled calls, clearer ownership of consent records, and steeper penalties.

What this means for AI voice agents specifically

The good news: AI voice agents are not in a separate, more restrictive category. The rules apply the same way they apply to any outbound commercial call. What differs is that AI lets you call far more numbers far faster, so the discipline that humans applied informally now has to be designed into the system.

The practical implications:

• The agent must identify itself as an AI assistant at the start of the call, and identify the business it is calling on behalf of.
• DND must be honoured automatically. Any campaign needs an active DND filter against the latest TRAI-published register, not last month's snapshot.
• Calling windows must be enforced by the dialler, not relied on operators to remember.
• Consent must be recorded with a clear basis for each call (transactional, service, promotional, customer-initiated).
• Opt-out must be honoured immediately. If the caller says "do not call me again", that number must move to your internal block list within minutes, not next week.

Categories matter

TRAI splits commercial communication into categories: service messages, transactional messages, promotional, and so on. The rules differ across them. For example, transactional calls (an OTP, a delivery update) have different windows than promotional calls. AI agents must know which category their campaign belongs to, and the dialler must enforce the corresponding rules.

Most failed audits come from misclassification, not from someone deliberately ignoring a rule. Pick the category honestly and your job is much easier.

The practical compliance checklist

A working AI calling operation needs all of these, not just some:

• A clear basis for each call (consent, transactional, customer-initiated).
• A DND filter applied at dial time, refreshed daily.
• A calling-window filter that respects 09:00 to 21:00 (or the right window for your category).
• The agent introduces itself as AI and as the business it represents.
• The caller can opt out by voice, and that opt-out is honoured immediately for that number.
• All calls are recorded and transcripts retained for the period your category requires.
• Sender header and content template are registered where required.
• A complaints inbox the regulator can reach you on.

The vendor you pick should implement these by default, not as something you bolt on later.

What gets campaigns shut down

Three things, almost always:

• Mass calling outside permitted hours. A 22:30 promotional dial is the fastest path to a complaint and a regulator action.
• Ignoring opt-outs. If a number says "stop" today and gets called again next week, you will hear about it.
• Pretending to be a human. If the AI does not identify itself, regulators and consumer forums take that very poorly.

Each of these is preventable in software. There is no reason to build a campaign that depends on humans remembering rules.

markAIble's stance

markAIble enforces all of the above by default in every campaign. DND filtering, calling-window enforcement, AI self-identification, voice opt-out routing, full recording and configurable retention are not feature flags; they are part of how the system works. The benefit is that your team focuses on the conversation, not on staying out of regulator letters.

---

If you are unsure how your specific use case maps to TRAI rules, book a 15-minute compliance call. We will tell you exactly which category you fall in and what to set up.