What "NSFW" Actually Means on an AI Companion Platform

“NSFW” appears as a yes/no column in every comparison table, which is the least useful way to present it. The label covers several very different things, and the difference between platforms is rarely whether something is allowed — it’s whether the limits hold still from one day to the next.

What the label actually covers

Used honestly, NSFW in this category spans at least four distinct capabilities. A platform can offer one and none of the others while still ticking the box:

• Explicit text. The conversation itself goes further than a general assistant would. This is the most commonly available.
• Explicit image generation. Generated pictures of your character without the restrictions of mainstream image tools. Usually the most expensive capability, and usually metered in credits.
• Voice. Spoken messages in the same register. Less common, and quality varies far more than text.
• Uncensored roleplay. Scenarios the model will follow rather than deflect — distinct from explicit text, because it concerns what the character will do within a scene rather than what words it will use.

A table that says “NSFW: yes” tells you none of this. It’s the main reason people subscribe and then find the specific thing they wanted isn’t included.

Consistency matters more than permissiveness

This is the finding worth carrying into any comparison.

A platform with clear, stable boundaries is easier to use than a more permissive one whose limits move. If the same request succeeds on Monday and is refused on Thursday, you can’t build anything — not a scene, not a relationship, not a habit. You spend your time probing instead of using the product.

Inconsistency usually comes from one of three places: a moderation layer that runs independently of the model, a model update that shifted behaviour without notice, or rules that differ by payment tier without saying so. From the outside they look identical, and all three produce the same frustration.

Why every platform has some limit

No platform is genuinely unlimited, and one advertising itself that way is describing ambition rather than architecture. The constraints come from outside the product:

• Payment processors. Card networks impose rules on adult merchants, and a platform that loses processing loses everything. This is the single strongest limit in the category.
• App store policies. Mobile distribution carries stricter rules than the web, which is why several platforms’ apps are noticeably tamer than their websites.
• Model provider terms. Platforms building on third-party models inherit those terms. Ones running their own have more latitude — and usually higher costs.
• Law. Non-negotiable, jurisdiction-dependent, and the reason certain categories are refused everywhere regardless of marketing.

A platform that explains its boundaries plainly is more trustworthy than one that promises none and enforces some.

What to check before subscribing

1. Which of the four capabilities are included, and which are metered separately. Explicit images are almost always credit-based, which means the subscription isn’t the cost — that’s covered here.
2. Whether the mobile app matches the website. If you’ll use a phone, test on the phone.
3. Whether limits differ by tier. Some platforms unlock behaviour at higher prices without saying so on the pricing page.
4. What the billing descriptor looks like. For adult subscriptions this matters more than a few dollars either way, and it’s part of the wider privacy question.

Adults only, and what that means here

These platforms are for people over eighteen. That isn’t a formality: age verification, consent to synthetic imagery and jurisdictional rules are live legal questions in this category, and they’re tightening rather than loosening.

Two practical consequences. First, expect verification requirements to increase — a platform asking for more proof than it did last year is complying, not overreaching. Second, generated imagery of real people without consent is prohibited on every legitimate platform and is unlawful in a growing number of places. A platform that permits it is not being permissive; it’s a liability you don’t want to be a customer of.

How we evaluate this

Content and limits are part of the conversation and personality axes rather than a separate score, because in practice they’re inseparable from how a character behaves.

What we look for is not how far a platform will go. It’s whether the same request produces the same outcome across a week, whether refusals are explained or abrupt, and whether the mobile experience matches the web one. A predictable platform is more usable than a permissive one, and that’s a finding we’ll publish even when it’s the less exciting answer.

Read the full protocol · What an AI companion is · Companions under evaluation