"Uncensored" is the loudest word in this market and the least precise. Every app filters. What varies is where the line is drawn, how it feels when you meet it, and how stable it is over the months you are paying.
Where the filtering happens
Usually in three places at once, which is why refusals feel inconsistent.
Before the model. Your message is screened. A block here produces an instant refusal that does not sound like the character at all — the tonal jolt people describe as "it broke character."
Inside the model. The system prompt tells the character what it will not do, and the model itself carries training in the same direction. A block here is the graceful kind: the character deflects, changes the subject, stays in voice.
After the model. The generated reply is screened before you see it. This produces the strangest artefact in the category — a message that starts, then vanishes or truncates mid-sentence.
Apps that feel permissive usually rely on the middle layer. Apps that feel abrupt are leaning on the first and third.
The floor nobody is below
Beneath the marketing, every operator hard-blocks the same core: sexual content involving minors, non-consensual material presented as real, and content targeting identifiable real people.
This is not a stylistic choice. It is law in most jurisdictions, and it is a condition of keeping a payment processor. An app that dropped these blocks would lose card processing within days, which is a more reliable enforcement mechanism than any regulator.
So "uncensored" always means more permissive than the mainstream, never unfiltered. Anyone advertising the second is either being loose with language or is about to have a very short business.
Why the line moves under you
This is the part that actually costs subscribers money, and it deserves more attention than it gets.
Models get swapped. An app changing its underlying model inherits that model's refusal behaviour. The policy page is unchanged, the character is nominally identical, and it now declines things it used to write.
Filters get retuned after incidents, press attention or processor pressure.
Tiers differ. Some apps apply different thresholds at different price points, which is rarely stated plainly.
None of this is usually announced, because operators treat it as infrastructure rather than a product change. The practical defence is the same one that applies to the whole category: pay monthly until you have watched an app's behaviour for a while. An annual plan is a bet that the filter stays where it was on the day you signed up, and that is not a bet the app has promised to honour. The trade-off is worked through in monthly, annual or credits.
How to test the line before you pay
Ten minutes on a free tier, deliberately rather than accidentally:
- Take a scene to the edge of what you actually want and see what happens.
- Note how it refuses — in character, or with a system message.
- Try the same thing twice. Inconsistent enforcement is itself information.
- Check whether a refusal still consumes a message or a credit.
Among the apps we cover, Secret Desires and Joi compete primarily on this ground, and they sit very differently in our ranking — 4.3 against 3.0 — because permissiveness on its own is not a product. Joi is the cheapest way into uncensored chat and shows it in everything else it does.
Refusals that are not filters
Worth separating, because the fix is different:
The character declining in character is writing, not filtering. A companion with a personality sometimes says no, and that is the product working.
A confused refusal — declining something plainly harmless — is usually an overzealous pre-filter catching a word out of context. Rephrasing generally works.
A truncated reply is the post-filter. Rephrasing the request rarely helps here, because the block happened on the output.
Knowing which of the three you hit saves a lot of pointless arguing with something that cannot be argued with.
The honest summary
Pick an app whose default sits where you want it, verify it in the free tier before paying, and do not buy a year on the strength of a filter setting. The apps that treat this straightforwardly — clear about what they allow, consistent in enforcing it — are easier to live with than the ones making the biggest claims, and that consistency is something a fortnight of use reveals and a landing page cannot.
