Character.AI Not Working? Most of It Isn't Broken (2026)

Most of what people report as Character.AI being broken is Character.AI working exactly as designed, on the other side of a boundary nobody told them about. An age bracket. A login method that was switched off. A feature that was always paid. A queue.

That is not a defence of the platform. It sits in the lower half of our directory, and we have written at length about why it feels worse than it used to. It is a diagnostic point, and it matters because the fixes are completely different. If you are inside an account state, no amount of clearing cache, reinstalling the app, or switching to mobile data will do anything at all — and those are the first four things every other page on this subject tells you to try.

So this page sorts the symptoms by what kind of thing is actually happening, using Character.AI’s own help centre and company blog as the source. Everything below was checked on 18 September 2026.

Start Here: Four Questions That Sort Almost Everything

Work down this list. Each one rules out a whole class.

Can you browse and read, but not send a message? That is reading mode, and it is an age bracket rather than a fault.
Does it fail on every character, on every device, at the same time? That is theirs. Nothing on your side will fix it, and there is no status page to confirm it with.
Did it stop working after an update, a new phone, or a long absence? Suspect the login change, which retired an entire generation of accounts’ sign-in method.
Is the thing that “doesn’t work” a feature you have never actually had? Several of the most-searched failures are paid features on a free account, which show up as buttons that do nothing rather than as errors.

If none of those four fit, you are probably in the genuine-fault section near the bottom, which is shorter than you would expect.

The Age Bracket: Reading Mode and Verification

This is the single largest category of “Character.AI not working” searches, and almost nobody searching it knows the term for what has happened to them.

Reading mode is the under-18 experience. In Character.AI’s own words it is “the experience for users who are verified as under 18”, and in it “chat features are not available”. The platform does not present this as a punishment or an error, and it is not temporary in the way a bug is temporary — it is a state your account has been placed in.

What still works inside it, per the help centre: all past chats remain saved and viewable, you can still create Characters, scenes and videos, Imagine Chat and AvatarFX are available, and feed activity such as stickers, reactions and supporting creators continues. What does not work is the thing you came for.

You get there through age assurance, which runs on all accounts without asking. Character.AI describes it as looking at “a number of signals, including login info, activity on the platform, and some signals from third parties”. If that process reads you as an adult, nothing happens and you never see any of this. If it reads you as under 18 and it is wrong, the route back is age verification, run by a third-party provider called Persona.

Two things worth knowing before you start, because both are widely misunderstood:

Most people never have to show an ID. Character.AI is explicit that a selfie is the normal path, that an ID is requested “only as a final step” when Persona cannot resolve the selfie, and that Character.AI itself does not collect or have access to that ID. Persona keeps IDs it does collect for one week, and any biometric information is deleted within seven days.
Paying does not bypass it. The help centre states directly that a Character.AI+ subscription “does not bypass or override the age assurance process”, and that a Google account with an age on file does not either.

If you cannot find the option at all, it is not hidden: Settings → Advanced → Verify Age, on web or in the app, and on mobile you may need the latest app version first. Character.AI also notes that the prompt rolls out gradually, so not seeing it yet is normal.

Because this family is large enough to deserve its own page, we have written one: Character.AI reading mode covers what it is, every documented route out, why selfie checks fail, and what happens to the data.

If your password stopped working and you are sure it is correct, it is correct and it is also irrelevant.

Since 30 July 2024, Character.AI no longer supports the old login methods — Discord, Facebook, and email plus password are all gone. The supported routes are Continue with Google, Continue with Apple, or an emailed verification link. The company’s stated reason is that link-based and identity-provider logins “help with account security, while avoiding the additional effort associated with multi-factor verification”.

This produces a very particular kind of confusion: people who return after a break, or set up a new phone, experience a working account as a broken one. There is no password to recover because there is no longer a password.

If you pick the email link and it never arrives, the help centre’s documented causes are worth running in order, because they are unglamorous and they are usually it: the address is mistyped or is not the one on the account; the message is in spam or junk, in which case mark it Not Spam so it stops happening; your inbox is full and cannot receive anything; or you are on an email domain the system does not support, in which case a different provider will work.

The Tier Boundary: Features You Never Had

A surprising share of “not working” reports are features that were never switched on. They fail silently — a button that does nothing, a panel that stays empty — which reads exactly like a bug.

What people report	What it actually is
Auto memories not working	Facts, which captures details automatically as you chat, is a c.ai+ feature
Lorebook missing or not showing	Launched July 2026 in beta, with early access for c.ai+ subscribers
Memory bar or usage view missing	Free accounts get a simple version; the full visualisation is paid
Muted words not working	Real, and partly documented: the limit is 4 words, rising to 20 for c.ai+, and the company warns “some multilingual or complex words may not be accurately removed”
Ads appearing, or ad changes	Free carries ads by design; c.ai Lite reduces them and c.ai+ removes them
Chat quality dropped mid-conversation	On Lite, the daily premium allowance ran out and the chat continued on the standard model

That last row is the one worth dwelling on, because it is the least intuitive. Character.AI introduced c.ai Lite at $4.99 a month — below c.ai+ — and it works by giving you a daily quantity of premium-model messages rather than unlimited ones. When they run out, the company’s own FAQ says “your chat continues on the standard model — nothing is interrupted — and your premium messages reset each day”. A usage indicator is meant to show this, and the FAQ explicitly invites reports from anyone who saw the model change without one.

So a chat that gets noticeably worse partway through an evening, on a paid plan, is not a fault. It is the product.

One inconsistency to be aware of if you go looking: the help centre’s older Pinned Memories article describes pinning 5 messages per chat, while the newer c.ai Lite FAQ — updated 16 September 2026 — describes free accounts as having 15 memory pins and Lite as having 30. The newer document is the one to trust, but the disagreement is live in their documentation as of today, so do not be surprised by either number.

Behaviour That Looks Like a Bug and Isn’t

These are the searches phrased as complaints rather than errors — the character repeating itself, speaking for you, forgetting, refusing. They are worth separating out because the fix is never technical.

Repetition, drift, short replies and cut-offs are model behaviour. Character.AI’s own creator guide is unusually frank about the diagnosis, offering three causes in order: your Character’s Definition may need work; “it might just be a bad response”, in which case swiping to regenerate is “the fastest fix — and often the right one”; or it is model-level, and “if you’ve built a strong Definition and the problem persists across multiple chats, it’s probably not you”.

That last clause is a vendor admitting the thing players are usually told they are imagining. It is also consistent with what the company shipped: the April 2026 model update explicitly targeted “gibberish and cutoff responses” on DeepSqueak, and sold PipSqueak 2 on “less breaking, less drift”, better retention of earlier details, and “less repetitive” writing with “fewer loops”. Those are the exact complaints, named by the company, as things it was fixing.

Speaking for your character is the other big one, and it is structural rather than incidental — it is one of the two failure modes we have written about separately in railroading and puppeting. No setting turns it off.

Filtering has a clear documented answer and it is not a settings problem. Character.AI states that “content filtering happens at the platform level” and, plainly, “this isn’t a Definition dial you can turn”. It also concedes that filters “don’t always explain exactly why something was blocked” and that the company is adjusting them “to be less restrictive where possible while maintaining safety”. Whether that trade is the right one is a judgment for you; the factual position is that nothing you write in a Character will change it, and there is an appeals route if you believe something was flagged in error.

Forgetting has the simplest official answer on the site: “There is a limited amount of conversation context that the character can consider, so it will appear to forget things if they were not mentioned recently.” What is new is that there are now handles on it. Pinned messages and Story Memory are protected from the background tidying that trims older context — Character.AI says anything you pinned or wrote into Story Memory “stays put no matter how full the bar gets” — and both are available without paying.

This is, for what it is worth, the axis where we score Character.AI lowest, and the general problem is covered platform-agnostically in why your campaign falls apart at turn 50.

The Waiting Room Is Not an Outage

If you are being held before you can chat, that is deliberate. Character.AI runs a Waiting Room to shed load: “High traffic can strain our server capacity and cause messages to generate more slowly”, and the queue exists “to manage peak loads and preserve a relatively low-latency experience for everyone”. Documented wait: “usually just 1-2 minutes or less”, longer if traffic spikes suddenly.

There is no trick for it, and the company says it hopes to remove waiting rooms entirely once it has the capacity. Refreshing does not advance you.

Real Outages, and Why You Cannot Confirm One

Genuine server-side failures do happen — a chat error that appears on every character, a 500 internal server error, a page that will not load anywhere.

The notable thing is what Character.AI tells you to do about it. Its help centre article on the site being down does not link a status dashboard; it points you at community channels for “announcements regarding service outages, scheduled maintenance, and website status”. There is no first-party status page. For a platform of this size that is a real gap, and it is why the question “are the servers down” is one of the most-searched things about the product: there is nowhere authoritative to look.

In the absence of one, use the universality test. Try a different character, a different device, and a different network. A fault that survives all three is theirs and waiting is the only move. A fault that does not survive all three is not an outage, which sends you back up this page.

Device and App Failures, Ranked by How Often They Are It

This is the shortest section on purpose, because it is where most troubleshooting pages start and it is where the smallest share of real causes live.

Out-of-date app. This one genuinely matters, and not only for stability: the age verification option does not appear in older app versions, so “I can’t find Verify Age” and “my app is old” are frequently the same problem.
Sign out and back in. The documented fix for account state that has gone stale, including a verified age bracket that has not caught up.
A different browser or the app instead of web. Useful as a test rather than a fix: it tells you whether the fault follows your account or stays with the client.
VPN and network. Worth toggling once, since age assurance draws on signals that a VPN can muddy, but do not spend long here.

If all four come back clean, the cause is upstream of your device and the rest of this page is where the answer is.

When the Answer Is That It Is Not For You Any More

There is a class of report this page cannot fix, because nothing is broken and nothing is a setting: the product changed and you preferred it before.

Character.AI has been candid about several of these changes at once. In April 2026 the company acknowledged rolling out “age restrictions to more regions”, introducing “usage limits on some features” and adding “more ad placements for free users”, said those landed close together, and wrote: “We know your experience took a hit. We hear you.” Its stated reason is cost — “a small team serving millions of users every month with no outside investors” keeping a free tier viable.

Since then the company’s published direction has been comics, short-form video, image models and a broader entertainment platform, which is a strategy we have examined in detail in why Character.AI feels so bad now. If what you want is a long, remembered roleplay campaign, that is worth reading before you spend another week troubleshooting — and the alternatives list is organised by what you would be leaving for, not by rank.

For everything else: check whether you are in an account state before you check your cache.

Frequently Asked Questions

Why is Character.AI not working? In most cases it is working, and you are on the other side of a boundary. The four that account for the bulk of reports are an age bracket you have been placed in, which disables chat and is called reading mode; a login method that was discontinued in July 2024, which is why an old email and password no longer works; a tier limit, since several features people assume are broken are paid ones that were never switched on for a free account; and the Waiting Room, which is deliberate queuing during peak traffic. Genuine outages exist but are a smaller share than the search volume suggests, and they look different: a chat error or a server error that appears on every character, on every device, at the same moment.

Why is my Character.AI stuck in reading mode? Reading mode is not a bug and not a stuck screen. It is the experience Character.AI gives to users its age assurance system has assessed as under 18, and in it chat is unavailable while browsing, past conversations, character creation and feed activity all still work. The only documented way out is to complete age verification, which is run by a third-party provider called Persona and normally needs only a selfie. If you cannot see the option, it is under Settings, then Advanced, then Verify Age, and on mobile you may need to update the app first.

Why won’t Character.AI let me log in? Because the old login was retired. Since 30 July 2024 Character.AI no longer supports Discord, Facebook, or an email and password combination. The supported methods are Continue with Google, Continue with Apple, and an emailed verification link, and the company’s stated reason is account security. If you chose the email link and nothing arrives, the documented causes are a mistyped address, the message sitting in spam, a full inbox with no room to receive it, or an email domain the system does not support.

Are Character.AI servers down right now? There is no first-party status page to check, which is itself worth knowing. Character.AI’s own help centre answers the question by pointing people at its community channels for outage announcements rather than at a dashboard. The practical test is whether the failure is universal: if every character fails on every device and network at once, it is theirs, and waiting is the only fix. If one chat fails while another works, or a phone fails while a browser works, it is not an outage and something on your side or in your account is the cause.

Why does Character.AI keep forgetting things? Character.AI’s own answer is that the amount of conversation a character can consider is limited, so anything not mentioned recently appears to be forgotten. What has changed is that there are now controls over it. Pinned messages and Story Memory are protected from the background tidying that trims older context, and are available without paying. Facts, which capture details automatically, and private Lorebooks are paid features, so an account that has never seen them working is not broken. This is also the axis on which we score the platform lowest.

Why did Character.AI change model in the middle of my chat? On the c.ai Lite plan that is the daily premium allowance running out, which the company documents as working as designed: the conversation continues on the standard model rather than stopping, and the allowance resets the next day. A usage indicator is supposed to show it happening. Separately, the free and paid chat styles are genuinely different models, currently PipSqueak 2 for free users and DeepSqueak for subscribers, so a change in quality after a plan change or a rollout is a real difference and not something you can adjust in settings.

Checked against Character.AI’s help centre and company blog on 18 September 2026. Character.AI is a trademark of Character Technologies, Inc.; Arcanum is an independent publication with no affiliation to it, and references here are nominative. Documented behaviour, tiers and prices are accurate as of that date and change often on this platform.