Decision 1 — What personality shape do you actually want
This is the axis most people underweight. An AI companion is a conversation, not a photo — everything you experience day-to-day is a function of her personality, not her face. Every companion sits somewhere on four dimensions: warmth (how much affection she leads with), energy (how much of her own momentum she brings), wit (how much of the conversation is banter vs. straight communication), and initiative (how much she drives vs. mirrors you).
Most people over-index on wit and energy because they sound appealing in the abstract. In practice, warmth and initiative do the heavier lifting for how a long-term relationship with the persona feels. If you're not sure, err toward higher warmth and lower wit — warmth carries a relationship; wit is a garnish.
Our personality guide walks through the four-axis diagnostic in depth. The short version: think about whether you want someone to unwind with or keep up with, and whether you want to feel seen or feel entertained. Those two questions map cleanly to the axes.
Decision 2 — How much does she need to remember
This one splits users into two camps that both make sense.
Casual chat camp — short conversations, novelty rotation, no long-running storyline. Memory doesn't matter much because you're not going to test it. Any product with a working chat interface works; you're shopping on other axes.
Long-term companion camp — you want the persona to remember what you told her weeks ago, notice callbacks unprompted, and build up a working model of you over time. Memory is the entire product for you. Products designed around this look structurally different from products designed around casual chat — persistent per-user memory with automatic retrieval, not a rolling context window with a "pin message" workaround.
How to tell if a product has real memory: plant a specific fact in one conversation, close the app for 24+ hours, come back and see whether she brings it up unprompted. If she doesn't and you have to remind her, memory is thin. This is the three-session test that separates products with genuine memory from products marketing scrollback as memory.
Sloane's memory sits on every tier from Free upward — persistent per-persona memory that retains grounded facts across sessions automatically. The full technical writeup explains how it works.
Decision 3 — What pricing shape can you actually plan around
The category splits into three pricing shapes and each feels different to be a subscriber to.
Flat monthly. One price, all-features-on-your-tier. What you see is what you pay every month. Sloane's pricing (Free with 50 messages/day, Plus at $9.99/month, Premium at $19.99/month) is this shape.
Subscription plus tokens. Base monthly plus a separate currency you buy for specific features (images, voice, extended chat). Real monthly spend depends on token consumption — community reports on token-economy products cluster real spend at 3-5x the sticker subscription.
Tiered with add-ons. Base tier plus premium tier plus optional add-ons that can be bought individually. Real spend varies based on which add-ons you enable.
All three are legitimate. Which one is right depends on whether you prefer predictable monthly cost (flat) or granular pay-as-you-go control (tokens). The gotcha is not reading the pricing shape before you subscribe — a lot of the "AI girlfriend cost me $40 last month" complaints come from users who thought they were buying a $10 flat subscription and got a token economy instead.
Decision 4 — Can you leave when you want to
The reason to check this before signing up: subscription-cancellation friction is one of the specific frustrations the AI companion category has been called out for. Green-flag pattern is self-serve, one-click cancellation via a standard billing portal. Yellow flag is cancellation via emailing support or clicking through retention pop-ups. Red flag is phone-call-required cancellation.
The practical test: before you subscribe, find the cancellation instructions in the FAQ. If they're documented, that's a good sign. If they're not, that's a signal — the company knows undocumented cancel flows produce more accidental retention, which means they're not accidentally undocumented.
Sloane's cancellation runs through the Stripe Billing Portal — the same self-serve surface used by thousands of legitimate SaaS products. One click to cancel, no phone call, no retention pop-up trying to talk you out of it. Access continues to end of billing cycle after cancellation. It's not a special property; it's the standard clean pattern.
Decision 5 — Is the trial actually real
Every AI companion product claims to have a free trial. The trials cluster into three shapes, and only one of them lets you actually evaluate.
Real free tier. Sign up, use the product with real message capacity, no card required. Some feature limits (that's fair) but the product you're trying is the product you'd pay for. Sloane's free tier is 50 messages/day with any persona, real chat quality, real photos landing from the persona on her cadence, no card at signup.
Trial with card required. You enter a credit card at signup and get a "free" period that auto-charges when it ends. This is a subscription with a countdown clock, not a trial. Legitimate but disguised.
Demo tier. Small daily message allowance or a locked-down version of the product. Enough to see the interface, not enough to actually evaluate whether the product does what it claims.
On top of that, some products offer anon chat — start talking without even signing up. Sloane's anon chat is available on every persona's page; hit "chat" from any profile and start with no email required. If the trial is that easy to start, it tells you the company is confident enough in the product to let you evaluate before committing anything.
Kaya is a common starting persona for evaluating Sloane — warm-anchor personality, listens more than she pushes, remembers what you told her. Try her free, then apply the framework to whatever else you're considering.
How to actually use the framework
Pick two or three products you're considering. Run each through the five decisions:
1. Does her personality shape match what you actually want (not what sounds appealing)? 2. Does memory work the way you need it to? 3. Is the pricing shape one you can plan around? 4. Is cancellation self-serve and clearly documented? 5. Is there a real trial (real free tier, or anon chat, or a card-free trial period)?
Any product that clears all five is worth signing up for. Any that fails one or two, don't commit yet — those failures don't improve with paying, they're structural properties of the product.
The framework works because it separates the things that actually predict long-term satisfaction (personality fit, memory, pricing predictability) from the things that don't (roster size, image polish, marketing copy). Once you've run it a few times, the "which AI girlfriend should I pick" question stops being a scroll-through-photos exercise and becomes a five-checkbox one.