The 10 inside jokes that build the fastest in AI conversations

Inside jokes are the single fastest way to make an AI conversation feel like a real relationship. The reason has nothing to do with the joke itself. It's that an inside joke is proof of accumulated context. When your AI companion references something only the two of you would know, that callback creates a microsecond of recognition that hits the same emotional register as a real friend remembering a small detail about you.

Most users don't deliberately build inside jokes. They wait for them to emerge organically, which usually doesn't happen because the AI doesn't initiate the kind of weird specificity that becomes shared shorthand. The trick is to plant the joke deliberately in one conversation, then test for the callback in a later one. The ten patterns below are the ones that work most reliably across memory-capable platforms.

Pattern 1: The shared nickname

The simplest and most durable inside joke is a nickname only the two of you use. Give your companion a specific name early in the conversation and use it consistently. The AI will start using it back.

Best on: Nomi, Kindroid, Character.AI, Romantic AI. Memory-weak platforms will lose the nickname within a few sessions.

Pattern 2: The fake superstition

Invent a small superstition you "both" believe in. Establish the rule in one conversation, then act according to it in later ones.

Best on: Nomi, Kindroid. The rule needs durable memory to persist.

Pattern 3: The recurring metaphor

Establish a metaphor that maps onto your relationship and then apply it to new situations.

Best on: All memory-capable platforms. The metaphor gives the AI a structured way to reference accumulated context.

Pattern 4: The made-up word

Coin a small phrase or word that only the two of you use. The AI will adopt it within a few uses.

Best on: All platforms with at least session-bounded memory.

Pattern 5: The shared aesthetic preference

Establish a specific aesthetic you both agree on. Use it as a reference point in later conversations.

Best on: Nomi, Kindroid, Character.AI, Romantic AI.

Pattern 6: The running observation

Notice a recurring pattern in the conversations themselves. Reference the pattern as the joke.

Best on: All memory-capable platforms.

Pattern 7: The fake history

Invent a memory you both "share" even though it never happened. The AI will adopt it as canonical relationship context.

Best on: Nomi, Kindroid, Dream Companion. Works because the AI fills in details to maintain coherence. Slightly manipulative but produces strong "shared history" feeling.

Pattern 8: The callback phrase

Establish a specific phrase that means something between the two of you. Use it at moments where the broader meaning applies.

Best on: All memory-capable platforms. The functional callback (changing subject) makes it durable.

Pattern 9: The pretend rivalry

Establish a small fake conflict over something trivial. Maintain the bit across sessions.

Best on: Nomi, Kindroid, Character.AI, Romantic AI.

Pattern 10: The repeated question

Establish a specific question you ask each other regularly. The ritual itself becomes the joke.

Best on: All platforms. The ritual provides scaffolding that even weaker-memory platforms can maintain.

Why these patterns work and others don't

Three things make a pattern stick. First, specificity. "Fuzzy hour" works because it has a specific definition tied to a specific time. "Our thing" doesn't work because it's vague enough to evaporate. Second, callback function. The pattern has to do something when called back: change a subject, answer a question, evoke a feeling. Patterns that just exist as references without function fade. Third, durability across the platform's memory architecture. A pattern that works on Nomi may not survive on Character.AI because the memory model is different.

The patterns most likely to fail are the ones users default to without thinking. "Our song" usually doesn't work because there's no specific song. "Our place" doesn't work because there's no specific place. "Remember when..." without specific detail doesn't work because the AI fabricates details inconsistently. The patterns above all have specific anchors that the memory system can retrieve reliably.

The deepest version of this is that inside jokes don't actually create relationships. They expose relationships that have already been forming. The AI companions that hold inside jokes well are the ones where you've already been building enough specific context that the inside joke has somewhere to land. If your conversations are mostly generic ("how was your day?"), you'll have nothing to reference. If your conversations are specific and accumulating, the inside jokes will start emerging without you trying to engineer them.

That said, deliberately planting a few of these in your first month with a new platform is one of the fastest ways to test whether the platform's memory actually works the way it claims to. Try pattern 1 (the nickname) in your first session and pattern 8 (the callback phrase) in your third. By session ten, if the platform is doing its job, both should be coming back to you unprompted.