What each platform does about AI

What gets AI content removed on X?

X does not ban AI content. Its authenticity policy goes after deception:

• Fake personas: AI-generated profile photos and made-up identities used to deceive are prohibited.
• Unauthorized automation: bot accounts that do not follow the developer policy.
• Impersonation: posing as a person or brand you are not.

Can you stop AI models training on your newsletter?

Partly. Substack has a setting under Settings > Privacy: "Tell AI tools not to train their models on your content." Substack says it only applies to tools that respect the signal, and warns that blocking training may reduce how often you show up in AI-generated results.

Can you clone a voice for a campaign?

Only with the rights to that voice, and the terms take something back:

• ElevenLabs prohibits cloning a person’s voice without their consent or a legal right to it.
• Its terms take a license to use your voice models to run and improve the service and to develop new products.
• Voice cloning is also where US state likeness laws and the FCC’s robocall rules bite, so consent from the speaker is not the only permission you need.

Runway

Can you use Runway video in client work?

Yes, per Runway’s own help center: you keep ownership of what you upload and generate, and there are no non-commercial restrictions from Runway. That covers monetized YouTube, Reels and TikTok, social posts, festival entries, and product advertising.

Is Sora still available?

No. OpenAI has discontinued it, and there is a deadline that matters if you made anything there:

• The Sora web and app experiences were discontinued on April 26, 2026.
• The Sora API shuts down on September 24, 2026.
• Export what you made at sora.chatgpt.com/sunset. OpenAI says data tied to Sora will be permanently deleted after any final export window.

Does Midjourney train on your prompts and images?

Its terms grant Midjourney a perpetual, worldwide, irrevocable, royalty-free license to reproduce, adapt, display, and distribute the content you put in and the assets you make. That is a broad grant. Read it before you put a client’s unreleased product into a prompt.

Adobe Firefly

Does Adobe train on your work?

Adobe’s AI ethics page states: "We do not and have never trained Adobe Firefly on customer content." Adobe says it trains on licensed content such as Adobe Stock and public domain material, and does not scrape the web.

Canva

Does Canva train AI on your designs?

It can, and it is a setting you control:

• Canva’s privacy policy says it may analyze your activity, content, and uploads to train its models and AI products.
• You can change that in the privacy controls under your account’s privacy settings.
• Canva says it does not use Canva Education user content for AI training.

Does OpenAI train on what you put in?

OpenAI’s terms say it may use content to provide, maintain, develop, and improve its services. Consumer accounts have a setting to opt out of model training, and business and API terms are written differently. Check the plan you are actually on before pasting client material.

Do you own what Claude writes?

Anthropic’s consumer terms say: "Subject to your compliance with our Terms, we assign to you all of our right, title, and interest, if any, in Outputs." As with every tool, being assigned those rights is not the same as the text being protected by copyright.

Can a human read your Gemini chats?

Google’s Gemini Apps Privacy Hub is the page that answers this: what is collected, how long it is kept, how to turn activity off, and when reviewers may see conversations. Read it before putting client data in a prompt.

Adobe Firefly

Does Firefly tag images with metadata?

Yes, automatically, when the whole image came from Firefly:

• Adobe attaches Content Credentials to assets where 100% of the pixels were generated by Firefly, such as text to image.
• The credential records the issuer (Adobe), the date, the app or device, the AI tool used, and the actions taken.
• Adobe calls the metadata tamper-evident, which means removal shows up rather than being impossible.

Canva

Do you own what Canva AI makes?

Mostly, with one catch worth knowing before a client hand-off:

• Canva’s AI terms say you keep your input and own your output.
• The exception is output that uses or modifies Canva library content, which stays under the license you were granted for that content.
• Your ownership does not extend to other users’ outputs.

Canva

Can you strip the AI tag off a Canva export?

Not under Canva’s rules. Its AI terms prohibit removing, altering, or disabling provenance or metadata tags on AI-generated content, C2PA included, and prohibit passing AI output off as human-made.

Does ChatGPT mark the images it makes?

Yes, twice over, and neither mark is visible:

• C2PA Content Credentials: metadata naming the tool and when the file was made. Platforms, editors, and file conversions can strip it.
• SynthID: an invisible watermark inside the image itself, which OpenAI says can survive some edits that remove metadata.
• Audio gets a SynthID watermark too. A visible watermark only appears if you ask the model for one.

Do you own what ChatGPT produces?

OpenAI’s terms say you keep your input and own the output, and OpenAI assigns you its rights in that output. Two catches: output is not always unique, so another user can get something similar, and owning a file is not the same as it being protected by copyright.

Does Anthropic train on what you type into Claude?

On consumer plans (Free, Pro, Max), Anthropic says your chats are used to improve models only when:

• You turn that setting on.
• A conversation is flagged for safety review.
• You opt in another way, such as a tester program.

Does Google watermark AI content from Gemini?

Google applies SynthID, an invisible watermark embedded in the content itself, to media from its generative models. It is designed to survive some edits, and Google publishes a detector for it. Google’s generative AI terms cover what you may do with the output.

Do you own Midjourney images at work?

Usually yes, with exceptions worth knowing before a client campaign:

• If your company makes over $1,000,000 a year, you need a Pro or Mega plan to own what you create.
• If you upscale someone else’s image, it stays theirs.
• Ownership under the terms is separate from whether copyright protects the image at all.

Does AI content hurt our rankings?

Not by itself. Google has said its focus is the quality of content, not how it was produced, and that using AI to help make helpful content is fine. What gets penalized is content made mainly to game rankings.

Does Google label AI content in results?

Google does not put an "AI" badge on ordinary search results. Its guidance points to disclosure where it helps readers, such as saying who or what produced a page, rather than a required label.

Is fully AI-written content allowed?

Yes, with a limit:

• Allowed: AI-assisted content that is useful and original.
• Spam: content produced at scale mainly to manipulate rankings, whether a person or a machine made it.

Do we have to say a video used AI?

Yes, when it looks real. YouTube asks creators to disclose realistic altered or synthetic content when uploading. Ordinary production help, like beauty filters or lighting fixes, does not need it.

Does YouTube label the video for viewers?

It can. When you disclose, YouTube shows a label in the description, and a more prominent one on sensitive topics. YouTube can also add the label itself if it decides viewers need it.

Does TikTok require an AI label?

Yes for realistic AI content. TikTok requires creators to label realistic AI-generated content, and it also reads provenance metadata to label some uploads automatically.

Is fully AI content allowed on TikTok?

Allowed, with limits. It has to be labeled when realistic, and the community guidelines still bar synthetic media that misleads about real people or events.

Does Meta label AI images on Facebook and Instagram?

Yes. Meta applies AI labels to images it detects as AI-generated across Facebook, Instagram, and Threads, using industry provenance signals attached to the file.

Will AI posts get us less reach on Meta?

Meta has not published a rule that AI content is downranked for being AI. The label is about provenance. Reach still comes down to the usual signals and the content rules everyone follows.

Does LinkedIn allow AI-written posts?

Yes. LinkedIn does not ban AI-assisted posts. Its policies target inauthentic behavior, misleading content, and spam, so the risk is volume and deception, not the tool.

How does Pinterest treat AI content?

Pinterest expects Pins to be authentic and not misleading, and it has been adding provenance labels to generated images. Its community guidelines are the rule that applies.