A2P, P2A, P2P: why the direction of a message decides everything about it

A2P means application-to-person: software sending to a human. P2P is two humans. P2A is a human writing to software — an inbound enquiry, a keyword, a reply.

These look like taxonomy for its own sake. They are the reason your messages cost what they cost and arrive when they arrive.

Carriers built their networks for people texting people, and priced accordingly — bundles, flat rates, unlimited plans. Then businesses discovered they could reach anyone for the price of a personal text, and the economics stopped working. Worse, the abuse arrived: at scale, a channel with no gatekeeper becomes a channel nobody reads.

The response was to separate the traffic. A2P is identified, registered, filtered and priced differently. That is the entire history in a paragraph, and it explains almost every rule you will meet.

You must be identified. In the United States that means registering a brand and a campaign; since February 2025 carriers block unregistered 10DLC traffic outright rather than degrading it. In India it means registering headers and templates on the DLT platform. In the UAE it means a registered sender ID and, for promotional traffic, the mandatory AD- prefix.

Your content is inspected. Filtering is real, aggressive, and largely undocumented — deliberately, because a published filter is a filter that can be evaded.

Your price reflects it. A2P rates are not consumer rates, and the gap is where a carrier recovers what bundles gave away.

Inbound is the direction with the fewest restrictions, because the customer chose to start it. Almost every constraint discussed on this site — WhatsApp's window, consent requirements, promotional categorisation — relaxes or disappears when the person contacted you first.

That is not a loophole. It is the system working as designed: permission granted by action is the strongest permission there is.

The practical consequence is a design principle. Programmes that create reasons for customers to write first are structurally cheaper and structurally more deliverable than programmes that do not. Not because of a clever tactic, but because the entire regulatory and commercial apparatus is built around who started the conversation.

The part that surprises teams most is that content matters, and the rules are deliberately unpublished.

Carrier filters look at message bodies. Certain words, certain link shapes, certain patterns of repetition raise a score, and above a threshold the message is dropped — usually silently, sometimes with a status code that means "spam" in a way nobody will confirm.

Two practical implications. Shorteners are a liability. A generic link shortener is shared with every other sender using it, including the bad ones, and reputation attaches to the domain. Use a domain you control. Test with real content. A campaign that passes with lorem ipsum tells you nothing about a campaign containing the word "free" and a discount code.

Not all senders are equal, and the choice affects both cost and how the message is received.

• A long code is an ordinary number. Cheap, familiar, two-way, and rate-limited — it is a phone number, and networks expect phone-number volumes from it.
• A short code is a premium leased number with high throughput and heavy vetting. Expensive and slow to obtain, and worth it only at volume.
• An alphanumeric sender shows your brand name rather than a number. Recognisable, and one-way — the recipient cannot reply, which quietly disqualifies it for anything conversational.

The one-way property of alphanumeric senders is the trap. A brand chooses it for recognition, then discovers that the cheapest and least constrained direction of travel — inbound — has been designed out of the programme.

A2P delivery is rate-limited per sender, and the limit is a property of your registration and reputation rather than of your budget.

This catches out anyone planning a large simultaneous send. Ten thousand messages do not go out at once; they queue, and the tail of that queue can be hours behind the head. If the message is time-sensitive — a flash sale, a service alert — the last recipient is receiving something different from the first.

Design for it: stagger deliberately, put the time in the message rather than implying it, and never assume a broadcast is instantaneous.

• Direction decides price, routing, filtering and regulation. Model it explicitly.
• Content is inspected. Use a domain you control and test with the real copy.
• Alphanumeric senders cannot receive replies — check that against your programme before choosing one.
• Throughput is capped by reputation. A large send is a schedule, not an event.