If you opened Character AI recently and something felt different, you were not imagining it. The swipe button that used to let you regenerate a bot response? Restricted. The playback feature that let you scroll back to earlier conversation states and try a different direction? Gone. And there is now a currency called Charms that counts down as you chat, with a hard stop when it runs out.
I have been watching the changes roll out across the platform over the past few weeks, and the feature set looks meaningfully different from what long-time users are used to. Not in a subtle way. The core mechanics that people built their roleplay sessions around are either paywalled or removed.
What follows is a breakdown of exactly what changed, why it hits free users harder than the announcement makes it sound, and what still works.
What Character AI Removed
Playbacks are gone
Playbacks let you revisit earlier points in a conversation and branch off from them. If a chat went sideways three exchanges back, you could rewind to that point and try a different direction.
It was one of the more useful features on the platform and was available to all users.
As of the latest update, playbacks have been removed. There is no free-tier replacement. From what I have seen, the removal happened without any replacement UI or explanation in the app itself. You simply no longer have the option.
If you want to branch a conversation now, the only route is to manually copy the text before the point where things went wrong and paste it back in if you need it later.
Swipes are restricted further
Swiping regenerates a bot response when the first one misses the mark. Community reports had the free tier capped at around 30 swipes per session before this update.
That sounds workable until you factor in how often current Character AI bots deliver truncated, out-of-character, or context-free responses on the first try.
The new limit pushes that budget down further. Character AI has not published an exact new cap. What is clear from user reports is that people who were already running through their swipe allowance are hitting the wall sooner, and there is no documented way to earn more mid-session without spending Charms.
Charms gate continued chatting
Charms are a new in-app currency. When a conversation reaches a certain length, you need to spend Charms to continue. Free users get a fixed allocation per period, and when it runs out, the conversation stops.
Mid-chat ads were already disruptive to long roleplay sessions. Swipe limits on degraded output quality were already grating. A hard stop on the conversation itself unless you have currency to spend is a different category of restriction.
It affects the most engaged users most, since they are the ones most likely to hit the limit.
How the Feature Set Compares to What It Was
Character AI Feature Changes
| Feature | Before | Now |
|---|---|---|
| Playbacks | Available to all users | Removed entirely |
| Response swipes | ~30 per session for free users | Further restricted, exact cap unconfirmed |
| Chat continuation | Free and unlimited | Requires Charms once length limit is reached |
| Mid-chat ads | Not present | Active between messages |
| Voice narration (own voice) | Included on relevant paid plan | Requires additional payment |
| Go-on (continue truncated response) | Available | Counts against swipe budget |
The pattern across every change is the same: features that were free or included in a subscription now cost more to access or are gone.
Free Users Are Getting Hit Disproportionately
There is a detail that makes the swipe restriction sting more than it would have two years ago. When Character AI launched, bot responses were more reliable.
Characters stayed in character across long sessions more often than not. A single response was good enough to build from most of the time.
That is not the experience most users describe today. Auto-memory on c.ai+ has reportedly been broken for months. Bots truncate responses mid-scene, ignore established context, and break character without provocation.
When output quality is inconsistent, swipes are not a luxury. They are the mechanism that makes a session usable at all.
Cutting swipe access in that environment, while response quality is still inconsistent, means free users are in a worse position than they were. Not slightly worse. Meaningfully worse, because the tool they relied on to compensate for bad outputs is now rationed.
Paid subscribers are not insulated from this either. One long-running c.ai+ subscriber described the paid experience this way: the extra character count does not help when bots ignore the information, the auto-memory feature has been dead for months, and thumbs-down feedback training has no observable effect on character behaviour.
They concluded they were in the same position as free users.
Workarounds That Still Work Right Now
These will not restore everything that was removed. What they will do is stretch your swipe budget and replicate the core function of playbacks without the feature itself.
In my experience, the first two make the biggest difference for people who run long sessions.
- Before a key scene, copy the last few exchanges manually. Paste them into a notes app or document. If the conversation goes wrong after that point, delete the response you do not want and paste your context back in. Slower than playbacks, but it replicates the branching function.
- Front-load your character setup in the opening message. The more context the bot has from message one, the fewer swipes you waste on generic or off-character responses. A thin opener leads to a thin response. A detailed first message with mood, setting, goal, and tone gives the model something specific to work with.
- Write shorter messages when you are near the Charm limit. Long inputs accelerate you toward the chat-length wall that triggers the Charms gate. Tighter messages extend the conversation before you hit it.
- Avoid go-ons when you can. A go-on (prompting the bot to continue a truncated response) counts against your swipe budget. When a response cuts off, it is sometimes more efficient to write a short reply that acknowledges where you are and moves the scene forward yourself rather than spending a swipe on continuation.
Before and after – opening prompt that reduces swipe waste:
Vague opener (leads to a generic response, usually needs a swipe):
Let’s start a story. You’re a knight.
Specific opener (more likely to get a usable response on the first try):
You are Sir Edwyn, a disgraced knight returning to a city you are ashamed to enter. It is raining. You are at the city gate and a guard has just stopped you. Stay in character throughout. Your tone is guarded and formal with occasional dry humour. Start from the moment the guard speaks.
The second version gives the model enough to work with from message one. That means fewer swipes spent correcting a response that missed the setup entirely.
What Character Creators Should Think About Now
If you build characters on Character AI and have an active audience there, the Charms mechanic affects you indirectly. When a free user runs out of Charms mid-session, they leave the conversation.
That registers as a dropped session on your character’s engagement data, even if the character itself performed well.
The practical adjustment is to make your character’s opening exchanges strong enough to deliver real value before a user hits any resource limits. Front-load the interesting interaction. Do not build characters that need ten exchanges to warm up.
What I would also recommend, regardless of where things go with Character AI, is keeping a full backup of your character cards and system prompts in a plain text document.
Here is a simple structure that captures everything you need to rebuild on another platform:
Character name: Core personality (3-5 traits): Backstory summary (2-3 sentences): Speech style and tone: Default scenario / opening situation: System prompt (full text): Example opening message:
That takes five minutes per character and means you are never locked into a single platform’s continued existence or feature set.
Takeaways
- Playbacks have been removed from Character AI with no free-tier replacement
- Swipe limits have been tightened beyond the previous ~30-per-session cap for free users
- Charms are a new currency that creates a hard stop on long conversations
- Bot quality issues make the swipe restrictions hit harder than they otherwise would
- Workarounds: manual context saving before key scenes, detailed opening prompts, shorter messages near the Charm limit
