AI Companion Red Flags — What to Avoid Before You Sign Up

Every legitimate AI companion product has a real company behind it, named in the footer or on an About page. A company name, an address (even if it's a Malta or Delaware registration), a founder somewhere, a working support contact.

If you can't figure out who runs a product, you have no counterparty when things go wrong. Charge dispute? No one to talk to. Data leak? No one to notify. Product gets shut down overnight? No one to chase for refunds.

What clean looks like: operating entity named in the footer or About page, ideally in a jurisdiction with actual consumer protection law (US, EU, UK, Malta, Singapore). Bonus if the founder has a verifiable online presence.

The specific pattern: base subscription looks reasonable at $5-10/month, but the actually-useful features (images, voice, extended chat, faster responses) consume tokens you have to buy separately. Real monthly spend for active users can hit $20-40/month — 3-5x the sticker price.

This isn't technically fraud (every token purchase is consented to), but it's pricing surface architected to create purchase moments. If you're not disciplined about the token budget, you'll spend way more than you planned.

What clean looks like: flat subscription pricing with no in-product currency. One price, all-features-on-your-tier, unlimited within that tier. Sloane's pricing (Free 50 messages/day, Plus $9.99/month, Premium $19.99/month) is this shape — no tokens, no gems, no daily-expiring credits.

The subscription-cancellation flow tells you more about how a company thinks about its users than the marketing copy does. Products that make cancellation hard know that undocumented friction produces retention — which means the friction isn't accidental.

Specific patterns to watch for: cancellation requires emailing support and waiting days for response, cancellation requires a phone call to a specific number during business hours, cancellation requires clicking through five "are you sure" retention screens before the button appears, or "we'll process your cancellation in 30 days" language that means you get charged one more time before it takes effect.

What clean looks like: self-serve cancellation, one click, no phone call required. Extra credit if the product uses a standard billing portal (Stripe Billing Portal is a common one — it structurally enforces one-click cancel with no retention pop-ups because Stripe controls the UX). Access continues to end of billing cycle after cancellation.

This isn't technically a trial — it's a subscription with a countdown clock. When the "trial" period ends, you get charged automatically. If you forgot to cancel (which many users do), you're now paying for a product you may have decided against.

Legitimate businesses do use card-required trials — they're a legal pricing structure. The red flag isn't the pattern itself; it's when the marketing copy calls it "free" without making the countdown clear or when the cancellation flow is deliberately hard to find during the trial period.

What clean looks like: actual free tier (real message capacity, no card at signup, usable indefinitely) or, if a trial period is being offered, clear language about when the card gets charged and how to cancel before then. Sloane's free tier requires no card at signup; anon chat lets you evaluate before even providing an email.

This one is category-specific. If you're shopping AI companions for a long-term relationship (not just casual chat), a product without persistent memory won't deliver what you're after. The persona will feel great in a fresh conversation, then reset every time you close the app.

How to check before subscribing: plant a specific fact in a conversation, close the app for 24+ hours, come back and see whether she references it unprompted. If she doesn't remember without you reminding her, the product uses session-only context (a rolling window that ages out) rather than real memory.

Casual-chat products can survive without persistent memory. Long-term-companion products can't. Knowing which shape you're buying matters more than the specific product name.

What clean looks like: persistent per-user memory that survives across sessions, ideally on every tier from free upward rather than being a premium upgrade. Sloane's memory architecture is documented in our memory guide.

Some products publish exactly what you get on each tier. Others deliberately keep limits vague and adjust them based on server load or business conditions. The vague pattern is where users hit walls they didn't know existed and can't plan around.

The specific complaint that recurs: subscribers who paid expecting "enough capacity for what I want to do" hit throttles during peak hours because the caps tighten under load. The paid tier turned out to be "higher but still variable" rather than "uncapped or predictable."

What clean looks like: documented usage limits (or documented uncapped tiers). Sloane's Free tier is 50 messages/day fixed — same at 8am Tuesday or 8pm Saturday. Plus at $9.99/month lifts the cap entirely. The number you signed up for is the number you get, regardless of when.

When a product's landing page leads with "no filter" or "uncensored" or "no restrictions" as its primary value proposition, it's signaling that its infrastructure is thin. Products built on the "we allow more" pitch historically correlate with a specific pattern: model swaps that break workflows, sudden platform takedowns when payment processors get uncomfortable, and users left stranded when the company loses its infrastructure.

This isn't about whether the product's content policy matches yours. It's about whether the product's only differentiator is "we're less filtered than the competition," because that's a shaky foundation for a subscription you're planning to build a relationship on.

What clean looks like: products that differentiate on their positive properties (character consistency, memory quality, dialogue polish, curated roster) rather than on what they allow the model to do. Sloane's pitch is dialogue quality + persistent memory + character consistency + curated roster — not what filters we don't have.

The AI companion category has significant churn. Products that appeared six months ago sometimes disappear in the next six. That's not necessarily a scam signal — it's a market signal — but it does mean a subscription to a brand-new anonymous product is a bet on continuity that's often not going to pay off.

What to look for as continuity signal: how long has the product been running (multiple years is a real signal), is there a named founder with a verifiable public presence, does the company communicate consistently (blog posts, product updates, incident notifications), and does the product feel like it's being actively developed vs. running on autopilot with a minimum crew.

What clean looks like: an established product with real operating history, communicated by real people, actively iterating on the product rather than treating it as a passive revenue stream.