Skip to main content

AI features — every tool, and how they fit together

Our AI can run fan conversations and sales for you — but "the AI" is really several distinct tools, and most mistakes come from confusing them or missing a quiet setting. This guide maps every one: what it does, where to find it, and the traps that break it. Start with two rules:

  • The big split: AI Agents run conversations and sell autonomously; AI Generate writes one draft for a human to send. They're separate systems — don't confuse them.
  • Everything runs on credits. All AI features draw on your org's credit balance (1 credit = $50). Below ~0.5 credit, every AI feature silently stops.

Group A — The autonomous AI chatter​

AI Agents — the core · /dashboard/ai/agents (Beta)​

One autonomous chatter per model + platform. It replies, sends free teasers, sells PPV and sends PPV reminders on its own.

  • Modes: FULL_AUTO (chat + sell + teasers) · CHATTING_ONLY (chat, no selling) · PPV_ONLY (sell only; ignores the message limit).
  • Key settings: message delay · max messages per chat · active hours (daily schedule, empty = 24/7) · max daily spend $ · spend-notification threshold · gender (the only "persona" control) · allowed languages · excluded-fan lists · default photo/video price.
Must-knows
  • An agent only runs once at least one platform is attached. No platform = it does nothing.
  • There is no on/off button. To stop one: remove its platforms, use the active-hours schedule, or delete the agent.
  • Set a max daily spend. Leave it empty or 0 and the agent silently never sends.
  • One agent per model + platform (a second assignment errors).
  • Languages: you can pick NL + EN (DE/FR/ES are disabled); empty = auto-detect.
  • AI sales appear under the name "Dirty Harry."

Free teasers — part of a FULL_AUTO / PPV_ONLY agent​

No separate screen: the agent sends any content you've priced at $0 as a free teaser to warm a fan up. Capped at 3 free items per fan, ever (max 1 per turn); only content with a base price of exactly 0 counts, and already-received content is skipped.

AI dashboard & stats · /dashboard/ai (Beta)​

Agent counts, how many are active (= have a platform), tag count, AI revenue/sales (day/week), AI messages today, and AI costs (day/week/month per agent).

Group B — AI help for human chatters​

AI Generate — a reply suggestion · in the chat (⋯ menu by the composer)​

The chatter clicks and the AI writes one reply to edit and send. It never sends by itself.

  • Per click: tone (Chatty / Flirty / Horny) · persona (Dirty Debbie / Dirty Harry) · language.
  • Tone and persona are not saved per model — they apply per click, in your browser (tone changes explicitness, persona changes gender framing, nothing more).
  • It always drafts in English, then translates. Pick a language the engine can't handle (see Languages) and you get an empty reply. It can also come back empty on a safety guardrail (identity leak, incoherent, "sounds like the agency"). Needs ≥ 0.5 credit.

Auto-Translate & humanizer · /dashboard/translators ("Translation")​

Translates incoming and outgoing chat and "humanizes" it (adds creativity and deliberate typos) so translated/AI text reads hand-typed. Used manually by chatters and automatically by AI Generate.

  • It's org-wide — there is no per-model language. Default target is NL.
  • The engine supports only 5 languages: EN, NL, DE, FR, SV (other languages some menus show will fail). The translation cache never expires, and it translates even when source = target.

Group C — Sales sequences & the content the AI uses​

PPV Scripts — multi-step sell funnels · /dashboard/ai/ppv-bundles (titled "PPV Scripts")​

Reusable, ordered sequences of PPV steps (each step = content + base/min/max price + a pitch instruction) that the AI pitches in order.

  • Targeting per script: min fan score · min excitement · min/max messages between · available from/to · active flag.
No menu link

You reach PPV Scripts only by URL (/dashboard/ai/ppv-bundles). The page title "PPV Scripts" differs from the route name "PPV Bundles." Content is per model (not per platform). The min/max price hides behind "+ Add min/max price range." An inactive script, or a step with no content attached, gives the AI nothing to run.

Content tags, Auto-Tag & pricing — the fuel the AI matches and sells on​

The tags, AI-descriptions and prices the AI uses all live in the vault, and that setup is the single biggest lever on AI quality — it has its own guide: → Set up your vault for the AI. In short: Auto-Tag (vision) proposes tags + a description (max 10 items per call; confirming costs credits), and content pricing drives PPV prices and $0 teasers (falling back to the agent default, then a built-in default).

Group D — Rule-based automated messaging (not the AI chatter)​

These send on triggers/segments, separate from the autonomous agent. All under /dashboard/automations:

Automated Messages ("Single Messages")​

Auto-send one message to a fan segment on a trigger (keyword / fan type; online-only and spend-ranges supported).

  • Editing is limited: a save can't change the delete option, repeat, delay, fan type, schedule, send-window or exclusion lists — to change those, delete and recreate.
  • Activation is blocked if another active message on the same model + platform has an overlapping day-window (no window = all day = overlaps everything). Send-windows are in UTC.
  • Throttle needs both fields ≥ 1. "Purge all" is a permanent hard-delete that also wipes analytics.

Message Sequences (drip)​

Step 1 fires on a keyword; later steps follow after a per-step delay, and each fan moves forward once. Progress never rewinds, and saving a new set of steps deletes and recreates all of them (there's no single-step edit).

PPV Follow-ups​

After a PPV, auto-send one reminder after a delay (optionally suppressed if the fan already replied). Delay is 1–30 minutes. One config per model + platform — saving a new one overwrites the existing.

AI Generate (for automations)​

A draft helper that generates a batch of sequence concepts to review; it doesn't create anything itself. Style is conversational or flirty only. On an error it quietly returns generic fallback text (the only signal is a fallback flag). Needs ≥ 0.5 credit.

Group E — Compliance monitoring​

Banned Words · /agency/banned-words​

An org watchlist of forbidden terms. If a sale message contains one, you get a BANNED_WORD_DETECTED notification plus audit views. Turning it on the first time seeds 50 default words. (Full walkthrough: → Banned Words.)

Banned words don't filter output

Banned Words does not block or filter the AI's or a chatter's output. It's detection and alerting after the fact — the words are not part of the AI's prompt. Use it to catch problems, not to prevent them in-message.

Message Templates / Quick Messages · /dashboard/message-templates ("Scripts & Templates")​

Canned messages for human chatters. A chatter convenience tool — not AI.

What the AI does not do (so you don't count on it)
  • No training or learning loop. Nothing you do teaches or fine-tunes the AI; there's no rating/feedback mechanism. You steer it with bios, tags, descriptions, prices and PPV scripts — not by "teaching" it. (Per-chat context + your static scripts/tags is not "training.")
  • No saved per-model persona or tone-of-voice. The agent gets only a bio, gender and languages; AI Generate's tone/persona is per click, not stored per model.
  • No per-model translation language — translation is org-wide.
  • No positive spend/fan-type targeting for the AI chatter. The agent has exclusion lists, a spend notification and a daily spend cap — but you can't point it at a spend range or fan type. (That targeting lives in Automated Messages, not the AI chatter.)
  • No per-model AI on/off switch — an agent with a platform attached is the switch.

Three traps that cut across everything​

  1. Languages: different screens offer different lists (AI Generate 9, agents 6), but the engine only does 5 — EN, NL, DE, FR, SV. Anything else comes back silently empty. Keep fans' languages to those five.
  2. Silent blanks: both AI Generate and the automation generator can return empty or generic fallback text with no error — usually low credit, a guardrail, or an engine hiccup. If output goes blank, check those first.
  3. Credits: every AI feature runs on the org credit balance (1 credit = $50). Below ~0.5 credit, they all stop.

FAQ​

Is there an on/off switch for the AI?

No. An AI Agent with at least one platform attached is "on." To stop it: remove platforms, use active hours, or delete the agent. And always set a max daily spend — without one, the agent silently never sends.

Does the AI learn from our corrections?

No — there's no training or feedback loop. You steer it with bios, tags, descriptions, prices and PPV scripts, not by "teaching" it.

Can I save a tone or persona per model?

No. AI Generate's tone and persona are chosen per click (tone = explicitness, persona = gender framing). The agent's only persona control is gender.

Why did an AI reply come back empty?

Most often: an unsupported language (the engine only does EN/NL/DE/FR/SV), low credit (< 0.5), or a safety guardrail.

Can I target the AI chatter at big spenders or a fan type?

Not positively — the agent only supports exclusion lists (plus a spend cap and spend notifications). Spend/fan-type targeting lives in Automated Messages.

Why is a PPV Script missing from the menu?

It has no menu link by design — reach it by its URL. (The page is titled "PPV Scripts" but the route is "PPV Bundles.")

Do banned words stop the AI from saying them?

No — banned words are monitoring/alerting only. They don't filter the AI's output and aren't in its prompt.

Last updated on