Why AI companions are moving from apps to WhatsApp

AI companions are moving to WhatsApp because users are already there and apps are hard to keep: WhatsApp has more than 3 billion monthly users, while a median app keeps about 4% of installs at day 30. The catch is policy. Since 15 January 2026, Meta’s rules keep general-purpose AI assistants off the official Business API in most markets, so a companion needs its own number.

How many people are already on WhatsApp?

WhatsApp has more than 3 billion monthly users. Mark Zuckerberg gave the figure on Meta’s Q1 2025 earnings call, reported by TechCrunch on 1 May 2025, and Meta said on the same call that WhatsApp sees its strongest Meta AI usage. The report adds that WhatsApp passed 2 billion in 2020.

The habit matters more than the head count. DataReportal’s October 2025 report, using Similarweb data, says the typical WhatsApp Android user opens the app more than 20 times a day. For a companion, that is the point: a contact in that list is one tap from a conversation, in an app people already open without thinking about it.

What does an app install cost a companion?

A standalone companion app asks for a store visit, a download, an account and later a notification permission, and most people drop out along the way. UXCam’s April 2026 roundup of retention benchmarks, which says it compiles AppsFlyer and Adjust data, puts the median app at 25% retention on day 1, 8% on day 7 and 4% on day 30. Strong apps (the 75th percentile) reach 5% to 8% at day 30. It is a secondary roundup, so read the numbers as a range, not a law.

AI apps have it harder. RevenueCat’s State of Subscription Apps 2026, covering 2025 data from more than 115,000 apps, finds that AI-powered apps earn 41% more revenue per payer but churn 30% faster. Its global median download-to-paid conversion at day 35 is 2.0%.

A WhatsApp contact skips the install, but we found no public retention benchmark for a chat-contact assistant. Treat the comparison as a bet with a clear logic, not a measured result. You can measure it on your own users from the first week.

Which assistants grew on messaging first?

Luzia is the clearest example. The Spain-based assistant raised a $10 million Series A led by Khosla Ventures, and TechCrunch reported in October 2023 that more than 17 million people had talked to it, 8 million were active monthly and it handled 13 million requests a day. Users saved Luzia as a contact on WhatsApp or Telegram, and the company said most people forward the contact so friends can chat with it too. It served Brazil, Argentina, Colombia, Spain, Mexico and, newly then, the United States.

Eight million monthly users out of 17 million who had tried it is not a retention rate, because the 17 million is cumulative. It does mean that, in October 2023, roughly half of everyone who had ever talked to Luzia was active that month, on a product whose growth loop was a contact card people forwarded.

Luzia’s own site now claims more than 85 million users worldwide and lists web chat plus iOS and Android apps. Its agent mode, LuziClaw, is described as a beta on WhatsApp in select countries. That is a useful counterpoint: a team that grew on messaging now also runs apps.

What did Meta’s 2026 Business API rule change?

Meta’s 2026 rule made the official WhatsApp Business API a poor home for companions in most markets. Meta announced in October 2025 that its WhatsApp Business terms would bar “AI providers” when AI is the primary functionality, starting 15 January 2026. TechCrunch listed Luzia among the providers likely to be affected. Businesses using AI to serve their own customers, such as a support bot, are not covered.

Regulators in Italy, Brazil and the EU then pushed back. Meta’s AI provider pricing page lists the markets where it now supports AI providers because it is legally required to: Italy, Brazil, 29 other European countries and, from 15 October 2026, Türkiye. Even there, only service messages are allowed, sent in response to users inside an open 24-hour window. The page does not list the United States, the United Kingdom, India or Mexico.

For a companion builder the result is simple. Outside those markets, the official API is closed. Inside them it is an inbound-reply channel, and in Brazil one with per-message fees. A personal number is the other route. The full story of the January 2026 rule has the dates and what is still open.

What do you give up by moving to WhatsApp?

Moving a companion to WhatsApp trades control for reach. The table compares the two on what a builder feels first.

Two items deserve a second look. First, platform risk is real: Meta’s terms changed in January 2026, and the exceptions depend on regulators. Numbers also get banned. whasp replaces a banned number at no cost, but chats do not move, so you need a second channel such as email to tell users the new link. Second, a chat has no home screen of yours, so account, billing and onboarding stay in your own product.

Moving does not remove your safety and disclosure duties either: they follow the product, not the channel. The safety rules for 2026 list them.

What should a companion builder do with this?

Treat WhatsApp as the front door and the daily conversation, not as your only home. A plan that matches the evidence:

1. Keep the account, the subscription and the memory in your own system, and collect an email at signup.
2. Send users to the chat with a link whose prefilled message carries a code, so the first message identifies them. Launching a companion on WhatsApp walks through it.
3. Start with one number and count returning chats yourself from message.received events, at day 7 and day 30, so you have your own benchmark instead of borrowed ones.
4. Pick use cases where the user writes first and comes back on their own; the 12 that work best are a good start.
5. Compare the other channels before you commit. WhatsApp, Telegram and iMessage differ a lot in reach and rules.

whasp is one way to get the number: $49 per number every 30 days, about 100 active users per number. The map of every option covers the rest.