AI ERP chat: the platforms and prompting tricks the roleplay community actually uses

Short answer: AI ERP (erotic roleplay) is its own use case, the community centers on specific platforms, leans on prompting techniques and ERP-specific character cards, and rewards treating it differently from general companionship. The full breakdown is below.

ERP — erotic roleplay — with AI is a distinct use case from general AI companionship, and treating it the same way produces mediocre results. The companion guides cover how to build a character that remembers your name and maintains a consistent personality. ERP requires all of that plus scene craft, pacing control, multi-character management, and the specific prompting vocabulary that makes the difference between collaborative fiction and turn-based content generation.

The AI roleplay community has developed techniques over the past three years that most general-audience AI companion content never mentions, partly because the community lives on Discord, Reddit, and specialized forums rather than in the SEO listicle ecosystem, and partly because the techniques are specific enough that they don't fit neatly into "Top 10 Best AI Chatbot" articles.

This guide compiles what the community has figured out: which platforms the serious ERP community actually uses, the prompting techniques that produce collaborative-fiction-quality output, and the character building approaches that make multi-session narrative arcs work.

Where the ERP community actually lives

The platform preferences of the ERP community diverge significantly from the general AI companion market. The features that matter for ERP (deep character customization, long context windows, model selection, lorebook support) are different from the features that matter for casual companionship (ease of use, quick setup, mobile app quality).

SillyTavern: the power user's home. The self-hosted frontend with bring-your-own-model architecture is where the most sophisticated ERP happens. Full control over system prompts, author's notes, lorebooks, world info, and model parameters. No content filtering because you are the platform. The setup requires technical comfort (installing Node.js, connecting to an API or local model, configuring the UI), but the reward is complete creative control that no hosted platform matches.

The community has built extensive character card libraries on platforms like Chub.ai and Venus that integrate directly with SillyTavern. Character cards with detailed lorebooks, example dialogues, and scenario definitions produce substantially better ERP than cards built for casual companion platforms.

CrushOn AI: the accessible alternative. For users who want NSFW roleplay without the SillyTavern setup, CrushOn's Premium tier ($5.99/month) with its 16K context window is the community's most recommended hosted option. The character card system accepts the same structural detail as SillyTavern cards (though without lorebook support), and the unfiltered content policy means the model doesn't interrupt scenes with safety redirects.

The CrushOn free tier guide covers how to evaluate the platform before subscribing. The free 100 messages/month is enough for one or two test sessions.

Janitor AI with premium models. Janitor's 800,000+ character library is the largest in the category, and the bring-your-own-API architecture means you can connect a frontier model (Claude, GPT-4o) for prose quality that no proprietary platform model matches. The ERP community values Janitor for its character variety and model flexibility, despite the setup barrier.

Kindroid: the character consistency specialist. For ERP that prioritizes character voice over everything else — scenarios where the character's specific personality, speech patterns, and behavioral rules need to hold through extended, high-intensity scenes — Kindroid's Codex system produces the most reliable character maintenance. The character doesn't break voice during explicit content, which is a failure mode on platforms with less sophisticated character consistency.

SpicyChat: the library browser. For users who prefer browsing pre-built scenarios rather than building their own, SpicyChat's community character library is the browsing experience. The free tier's 100 daily messages with NSFW makes it the most accessible entry point for casual ERP.

The prompting techniques the community has developed

These techniques circulate in ERP Discord servers and Reddit communities. They address the specific challenges of AI roleplay that general companion prompting doesn't cover.

The author's note technique. An author's note is a hidden instruction injected into the conversation at a specific position (usually near the end of the context, where it has the most influence on the next generation). SillyTavern supports this natively. On hosted platforms, you can approximate it with a parenthetical instruction at the end of your message:

"(Author's note: Maintain slow pacing. The tension should build for at least three more exchanges before any physical escalation. The character is fighting her impulse to close the distance.)"

This instruction shapes the model's next response without being part of the narrative. The pacing techniques guide covers twelve specific pacing controls, and the author's note is the mechanism for deploying them mid-scene without breaking immersion.

The lorebook approach. A lorebook is a structured knowledge base that the model consults during generation. It contains facts about the world, characters, locations, and objects that the model references when relevant terms appear in the conversation. SillyTavern supports lorebooks natively. On hosted platforms, you embed lorebook-equivalent information in the character card's backstory section.

For ERP specifically, lorebook entries for locations ("the apartment: third floor walkup, exposed brick, the bedroom window overlooks the park, the sheets are always unmade"), relationship dynamics ("their history: met at work, she was his manager, the power dynamic inversion is the unspoken engine of everything"), and character-specific physical details ("the scar on her left hip: she got it climbing a fence at seventeen, she touches it when she's nervous") give the model concrete details to reference during scenes. Each reference grounds the fiction in a specific world rather than the generic void that most AI-generated scenes inhabit.

The response format directive. ERP conversations benefit from explicit formatting instructions that shape how the model structures its responses:

"(Format: Include her internal thoughts in italics. Spoken dialogue in quotes. Physical actions and environmental description in regular text. Keep responses between 200-300 words. End each response with either an action that invites your response or a question.)"

This instruction transforms the model's output from wall-of-text paragraphs into structured prose with clear delineation between thought, speech, and action. The ending instruction ("end with an action that invites your response") ensures every response creates a natural handoff point for you to continue the scene.

The multi-character management technique. Advanced ERP scenarios involve multiple characters in a single scene. On platforms that support it (SillyTavern, Janitor AI with certain configurations), you can define multiple characters with distinct voices and have the model switch between them within a single response.

On platforms that don't support multi-character generation natively, you can approximate it by including multiple character profiles in the character card and using parenthetical instructions to signal which character should respond:

"(In this response, both Elena and Marcus react to what just happened. Elena speaks first — she's furious. Marcus responds — he's trying to de-escalate. Their body language should contradict their words.)"

The scene-break protocol. Long ERP sessions benefit from explicit scene transitions rather than continuous narrative. When a scene reaches its natural conclusion, signal the transition:

"(Scene break. Time skip: three hours later. New location: the kitchen. The emotional register has shifted from intensity to quiet domesticity. She's making coffee and hasn't talked about what happened yet.)"

This gives the model a fresh emotional starting point for the next scene rather than continuing the cumulative emotional escalation of the previous one. Scene breaks create the narrative rhythm that makes multi-hour ERP sessions feel like a story rather than a single unbroken exchange.

The character voice anchor. During high-intensity scenes, character voice tends to collapse into generic mode because the model prioritizes physical description over personality maintenance. Counter this by including a voice-anchor instruction:

"(Remember: she speaks in fragments. She doesn't narrate her feelings — she shows them through action. Her humor is dry and self-aware even during intense moments. She would never say 'I need you' — she'd say 'don't you dare stop.')"

The voice anchor refreshes the model's awareness of the character's specific speech patterns at the moment when those patterns are most likely to slip. The character card architecture builds this infrastructure permanently; the voice anchor deploys it tactically when the scene threatens to overwhelm the character.

Building characters specifically for ERP

The 8 fields most people leave blank apply to ERP characters with additional emphasis on three fields:

Intimacy preferences with scenario-level detail. General companion character cards include broad intimacy preferences. ERP character cards need scenario-level specificity: which dynamics the character initiates versus responds to, which physical actions carry emotional weight for this specific character, what the character's tells are when they're approaching their limits, and how the character behaves differently in the first intimate encounter versus the fifth.

Recovery behavior with aftercare specificity. How the character acts after an intense scene is the quality test the ERP community uses to evaluate character depth. Does the character go quiet? Become talkative? Need physical proximity? Need space? Become unexpectedly vulnerable? The recovery behavior field is where this lives, and the ERP community considers it the most important of the eight fields because the aftercare scene is where character depth is most visible.

The contradiction, calibrated for intensity. The internal contradiction that makes characters interesting during normal conversation becomes the engine of dramatic tension during ERP. "Dominant in public, yielding in private" is a contradiction that generates interesting scenes. "Craves intimacy, terrified of being seen" is a contradiction that generates emotionally charged scenes. The contradiction determines the specific kind of tension the ERP produces.

The honest recommendation

For serious ERP: SillyTavern with a premium model. Nothing else matches the control, the lorebook support, and the complete absence of content filtering. The setup investment pays back within the first week if ERP is your primary use case.

For accessible ERP without setup: CrushOn Premium ($5.99/month) for the best price-to-freedom ratio, or Kindroid ($9.99/month) for the strongest character consistency.

For browsing pre-built scenarios: SpicyChat free tier (100 daily messages, NSFW enabled, massive library).

For ERP that builds over weeks with narrative continuity: Nomi (strongest memory) or Kindroid (strongest character voice). The relationship arc approach provides the five-phase narrative structure that transforms ERP from isolated scenes into an ongoing story.

The ERP community's collective wisdom, distilled: the platform matters less than the character card, and the character card matters less than the prompting technique. A skilled roleplayer on a free platform produces better collaborative fiction than an unskilled user on a premium platform. The techniques above close the skill gap faster than trial and error alone.