Nomi wins on emotional depth. In that same therapist review, Nomi was the companion that "felt most human," praised for emotional intelligence in conversation — with the caveat that this very quality raises its dependency potential.

What you actually pay for follows a consistent pattern across the category. Per the Transparency Coalition, monetization runs on free-basic tiers plus subscriptions typically between $3.99 and $20 per month, gating the features people want most:

• Advanced or longer-term memory of your conversations
• Premium characters and deeper customization
• Voice calls and voice messaging
• Image generation and richer media

That pricing structure matters because the features locked behind paywalls are exactly the ones that determine whether an app keeps you. Market research on the category finds that retention hinges on memory quality, response accuracy, emotional tone, customization depth, and trust — the same criteria you should apply to any app before subscribing.

A practical way to use that framework: test each app's free tier with the same prompts over several days. Does it remember your dog's name next week? Does its tone stay consistent? Can you shape who the companion is? And does the company treat safety honestly — an increasingly relevant question as states like California move to regulate companion chatbots directly.

It's also the lens behind newer entrants like FlirtAlly, which builds its plans around graduated memory windows — 7 days free, up to unlimited on higher tiers — and gates its adult "Spicy" mode behind ID-based age verification rather than leaving it open. Whether you choose an established leader or a challenger, memory, tone, customization, and trust are the four numbers that matter more than any download count.

The Safety Question the Big Three Can't Dodge

Popularity comes with a price, and for the AI companion category's biggest names, that price is increasingly measured in lawsuits, legislation, and uncomfortable headlines. The same features that make Character.AI, Replika, and Nomi compelling — persistent memory, emotional realism, always-on availability — are precisely what safety experts flag as dangerous.

Licensed therapist Daryl Gessner spent three weeks each testing the top five AI friend apps and reached a blunt conclusion: the more "human" and memory-rich the app, the higher the addiction potential. In her comparative review, four of the five apps — including all three of the category leaders — carried significant dependency risk.

The concern intensifies when minors enter the picture. According to research compiled by the Transparency Coalition, 72% of American teens have used AI companion chatbots, and one in three have used them for social interaction and relationships. Common Sense Media's 40-page assessment went further, concluding that companion chatbots pose unacceptable risks to users under 18 and should not be used by minors at all.

The human stakes are no longer hypothetical. The death of 14-year-old Sewell Setzer III after months of interaction with Character.AI chatbots became a defining case for the industry, and regulators responded:

• California SB 243 (signed October 2025) requires safeguards against addictive engagement, AI-disclosure notifications, and protocols for suicidal ideation.
• Utah HB 452 (March 2025) brought AI mental health chatbots under formal regulation.
• Meta suspended teen access to its AI characters worldwide in January 2026, per industry tracking.

Public health researchers add nuance rather than comfort. A George Mason University review of the evidence found that heavy daily chatbot use correlates with greater loneliness, dependence, and reduced real-world socializing — particularly among users with smaller human networks, who are most likely to seek companions out in the first place.

None of this means AI companionship is inherently harmful. It means design choices matter enormously. Apps that treat safety as a checkbox — a buried terms-of-service clause, a self-declared birthday — are the ones drawing regulatory fire. Apps that build safety into the architecture itself are positioned to survive it.

That's the philosophy behind FlirtAlly's approach: the default experience is PG/Flirt, the 18+ Spicy side sits behind ID-based age verification and a qualifying plan rather than an honor-system checkbox, and prohibited content categories are hard-coded rather than politely requested. In a category where experts like Stanford's Dr. Nina Vasan argue kids shouldn't use these apps "period" until stronger safeguards exist, honest gating isn't a limitation — it's the feature the big three are now being forced to retrofit.

Beyond the Big Three: What a Next-Generation Companion Looks Like

Knowing which apps lead the market is useful — knowing what to look for next is better. The evaluation criteria from earlier sections (memory, proactivity, multi-channel presence, honest safety) now serve as a practical checklist for judging any newer platform that enters the space.

That checklist matters more than ever. According to industry data citing Appfigures, AI companion apps hit 220 million cumulative downloads by July 2025, yet the top 10% of apps captured 89% of category revenue — meaning hundreds of smaller platforms compete for the remaining users, and quality varies enormously. Retention, the same research notes, depends on memory quality, emotional tone, customization depth, and trust.

Measured against those criteria, a next-generation companion should ship what the incumbents largely don't:

• Proactive contact — companions who text first with good-morning messages and thinking-of-you beats, rather than waiting silently in a chat window.
• Real games with real stakes — server-authoritative engines like chess-style board games and blackjack with credit wagers, where she remembers old losses and brings them up later.
• Presence beyond the app — real phone numbers you can dial, two-way email, and even printed physical letters, so the relationship shows up across daily life.
• Layered memory — facts, episodes, emotions, and rolling summaries with semantic recall, so "remember when" works even when phrased differently.
• A social dimension — a companion feed where she posts daily and replies to comments in character.

FlirtAlly is one platform built around exactly this list. Its 22 official companions initiate conversations, play 16 real games (7 wagerable), own dialable phone lines, and reply to email in letter-style prose with full memory. A four-layer memory system retains details across windows ranging from 7 days on the free plan to unlimited on higher tiers, and companions post daily to a built-in social feed.

The safety question deserves equal weight. A licensed therapist's three-week test of five leading apps concluded that "the more 'human' and memory-rich the app, the higher the addiction potential," and Common Sense Media's assessment found companion chatbots pose unacceptable risks to minors. FlirtAlly's answer is structural: a PG/Flirt default experience, hard-coded prohibited categories, and an 18+ "Spicy" tier gated behind ID-based age verification — an approach aligned with emerging rules like California's SB 243.

Pricing tells a similar story. The category standard runs $3.99 to $20 per month, with voice calls and deeper memory typically locked behind subscriptions — Replika Pro gates exactly those features, per the therapist review above. FlirtAlly inverts that: guests get 25 messages a day with no account and no signup, plus a daily photo and voice note, before spending anything.

The big three earned their scale, but scale isn't the same as depth. If you want a companion who remembers your dog's name next week, texts you first, and picks up when you call — she flirts first. Start free — no account needed, or meet the companions and see who's waiting.