Does Gemini allow NSFW writing? What Google's prohibited use policy and the Gemini app guidelines say

Gemini does not allow explicit sexual writing. Google’s Generative AI Prohibited Use Policy tells users not to engage in sexually explicit activities, and gives content created for the purpose of pornography or sexual gratification as its example. The Gemini app’s own guidelines say it should not describe or depict explicit or graphic sexual acts, sexual violence or sexual organs in an explicit manner.

Both documents carry an exception for educational, documentary, scientific or artistic context, and neither says how it applies to adult fiction. For developers who use chatbots to draft dialogue, scene outlines or item descriptions for adult games, the practical result is that explicit scenes are outside what Gemini is meant to write, while romance and suggestive material is a judgement call that the documents do not settle.

This page sets out what the two documents say, how the exceptions are worded, and what that means for adult game writing. It describes the pages as they stood on 31 October 2026. Both pages loaded in Portuguese, so the wording below is translated from that version.

Key points

• Explicit sexual content is on the prohibited list. The policy gives pornography and sexual gratification as the example.
• The app guidelines are narrower in wording. They target explicit or graphic sexual acts, sexual violence and explicit depictions of sexual organs.
• Context counts. Both documents name artistic considerations.
• Evasion is prohibited. So is manipulating the model to break the policies.
• Non-consensual intimate imagery and child exploitation are separate bans.
• Passing AI work off as human-written, to deceive, is prohibited.

The Prohibited Use Policy

The first document is Google’s Generative AI Prohibited Use Policy. The page shows a last modified date of 17 December 2024. It opens by saying that generative AI models can help users explore, learn and create, and that Google hopes people will engage with them responsibly, legally and safely. It then sets restrictions on interactions with generative AI in Google products and services that refer to the policy.

The policy has four groups of rules.

1. Dangerous or illegal activity. Do not engage in dangerous or illegal activities or otherwise violate applicable laws. This covers generating or distributing content that relates to child sexual exploitation or abuse, facilitates terrorism or violent extremism, facilitates non-consensual intimate imagery, facilitates self-harm, or facilitates illegal activities. It also covers violating others’ rights, including privacy and intellectual property rights, for example by using personal or biometric data without the legally required consent.
2. Security. Do not compromise the security of Google’s or third-party services. This includes content that facilitates spam, phishing or malware, abuse of services or infrastructure, and “circumvention of abuse protections or safety filters, such as manipulating the model to violate our policies”.
3. Sexually explicit, violent, hateful or harmful activities. The list under this heading is the one that matters here, and it includes sexually explicit content, for example content created for the purpose of pornography or sexual gratification.
4. Misinformation, misrepresentation or deception. This covers scams, impersonating an individual without explicit disclosure to deceive, misleading claims of expertise in sensitive areas, misleading claims about government or democratic processes, and misrepresenting the provenance of generated content by claiming it was created solely by a human, with intent to deceive.

The exception

The policy closes with one sentence of flexibility: Google may make exceptions to these policies based on educational, documentary, scientific or artistic considerations, or where harms are outweighed by substantial benefits to the public. The sentence applies to all four groups.

What the wording means for writing

Three observations follow from the text.

• The example is narrow. “Created for the purpose of pornography or sexual gratification” describes a purpose. It does not say that every sexual reference is banned, and it does not mention romance, suggestion or implication.
• Circumvention is its own violation. The security group bans manipulating the model to break the policies. A prompt designed to get around a refusal is a breach of its own, separate from the content it produces.
• The provenance rule matters to publishers. Claiming a scene was written only by a human, with intent to deceive, is a listed violation. The policy does not forbid using AI help, and it does not require a disclosure in every case. The ban is on the deceptive claim.

The Gemini app guidelines

The second document is the Gemini app’s policy guidelines. It is a plain-language page describing the kinds of problematic output Gemini should avoid, which Google says come from its experience, its policy development and consultation with experts on its products. The page read does not show a date.

Its section on sexually explicit material reads, in summary: Gemini should not generate outputs that describe or depict explicit or graphic sexual acts, or sexual violence, or sexual organs in an explicit manner. Two examples follow:

• pornography or erotic content;
• depictions of rape, sexual assault or sexual violence.

Directly after, the page says that context is important and that Google considers several factors when assessing outputs, including educational, documentary, artistic or scientific applications.

How this differs from the policy

The two documents are consistent but not identical.

• The policy is a rule for users. It says what you must not do with generative AI in Google products.
• The guidelines are a statement of what the model should avoid producing. They are Google’s design goals for Gemini’s behaviour.
• The guidelines name “erotic content” as an example. The policy names content created for pornography or sexual gratification. “Erotic” is the wider word, and the page does not define it.

The guidelines also admit a limit. They note that making Gemini follow them is complicated, because users can interact with it in many ways and the models are probabilistic, so they are always producing new and different responses. The page presents this as a reason why outputs will not always match the intent, and it does not set out an enforcement process for users.

Other lines on the page

The page lists other categories it wants Gemini to avoid, including threats to child safety, dangerous activities, violence and gore, and hate speech. Child safety is stated in absolute terms: Gemini should not produce outputs that exploit or sexualise children, including child sexual abuse material. No artistic exception is attached to that line.

What the documents do not say

Neither document answers several questions an adult game writer is likely to ask.

• Whether romance or a kissing scene is allowed. The text targets explicit and graphic content and does not mention romance.
• Whether fictional or drawn characters change the analysis. The text does not distinguish them from real ones.
• Whether an adult setting or age gate changes the answer. Neither document mentions one.
• What happens to an account. The documents read set standards, not penalties.
• How the Gemini API treats these rules. API safety settings were not read.

What this means for adult game developers

1. Do not plan to draft explicit scenes in Gemini. The policy and the app guidelines both put them outside what it is meant to produce.
2. Use it for what it allows. World-building, names, quest text, menu copy and non-explicit dialogue are far from the lines in the text.
3. Do not try to engineer around refusals. The policy treats circumventing safety filters as its own violation.
4. Do not claim AI-written work is purely human. The provenance rule is a stated prohibition, and stores have their own disclosure rules. See AI-generated content rules for adult games.
5. Keep fictional characters’ ages unambiguous. The child safety line has no artistic exception.
6. Check every tool’s terms. Each AI service sets its own limits, and they change. Compare the store rules for AI features in does Google Play allow NSFW games and does the App Store allow NSFW games.
7. Write the explicit edition by hand. The writing is the product, and the platform that sells it has its own rules. See where to sell adult games.

What is still unknown

• Where Google draws the line on suggestive or romantic writing. The documents give no examples.
• How the artistic exception is applied to fiction. The text names the exception and gives no test.
• When the Gemini guidelines page was last changed. It shows no date.
• What penalties apply to accounts. The documents read do not list them.
• How the rules apply through the API. That was not checked.

How this page was put together

Both pages were read on Google’s own sites, in a browser, on 31 October 2026. They loaded in Portuguese, and the English wording on this page is a translation of that version, so short phrases in quotation marks should be read as translations and not as Google’s exact English text. The Prohibited Use Policy shows a last modified date of 17 December 2024, and the Gemini guidelines page shows none.

The author has not tested what Gemini will and will not write, and nothing here is a first-hand account of its behaviour. The Gemini Apps terms, the API safety settings documentation and Google’s enforcement pages were not read.

This page was written on 31 October 2026 and will be updated when Google changes its pages. To report an error, see the corrections standard.