AI girlfriend content policies compared, dollnerd.

This is a clinical comparison, non-explicit by house policy. The goal is to map what each app permits, not to demonstrate any of it. "No restrictions" and "uncensored" are the loudest claims in the category and the least reliable, because the homepage marketing routinely contradicts the terms of service three clicks away.

Genuinely permissive text roleplay does exist: SpicyChat, Chai, Perchance. But "no restrictions" is a marketing line, not a legal reality.

Several apps advertise flirtation while their own terms ban explicit content. Almost none allow explicit imagery. And every one of them checks for minors with self-attestation that regulators keep calling inadequate. Read the terms, not the landing page.

A polished AI girlfriend render. No restrictions is the loudest claim in the category and the least reliable, because the terms rarely match the marketing.

The policy table

App	Text roleplay	Explicit imagery	The catch
SpicyChat	Permissive (uncensored lane)	Weak image gen	Content lives on their servers under a real account
Chai	Historically very permissive	No	17+/Mature; reputation is lore, not stated policy
Perchance	No hard filter	User-prompted images	Responsibility placed entirely on the user
Character.AI	Blocked at generation ("Bob")	No	Under-18 open chat removed Oct 2025
Replika	Romantic mode, Pro only	No, "suggestive not pornographic"	New accounts never got legacy ERP back
EVA AI	Marketed flirty	No	Terms actually prohibit "sexually explicit" content
Romantic AI	Toggleable NSFW advertised	Photo gen advertised	Terms disclaim being any kind of service; heavy trackers
Grok "Ani"	Unlocked by level NSFW mode	Outfit change at level 5	Launched at a 12+ rating; self-attestation only

The marketing-versus-terms gap

This is the gap that matters. EVA AI markets itself as a flirty, romantic companion, yet its iOS terms and conditions prohibit "sexually explicit, pornographic, or adult-themed content." Romantic AI advertises toggleable NSFW and "spicy replies" on its homepage while its terms disclaim being any provider of anything and, per Mozilla's testing, fired 24,354 trackers in a minute.

The homepage sells one thing; the terms reserve the right to do another. When they conflict, the terms win, and the terms are what get enforced with a ban.

China AI rules remove companion features (@Dexerto, 2026-07-18)

China ordering companion features stripped out. Regulators, not companies, are setting the limits in this category.

Flirty companion marketing art. Apps like these sell romance on the homepage while their own terms quietly prohibit explicit content.

Where the filters actually bite

Where AI girlfriend apps fall on content moderation, from hard-filtered to permissive

Character.AI is the strict end: explicit content is blocked at generation, there has never been an official NSFW toggle, and the community's long-running war with the filter (nicknamed "Bob") produced petitions and a permanent jailbreak culture. The permissive end (SpicyChat, Chai, Perchance) allows adult text roleplay, but even there, explicit imagery is weak or user-generated, not a polished product. Grok's Ani sits in its own lane: an NSFW mode that unlocks around affection level 5.

Age checks

Every serious content-policy conversation runs into the same failure. Replika until February 2023 "had not implemented any age verification mechanisms," and Italy's regulator found the later system still "deficient in several respects," fining the company 5 million euros in 2025.

Grok's app launched at a 12+ App Store rating despite explicit content, later raised to 16+ after backlash, with age verification that is a self-attestation pop-up. Character.AI went the other direction and removed open-ended chat for under-18s entirely in late 2025.

An AI companion render marketed to adults. Every app in the category checks for minors with self-attestation that regulators keep calling inadequate.

If you have kids with phones, "self-attestation" is the phrase to worry about.

How to read any app's real policy

Open the terms, not the homepage. Search the terms for "explicit," "pornographic," and "18." The terms are the enforced document.
Assume "no restrictions" means "text, behind a paywall, on our servers." Truly unlimited is rare; explicit imagery is rarer still.
Remember it is their server. Permissive-but-hosted means your content sits under a real account subject to their moderation, which is the privacy angle in what these apps collect.

A companion render of the sort these apps advertise. Judge the app by its terms of service, not the picture on its landing page.

For the imagery side specifically, see AI girlfriend image tools compared; for the level-locked case, Grok's AI girlfriend, Ani; and the whole field is on the honest app map.

"Uncensored" is a banner. The terms of service are the enforced document. When the two disagree, the ban comes from the terms.