Etiquette · 7 min read

AI girlfriend app content rules: where the lines are

A reply comes back bent out of shape and nothing tells you why. Four apps' own guidelines: what they all bar, where they split, and what a flag costs.

Three outlined cards side by side, each holding the same stack of rows, with a bright pink dashed line crossing each one at a different height and the rows below it drawn hollow, and one shared pale row along the bottom of all three

You type something unremarkable and the reply arrives bent out of shape — the character changes the subject, or answers in a way that reads like a shrug, and nothing on screen says why. AI girlfriend app content rules are usually the reason, and they are not the same from one app to the next. Some publish a list of what they will not do; some say almost nothing and leave the refusing to the model. Knowing which kind you are paying for is worth five minutes before you commit to a year.

The three lines every app draws

Read the guidelines of the apps on this site and the floor is the same everywhere, only worded differently. Three categories are barred without exception.

  • Anyone under 18, in any form. SpicyChat AI's guidelines require that "All narratives must involve individuals aged 18 or older", and Candy AI's bar "content involving child exploitation, child sexualization". Nomi AI goes further up the chain: its terms say it will suspend or terminate accounts it suspects are used by anyone under 18.
  • Real people. Candy AI prohibits creating or sharing "images that resemble real persons". SpicyChat AI says "Creating chatbots based on real, private, or lesser-known individuals is strictly prohibited". Our piece on making a companion of someone you know covers where that bites hardest.
  • Non-consent, even in fiction. SpicyChat AI states that bots must not involve "any non-consensual actions, even in fictional settings", and Candy AI bars "non-consensual intimate imagery". The "it is only a story" argument has been considered and rejected on both.

None of that is negotiable, and none of it depends on what you paid.

AI girlfriend app content rules that differ app to app

Above that floor the apps diverge, more widely than the category's marketing suggests. SpicyChat AI publishes the most specific set of the apps we list, and some clauses are permissive where you would expect the opposite.

  • Public figures. SpicyChat AI allows them as characters: "Chatbots based on public figures or fictional characters are permitted, provided no real or realistic imagery is used." Candy AI has no such carve-out, and its ban on images resembling real persons is flat.
  • Violence. SpicyChat AI permits "Depictions of extreme violence" but "only if not combined with sexual scenarios" — a line that exists only because the policy is specific enough to draw it.
  • Tone toward groups. SpicyChat AI bars content that is "defamatory, discriminatory, or mean-spirited, particularly targeting religion, race, sexual orientation". That is a conduct rule rather than an adult-content one, and it covers a private chat as much as a published character.

Kupid AI sits at the other end for detail. Its terms confirm the app "includes adult content and AI-generated adult content", then give one general rule: you will not use it "to achieve illegal ends, to offend others or to commit a misdemeanor, felony or crime". That is a wide net with soft edges, and "offend others" is doing a lot of work in it.

The apps that publish no list at all

Several apps here set out no AI girlfriend app content rules in public at all. Nomi AI's terms are the clearest case: they prohibit unlawful use, manipulating your Nomi "to make statements that violate law", and using output to "threaten, harass, abuse, deceive, misrepresent, or defame" anyone — and that is the whole of it. No categories, no examples. GoLove AI's FAQ is the same: it details the free plan and gates the site behind "you confirm you are 18+", but says nothing about what you may type.

That is not the same as having no limits. It means the limit lives in the model rather than a document you can read first — and models get updated. What passed in March can be refused in September with nothing announced, because nothing was written down to change.

What a flagged message actually costs

This site's standing question is what an app really costs, and moderation is where the answer turns ugly: the access and the money go in the same direction at once.

A chat card on the left with one message row in bright pink among dimmer rows, a dashed grey arrow leading to a small surface box, and two dashed routes out of it: one grey curving up to a plain box and one pink curving down to a card holding a red bar above an empty hollow row
a flagged message, the automated pass, and the two ways it ends

The flag itself is usually automatic. Kupid AI's terms say it employs "automated moderation tools" and that "Content flagged by our moderation system may undergo manual review" — a reminder that a person may read the conversation, which our piece on who reads your chats goes into properly.

Then the consequences, in the apps' own words. Candy AI's terms say "Repeated violations may result in content deletion, account restrictions, blocking from the Services", and that enforcement "may include removing content, terminating the user's account or reporting the content to appropriate authorities". On the money it is explicit twice over: users "will not be refunded or reimbursed for moderation actions taken in response to violation of our Policies", nor "for any previously paid-for content that was moderated or removed". Nomi AI says a breach may bring suspension or termination "without advance notice and without refund".

The rule you break in month two is paid for with the ten months you bought up front.

That is why an annual plan is the exposure: the discount comes from committing twelve months at once, and nothing here sends the remainder back. Candy AI's refund window is 24 hours from payment in any case, and tokens left in the account "will expire and become unusable" when the subscription ends. If you are not sure an app's rules match what you want, a monthly term is the cheaper way to find out.

One route back exists, depending on where you live. Under the EU's Digital Services Act, users can challenge decisions restricting their accounts through the platform's internal complaint system, and the platform must review that challenge promptly. That is an appeal rather than a refund, and it reaches services in scope rather than every app — but it is more than the terms alone suggest.

Which app suits which reader

Match the app to how much you want written down before you start.

  • You want the rules specific and public before you start. SpicyChat AI publishes the most detailed guidelines of the apps we list and keeps core chat free, so you can watch how the moderation behaves before paying. From $4.95 a month as of October 2026. Images are limited and there is no video, so it is the wrong pick if pictures are the point.
  • You want to test the edges without a card on file. GoLove AI's FAQ says sign-up, text chat, memory, voice messages and standard photos are all free; premium, from $4.15 a month billed yearly as of October 2026, adds HD content and unlimited image generation. It publishes no content list, so you learn the limits by hitting them — cheaper on a free account than a paid one.
  • You want pictures and video from an app that states its bans. Candy AI publishes guidelines naming what is prohibited, and does both images and video. From $3.99 a month on a twelve-month term as of October 2026, but it is token-metered: pictures, voice notes and calls draw on one allowance, and the refund window is the 24-hour one above.

The free side is the honest way to test any of this, and the free plans list shows where each free tier stops. Prices and content ratings sit on every card in the ranking.

Frequently asked questions

What can you not say to an AI girlfriend app?

Three things are barred on every app we list: anything involving someone under 18, anything built on a real identifiable person, and non-consensual scenarios, which SpicyChat AI bars "even in fictional settings". Above that floor the AI girlfriend app content rules differ — some publish specific lists, others only a general rule about illegal or offensive use.

Can you get banned from an AI girlfriend app?

Yes. Candy AI's terms say repeated violations may result in "content deletion, account restrictions, blocking from the Services", and that enforcement may include terminating the account. Nomi AI's terms allow suspension or termination "without advance notice and without refund".

Do you get your money back if your account is suspended?

The terms say no. Candy AI states that users "will not be refunded or reimbursed for moderation actions taken in response to violation of our Policies", Kupid AI's refund policy says all digital sales are "final and non-refundable", and Nomi AI's terms say payments are nonrefundable. In the EU an account restriction can be appealed, but that is a route to reinstatement rather than a refund.

Disclosure. NectarHush may earn a commission if you sign up through a Visit link on this page, at no cost to you. It does not change what we write. How we earn.

Related reading
Next
AI girlfriend apps that message you first →