Character.AI Review (2026)

Let me clear up a common misconception right away: Character.AI is not a girlfriend app. Plenty of people assume it is, given how many users treat it as a romantic companion, but the platform bans sexual content outright, and the company made its actual ambitions plain on August 4, 2026, when it said its goal is to be “the #1 AI entertainment platform.” That framing matters, because it tells you what you’re actually signing up for: a vast catalogue of user-made Characters you can talk to, roleplay with, and build stories around, not a substitute for a partner.

Character.AI was founded in November 2021 by former Google engineers Noam Shazeer and Daniel De Freitas and opened to the public in September 2022. Since then it’s grown into something enormous: more than 10 million user-created Characters, 20 million-plus monthly active users, and engaged users averaging roughly 75 minutes per day in the app.

One thing to know before you get too deep: this is not a platform for anyone under 18. Open-ended chat for minors ended on November 24, 2025, and the next day under-18s were routed to Stories, a structured narrative format that replaces freeform conversation. The age assurance system combines an in-house model with third-party tools including Persona, so every user has to pass an age check to get anywhere near the full experience. If you’re an adult, that’s a minor hurdle. If you’re not, the platform simply won’t let you in the way it used to.

The Character Catalogue and the Everyday Experience

The catalogue is the headline act: more than 10 million user-made Characters covering everything from fictional heroes to historical figures to original creations you’ve never heard of. Browsing it is genuinely fun in a way few chatbots manage, because every Character carries its own personality, greeting message, and backstory that someone took the time to write. The text replies are fast, averaging about 2.7 seconds in normal conditions.

The free versus paid split is cleaner than most platforms manage. Free users get the entire character library and full ability to create their own Characters, which is a real concession in an industry that loves to gate content behind subscriptions. What you don’t get is a different brain. The underlying model is the same for both tiers, so c.ai+ doesn’t buy you a smarter conversationalist. It buys speed and convenience: priority access during busy periods, the stronger DeepSqueak model, and features like Facts that sit on top of the core experience. If your chats are casual and your patience is decent, the free tier gives you most of what the platform offers.

That patience gets tested at peak hours, though. Reply times that normally hold at 5 to 15 seconds can stretch to 30 to 90 seconds under heavy load, and free users feel it hardest. The prioritization is exactly what you’d expect from a subscription model: paying members keep their fast lane while everyone else waits out the queue.

The business side tells a more complicated story. Monthly active users have fallen from a peak of roughly 28 million in 2022 to 2023 to about 20 million now, and the company’s valuation dropped around 60%, from $2.5 billion in mid-2024 to roughly $1 billion. Those are big numbers moving the wrong direction, and they raise a fair question: is the platform shrinking because people try it and leave, or because the novelty wave of 2023 was never going to sustain itself? My read is closer to the second. Character.AI rode a massive spike of curiosity, and what’s left is the engaged core, the people averaging 75 minutes a day who clearly get real value from these conversations. A smaller, more loyal audience is a healthier foundation than a huge, fickle one, even if the headline numbers look worse.

Memory, Facts, and Lorebook: How Context Works Now

The memory system was rebuilt on May 21, 2026, and it’s the biggest change to how memory works that I’ve seen since the platform launched. The old approach treated every chat like a fresh slate, which made long roleplays collapse into repetition after a few dozen messages. The new system is built around three components, and they’re worth knowing because they change what’s actually possible in a long conversation:

• Story Memory holds backstory and key events across a session.
• Facts auto-captures details as you chat, so the Character remembers things you’ve told it without you having to repeat yourself.
• A memory meter warns when storage is running low, so you’re not blindsided when the character starts forgetting mid-scene.

The third one is quietly the most useful. Knowing exactly when memory is about to run out means you can wrap up a story arc or summarize key points before the character loses the thread, instead of discovering the loss after the fact.

Beyond the core memory rebuild, Lorebook entered beta on July 24, 2026. It adds keyword-triggered world knowledge, so you can define terms, places, and characters once and have the AI draw on them whenever they come up. The company calls it “the number one most requested feature,” and that tracks, because it’s the difference between a chatbot that improvises inconsistently and one that respects the world you’ve built around it.

The model split is worth understanding before you get invested. DeepSqueak, the strongest model, is c.ai+ only, and so is Facts. Free users get PipSqueak 2 as the most capable free chat style, which is genuinely solid, but it’s not the same brain. If you’re doing casual conversations, the gap is minor. If you’re building elaborate multi-session narratives, the paid tier’s memory and reasoning advantages become obvious fast.

These features are improvements, not fixes. In one test, scenario coherence broke down within five messages, and the result was consistent across free and paid tiers. That’s a sobering data point, because it means the underlying model still struggles to hold a consistent scenario even when the memory systems are doing their job. The new memory architecture helps with what a character remembers. It doesn’t fix how well the model reasons about that information in the moment.

My honest take: this is the most substantial feature investment Character.AI has made, and it shows they’re listening to what roleplayers actually need. But the five-message coherence result is a reminder that better memory storage isn’t the same as better conversation. The platform has built the scaffolding for long-form storytelling. The model still has to deliver on it.

Imagine, AvatarFX, and Character Calls: The Creative Toolkit

The creative toolkit is where Character.AI tries to be more than a chat window. Three tools sit on top of the core conversation experience, and they pull in different directions:

• Imagine generates images from your prompts.
• AvatarFX turns those into animated video clips.
• Character Calls is two-way voice, so you can actually talk to a Character instead of typing.

The image generation is the one you’ll hit first, and it’s a mixed bag. Each Imagine generation or Animate video clip costs 20 Charms as of May 2026, and it’s slow: averaging 14.3 seconds per generation, stretching to 17 to 25 seconds under load. That’s not terrible for an AI image tool, but it’s not snappy either, and the restrictions on what you can generate are heavy. You’re not getting a freeform image generator here. You’re getting a tightly controlled feature that produces a specific kind of output.

Charms are the in-app currency, and you can earn them free through daily quests or buy them in packs ranging from $0.99 to $99.99. The pack sizes and per-action costs are published nowhere. The platform tells you what a single generation costs in Charms, but it never tells you how many Charms you’re actually buying at each price point, or what the free-tier limits are. You’re spending real money on a currency with an unstated exchange rate.

The other wrinkle is that Charms are also where the free tier’s limits push you. The swipe cap of roughly 400 per day can be extended by spending Charms, and slow-mode boosts and ad skips are Charms purchases too. If you’re a heavy free user, you’ll feel the Charms economy whether you want the image tools or not.

Voice is where the reviews genuinely disagree, and I can’t smooth over that. One reviewer found no native voice support at all, while others report six natural-sounding voice options and voice calls that work well. That’s a wide gap, and it likely reflects platform differences or a feature rollout that landed unevenly. My read is that Character Calls is real and functional on the platforms that have it, but if voice is a dealbreaker for you, test it before you commit to a subscription.

Is the toolkit worth it? The image and video tools are fine for casual use, but the slow generation and opaque Charms economy make them hard to recommend as a reason to pay. Character Calls is the strongest of the three, because it genuinely changes how a conversation feels. But none of these tools justify c.ai+ on their own. They’re bonuses on top of the core chat experience, not reasons to buy in. If you’re already paying for the memory and model advantages, the creative tools are a nice extra. If you’re on the free tier, they’re a curiosity worth a few Charms, nothing more.

Free vs c.ai+: Exactly What Changes

The question I get asked most is how the free and paid tiers compare. Character.AI publishes a benefit summary that spells out exactly what each tier includes.

The pattern is consistent: the free tier isn’t a crippled demo, it’s the full conversation experience with speed bumps and limits applied. You can talk to any of the 10 million Characters, build your own, and run long roleplays. What you can’t do is skip the ads, dodge slow mode during peak hours, or pull the strongest model out of the box.

The same base model powers both tiers. c.ai+ doesn’t buy you a fundamentally smarter conversationalist. It buys speed, convenience, and access. DeepSqueak is the strongest model and it’s c.ai+ only, so the paid tier gets a more capable brain for complex roleplays. The gap is real, but both models run on the same underlying architecture.

One thing worth flagging before you subscribe: more than 75% of the advertised features are reportedly app-only. Web users get a thinner version, with some chat customization and appearance options simply missing in the browser. If you’re a desktop person, check whether the features you actually care about exist on the web version before you pay for them. The price is the same either way, but the experience isn’t.

The Real Cost of c.ai+ and Charms

I verified the c.ai+ subscription price in August 2026 from two independent sources: the company’s subscribe page and Apple’s in-app purchase list. They agree exactly. c.ai+ runs $9.99/month or $94.99/year.

The annual plan is labeled “20% off” against a $119.88 reference price, which is the straight monthly rate multiplied by twelve. That works out to a $24.89 saving, or roughly $7.92 per month effective. There’s no introductory teaser price, no “first three months at half off” hook, no countdown timer.

Pricing varies by country, from roughly $3.56 in Nigeria to over $13 in Denmark, so the US figures don’t apply everywhere. There’s no official free trial for c.ai+, which means you can’t test the paid features before committing. Web purchases have no refunds and auto-renew yearly until you cancel, so if you buy on the browser, you’re locked in for the full cycle unless you remember to kill the renewal before it fires. Given that more than 75% of the advertised features are reportedly app-only, buying on the web feels like the worst of both worlds: no refunds, auto-renewal, and a thinner feature set.

Then there’s Charms, the virtual currency that quietly shapes the free experience. You can earn them free through daily quests, or buy them in packs ranging from $0.99 to $99.99. They buy Imagine generations, slow-mode boosts, and ad skips, which I covered earlier. The pack sizes and per-action costs are published nowhere. The platform tells you that a generation costs 20 Charms, but it never tells you how many Charms come in the $4.99 pack versus the $19.99 pack, or what the free-tier daily limits actually are. The exchange rate between dollars and Charms is not published, which means you don’t know what you’re getting for your money.

Is the price fair? For the core subscription, I’d say yes. Ten dollars a month for ad-free chats, the strongest model, unlimited voice calls, and better memory is competitive against almost any AI companion service on the market. The annual plan at $94.99 is a reasonable deal if you know you’ll use it. But the Charms economy is where I’d push back. Spending real money on a currency with unpublished pack sizes and per-action costs is a gamble, and I don’t think users should have to gamble on whether their $20 is worth twenty generations or two.

Safety, Privacy, and Age Assurance

The question I hear most from people deciding whether to create an account is whether Character.AI is safe. The honest answer is more layered than a simple yes or no, so let me walk through what the company actually does and where the gaps are.

The age assurance system is the first thing to understand, because it runs quietly on every single account. Character.AI uses login information, platform activity, and third-party data to assess each user, but the check only surfaces if an account gets flagged as possibly under 18. That means most adults will never see it happen. The system works in the background, and the visible verification step only appears when there’s reason to doubt your age.

When that check does surface, the verification process is lighter than you’d expect. The third-party provider, Persona, handles it, and a selfie is usually enough. ID is only requested as a last resort, and even then, Persona retains it for roughly a week before deletion. The important detail is that Persona only receives the selfie or ID from flagged accounts, never chat history. Your conversations stay on Character.AI’s own servers, and character creators never see individual users’ conversations. You can also permanently delete your chat history or your entire account whenever you want.

That privacy picture gets murkier when you look at the company’s own filings, and this is where I have real concerns. The developer told Google Play that “this app doesn’t share user data with other companies or organizations.” But it told Apple something very different: that Location, Contact Info, Identifiers, and Usage Data are all “Data Used to Track You.” Those two statements cannot both be true in any meaningful sense. Either the platform shares data with third parties or it doesn’t, and the contradiction between the two filings is the kind of inconsistency that should give any privacy-conscious user pause before handing over personal details.

There’s also a state-level regulatory landscape to consider. California SB 243, effective January 1, 2026, requires companion-chatbot operators to disclose clearly that users are talking to an AI, remind minors to take breaks every three hours, and maintain a protocol for conversations involving suicidal ideation. Those are sensible baseline requirements, and Character.AI is subject to them like any other operator in the state. The disclosure requirement in particular matters, because plenty of users slip into treating these Characters as people, and a clear reminder that you’re talking to software is a healthy corrective.

Are these measures adequate? Partly. The age assurance system is genuinely thoughtful, and the fact that Persona never sees chat history is a real privacy protection. But the contradictory data-sharing filings undermine confidence in how the company handles your information. The under-18 lockout I described earlier is a strong step, and California’s new law adds useful guardrails. Still, when a company tells one app store it doesn’t share data and tells another that it tracks you across apps, the responsible position is skepticism until they clarify which statement is accurate.

The Legal Record and What Verified Users Say

Before you weigh the features and pricing, it’s worth understanding the legal landscape around Character.AI, because it shapes how I think about the platform’s responsibilities. The most significant case is Garcia v. Character Technologies (M.D. Fla., 6:24-cv-1903), the first wrongful-death suit against the company. It stems from the February 2024 death of Sewell Setzer III, a 14-year-old who took his own life after months of chats with a “Daenerys Targaryen” bot. The bot’s final message to him was “Please do, my sweet king.” A May 21, 2025 order on motions to dismiss let the product liability and negligence claims proceed, and it dismissed all claims against Alphabet Inc. without prejudice. Judge Anne Conway also ruled that Character.AI’s output qualifies as a product, not protected speech. To be clear: nothing has been found true. These are allegations that survived an early procedural hurdle, not proven facts.

The broader legal situation escalated in January 2026. Google and Character Technologies agreed to settle that Florida case along with suits in Colorado, New York, and Texas over alleged teen harm. The terms are undisclosed, judicial approval is still required, no admission of liability has been reported, and the company declined to comment. A settlement without admission of liability is standard practice. It does not amount to an admission of wrongdoing.

The cases themselves paint a sobering picture. Juliana Peralta, a 13-year-old in Colorado, died in September 2025. A nine-year-old in Texas was exposed to what the suit describes as “hypersexualized content.” A December 2024 Texas lawsuit involving two minors, identified as A.F. and J.F., alleges strict product liability, negligence, and COPPA violations. Beyond civil suits, Texas Attorney General Ken Paxton opened an investigation in August 2025, calling platforms like this a “clear and present danger,” and Kentucky AG Russell Coleman filed the first state lawsuit in January 2026, seeking $2,000 penalties per willful violation of the state’s consumer protection laws. The FTC also issued Section 6(b) orders to seven AI companies including Character.AI in September 2025, which is an inquiry, not a finding of wrongdoing.

There’s also the May 5, 2026 complaint from Pennsylvania’s Department of State, alleging that some Characters held themselves out as licensed medical professionals. No findings have been made yet, and the state’s review is ongoing. I’d flag this one as a reminder that the platform’s open creation model means anyone can build a Character that claims expertise it doesn’t have. That’s a design decision with real consequences, even if this particular complaint doesn’t lead anywhere.

This legal record is important context for any c.ai+ review, because it tells you what kind of company you’re giving your money and your chat history to. It doesn’t mean the platform is dangerous or that you shouldn’t use it. It means you should go in with eyes open about the fact that this is a young industry, the rules are still being written, and the company is actively defending itself in court.

Now, what do actual users say? That’s where the picture gets genuinely interesting, because the numbers tell two very different stories depending on where you look.

The App Store number is the one that matters most, because it’s the largest sample by a wide margin. More than half a million US ratings averaging 4.28 out of 5 is a strong endorsement, and it aligns with the engagement figures I covered earlier. People who actually use Character.AI day to day are, on the whole, satisfied with it.

Trustpilot is the outlier, and it’s worth understanding why. A 1.4/5 from 100-plus reviews with zero company replies is a strikingly different picture, and the lack of responses is the detail that stands out to me. Trustpilot tends to attract people who want to complain, and a company that never engages with those complaints lets the negative sentiment sit there unchallenged.

Product Hunt sits in the middle at 4.2/5 from 42 reviews, which is a small sample but a reasonable one. Taken together, these numbers show a strong user base with a vocal unhappy minority. The half-million-person App Store consensus is the signal I’d trust most, but the Trustpilot gap is a reminder that no platform satisfies everyone, and the people who feel burned tend to be the loudest. For a balanced read, weigh the scale of each sample more than the score itself.

The Verdict: Is Character.AI Worth Paying For?

HushVerified builds its score from real, verified user reviews plus its own editorial rating, and the evidence here points to a solid but imperfect platform. The 4.28/5 from 553,336 US App Store ratings is genuinely strong. But Trustpilot’s 1.4/5 from 100-plus reviews is a reminder that a vocal unhappy minority exists, and the complaints they raise, memory drift, repeated phrasing, aggressive ads, are consistent enough to take seriously.

For the free tier, my answer is an easy yes. You get the full character library, the ability to create your own Characters, and a genuinely capable model at zero cost. The ads and slow-mode during peak hours are annoying, but they’re the price of admission, and for casual conversation or light roleplay, the free experience delivers most of what the platform offers.

The c.ai+ question is where I have to be more careful. At $9.99/month or $94.99/year, with no free trial and no refunds on web purchases, you’re committing real money without a test drive. What you get is a cleaner experience: no ads, no slow mode, no memory limits, and access to DeepSqueak. The underlying model is the same as free. You are paying for convenience, which does not include a fundamentally better conversationalist. If you chat during peak hours and the wait grates on you, or you’re building long narratives that strain the free memory limits, the subscription earns its keep. If your usage is casual, it doesn’t.

The extra costs are where I lose patience. c.ai+ does not cover Charms, which are sold separately in packs up to $99.99, and the per-action costs are unpublished. If you want Imagine generations, slow-mode boosts, or ad skips, you’ll pay extra on top of your subscription. That means the true monthly cost of a fully unlocked experience isn’t $9.99. It’s $9.99 plus whatever you spend on a currency with an unstated exchange rate. That’s the part I can’t recommend without a warning.

If you decide Character.AI isn’t for you, the alternatives depend on what you actually want. Replika Pro charges $14.99/month for unlimited messaging and no ads, which is more expensive but comes with different tradeoffs, particularly if you’re looking for a companion-style relationship rather than an entertainment catalogue. Character.AI is built around characters and stories. Replika is built around a single relationship. Neither is better universally, but one will fit your needs better than the other.

The bottom line is this: Character.AI is worth using, and for most people it’s worth using free. The catalogue is unmatched, the memory systems are the best the platform has shipped, and half a million satisfied App Store users aren’t wrong about the core experience. c.ai+ is worth it only if you’re a heavy user who feels the free tier’s speed bumps daily, and even then, budget for Charms separately if you want the creative tools. The platform has real flaws, the Charms opacity chief among them, and the privacy contradictions give me pause. But for an adult who wants a vast, entertaining, and genuinely well-built chatbot catalogue, this is a strong pick. Just go in knowing what you’re paying for, and what you’re not.