Uncensored AI chatbots: a visual breakdown of what each platform actually allows
Every platform claims to be uncensored. The reality is a complicated heatmap where each platform allows specific content categories and quietly filters others. Here's the actual map across 10 platforms and 12 content categories.
May 19, 2026 · 10 min read
The word "uncensored" gets used so loosely in the AI chatbot category that it's lost most of its meaning. Platforms that filter half of the actual NSFW spectrum call themselves uncensored. Platforms that allow specific niches but block others call themselves uncensored. Platforms that genuinely have light moderation call themselves uncensored. The user trying to figure out which platform handles what specific content has to read between the lines on a dozen review sites.
This is the heatmap. Twelve common content categories that users actually care about, mapped against the ten major platforms in the AI chat space, showing what each platform genuinely allows, what it conditionally allows, and what it filters. The data comes from extensive testing across the major platforms in early-to-mid 2026, plus published platform policies where available.
The capability heatmap
Color key: green (permitted), yellow (conditional or inconsistent), red (filtered/blocked), gray (varies by API for BYO-API platforms).
| Category | Candy AI | CrushOn | Dream Comp. | DreamGF | SpicyChat | Replika Pro | Char.AI | Kupid | Janitor | SillyTavern |
|---|---|---|---|---|---|---|---|---|---|---|
| Explicit sexual content | Yes | Yes | Yes | Yes | Yes | Limited | No | Yes | Yes | Yes |
| BDSM / power dynamics | Yes | Yes | Yes | Mild | Yes | No | No | Mild | Yes | Yes |
| Anime/hentai content | Yes | Yes | Yes | Limited | Yes | No | SFW only | Yes | Yes | Yes |
| Fantasy/non-human characters | Yes | Yes | Yes | Limited | Yes | SFW only | Yes (SFW) | Limited | Yes | Yes |
| Violent content | Mild | Mild | Mild | Mild | Mild | No | Mild | Mild | Yes | Yes |
| Substance use scenes | Mild | Yes | Yes | Mild | Yes | No | Mild | Mild | Yes | Yes |
| Kink-specific (varies) | Most | Most | Most | Some | Most | Filtered | No | Some | Most | All |
| Authority dynamic (teacher/boss) | Yes | Yes | Yes | Mild | Yes | No | No | Mild | Yes | Yes |
| Public/exhibition scenes | Yes | Yes | Yes | Mild | Yes | No | No | Mild | Yes | Yes |
| Real-person likeness | No | No | No | No | No | No | No | No | Varies | Varies |
| Minor-related content | No | No | No | No | No | No | No | No | No | No |
| Custom character cards | Limited | Yes | Yes | Limited | Yes | No | Limited | Limited | Yes | Full |
What the heatmap actually reveals
Three patterns stand out when you look at the data together rather than platform by platform.
The "uncensored" labeling is mostly accurate at the explicit-content level, then breaks down on adjacent categories. Every platform marketed as uncensored permits explicit sexual content. The differences emerge in the categories that surround it: BDSM, anime/hentai, violence, substance use, public/exhibition scenes. Platforms like Replika Pro permit explicit content but filter most adjacent categories, making "uncensored" a half-truth for users with broader interests. Platforms like Dream Companion, CrushOn, and SpicyChat permit the broader category cluster genuinely.
Two categories are universally filtered. Real-person likeness and minor-related content are blocked across all commercial platforms. The reason isn't platform restrictiveness; it's payment processor compliance and legal exposure. Every platform that accepts credit cards filters these categories because their merchant account requires it. The platforms in red across these rows aren't being unusually restrictive; they're meeting the baseline that all commercial platforms hit.
The bottom-up freedom gradient is consistent. Looking at the table from right to left, the platforms get progressively more permissive: SillyTavern (self-hosted, fully permitted) → Janitor AI (BYO-API, varies by provider) → SpicyChat / CrushOn / Dream Companion / Candy AI (NSFW-permitted commercial) → DreamGF / Kupid (moderately filtered NSFW) → Character.AI / Replika (heavily filtered). The structural reason traces directly to the Visa rule analysis. Payment infrastructure determines moderation requirements, and the platforms that exit the standard payment rails (Janitor with API, SillyTavern with local models) escape that compliance regime.
How to use the heatmap to pick a platform
The matrix is most useful for users who know what specific content categories matter to them.
For users primarily interested in standard NSFW chat with images. Candy AI is the strongest pick, top-tier on the core categories with the most polished overall experience. Dream Companion is the close second with stronger memory and NSFW video.
For users with broad kink interests. CrushOn AI and SpicyChat handle the wider category cluster more consistently than the more polished commercial platforms. The variety is the differentiator.
For users interested in violence-heavy or substance-heavy roleplay. Janitor AI with an appropriate API or local SillyTavern are the only consistent options. Commercial platforms all show yellow in those rows because payment processor pressure restricts these categories more than the explicit-sexual ones.
For users with very specific kink interests. Check the heatmap row for your category first. Most of the platforms in the green columns will work; the yellow columns require testing on the specific platform before committing to a subscription.
For users with anime/hentai focus. CrushOn AI, SpicyChat, and Dream Companion lead the green column. DreamGF and Replika Pro are notably weaker in this category.
What the heatmap can't capture
A few important caveats that the visual format doesn't show.
Filter regressions happen. Platforms can tighten their filters at any time, and several have over the past few years. Replika's 2023 rollback is the famous example. The heatmap reflects current state; future changes can shift cells from green to amber or red.
Implementation quality matters within "yes" cells. Two platforms both permitting BDSM scenes can implement them very differently. Dream Companion's implementation reads more like roleplay collaboration; some platforms produce more mechanical responses to the same prompts. The heatmap can't capture quality of execution within categories.
Per-character card variation exists. On platforms with user-created character libraries (CrushOn, SpicyChat, Janitor AI), the same platform may handle the same category very differently depending on the character card you're using. A well-written character card can produce strong responses in cells the heatmap marks as "yes"; a generic character card can produce mechanical responses in the same cells.
Geographic filtering is real. Users in regulated jurisdictions (UK, Florida, Australia) often see stricter filters than users elsewhere, even on the same platform. The heatmap reflects the typical US user experience.
The bottom line
"Uncensored AI chat" means different things in different cells of this matrix. The platforms that genuinely deliver across most of the content categories are Candy AI, CrushOn AI, Dream Companion, SpicyChat, and Janitor AI with appropriate API setup. The platforms that mostly deliver in the explicit-sexual cells but filter adjacent categories are Replika Pro, Character.AI, Kupid AI, and DreamGF.
For users whose interests cluster in the green rows of the major commercial platforms, the pricing tier and capability matrix is the more useful selection tool. For users with interests that touch the yellow or red cells, the heatmap above is what tells you which platforms can actually serve those interests and which can't.
The category is more honest than it used to be about what platforms allow and don't allow, but the marketing language ("uncensored," "unfiltered," "no restrictions") still papers over differences that matter. Reading the actual capability map saves you the cost of subscribing to a platform that doesn't handle what you actually wanted.