AI Content Policy

Version and validity

• Version and effective date: 30 October 2026
• Last updated: 30 October 2026
• Issued by: CS Digital Wonders LLC, a limited liability company organized under the laws of the State of Wyoming, USA, with its registered address at 30 N Gould St Ste N, Sheridan, WY 82801, USA ("CS Digital Wonders", "we", "us"). SemPost is a product of CS Digital Wonders.
• Contact for this policy: [email protected]
• Languages: this policy is published in English and Turkish with the same content. The Terms of Service say which version prevails.
• How changes are announced: we update this policy when the product, the law or a network's rules change. Before a material change takes effect, we tell users by email, 30 days in advance, or with shorter notice when a law or a network's rule requires a faster change . Smaller changes (wording, links, corrections) are listed in the "Changes" section at the end. You can ask us for an earlier version at the contact address above.

1. Scope and key terms

This policy is part of the SemPost Terms of Service. It applies to everyone who uses SemPost: on app.sempost.ai, inside Shopify through the app "SemPost: AI Social Media Posts" (embed.sempost.ai), and to the content you publish through SemPost. It sits next to the Acceptable Use Policy, the Copyright and Notice-and-Takedown Policy and the Platform Terms.

It explains what SemPost makes with AI, how SemPost marks it, what the law and the networks require, what you must do, how SemPost handles people and voices, who owns the results, whether your content trains AI models, how music works, and how to report misuse.

Key terms used in this policy:

• You: the person who uses SemPost and the workspace (the business or person) that person acts for. A workspace owner is responsible for how the members of the workspace use SemPost.
• AI content: text, pictures, video or sound that an AI model created or changed in SemPost, and files you mark as "Made with AI" when you upload them.
• Network: a social network or service that SemPost publishes to: Instagram, Facebook, Threads, TikTok, Pinterest, LinkedIn, YouTube, Google Business Profile and Bluesky.
• Content Credentials: a signed record attached to a file under the C2PA standard (Coalition for Content Provenance and Authenticity) that says how the file was made.
• Network AI label: a setting that a network offers to mark a post as made or changed with AI. SemPost sets it through its social gateway (see section 4.3).
• Synthetic person: a person in a picture or video who was made up by an AI model and does not exist.
• Deep fake: the definition in the EU AI Act, Article 3(60): AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful.

This policy explains our rules and summarises laws that often apply. It is not legal advice for your business. You decide what to publish and you are responsible for it.

2. What SemPost makes with AI

SemPost uses AI only for the features in the table below. Each AI step runs when you ask for it, or when an automation rule that you switched on prepares posts.

These parts of SemPost use no AI: slideshow videos, still shots and drafts in the Video Studio, collages, labels, badges and text drawn on pictures, your logo, the end card, and the cropping of pictures for each network.

What we send to AI providers. To produce a result, SemPost sends the provider what the task needs: your product data (for example name, description and price), product photos, your brand voice and banned words, your instructions, your voiceover text, and for ambient sound a small, silent copy of the video without the logo. Pictures go to video and sound providers through temporary links that expire after one hour.

Which providers. SemPost's administrators decide which providers and models are available. You choose among them in Settings › AI models. The providers in use are Anthropic (Claude), Google (the paid tier of the Gemini API), fal.ai (video clips, ambient sound and, as a fallback, images) and Stability AI (only for the SemPost Music library). The providers that receive data, and where they process it, are listed on the Sub-processors page.

3. How we instruct the models, and their limits

• Product kept the same. For scenes and clips we tell the model to keep the product exactly as it is in your photos: the same shape, colours, materials, proportions, label and details, and not to add logos, text or other products.
• No invented facts. For text we tell the model not to state anything that is not in your product data: no invented materials, sizes, origins, certifications, awards, prices, discounts or delivery promises, and not to use your banned words.
• Clip checks. Before a motion clip is charged, SemPost checks automatically that it starts from your picture and, for most camera moves, that the product is still in the last frame. A clip that fails is not charged and is made again once. This is a safety net, not a quality guarantee.

AI models still make mistakes. A model can change the product (its colour, shape, size or the text on its label), add people or objects, write claims that are not true, or produce something that looks like another person's work. Results are not unique: other users can receive similar results. That is why you must review every result before it is published (section 6).

4. How SemPost marks AI content

4.1 Which files count as AI content

SemPost records for every file in your media library whether AI made it:

• AI generated: a model's own output, for example a scene picture or a motion clip.
• With AI parts (AI composite): a file that contains an AI-made part, for example a Studio video that opens with a motion clip, a video with a synthetic voiceover or AI ambient sound, a picture or video made from an AI picture, a product video that contains a scene or a clip, and the TikTok version of such a video. A video that carries a SemPost Music track made with an AI model is also AI composite.
• Made with AI (uploads): a file you upload counts as AI content when you mark it as "Made with AI" while uploading it. SemPost cannot tell by itself whether a file was made or changed with AI outside SemPost, so mark such files when you upload them.

The file card in the media library shows "AI generated", "with AI parts" or "made with AI (marked)" and says that posts carrying the file are labelled on the networks.

Text. Post text written by the AI writer is not marked in a machine-readable way, and SemPost sets no network label for it. Where a law or a network requires a disclosure for AI-written text, add it yourself (section 6.2).

4.2 Content Credentials (C2PA)

• What is signed. SemPost attaches Content Credentials to AI generated and AI composite files in JPEG, PNG or MP4 format, and to files you marked as "Made with AI" when you uploaded them.
• What the record says. The record says that SemPost made the file, gives the IPTC digital source type ("trainedAlgorithmicMedia" for a model's own output, "compositeWithTrainedAlgorithmicMedia" for a file with an AI part), can name the AI models used, and carries the file's title. For a file you marked as "Made with AI", the record says that it was marked when it was uploaded and gives "compositeWithTrainedAlgorithmicMedia", because SemPost cannot tell how much of it AI made; it names no model. SemPost does not add your name, email address or account details, and does not embed a thumbnail.
• When. When the file is kept in your library, and for pictures again when they are published, because the post's shape, text and logo are drawn on the picture at that moment. A video kept before signing started, and a video you marked as "Made with AI", is signed when it is published.
• Who signs. A signing certificate issued from SemPost's own root certificate. SemPost is not on the C2PA trust list. Readers show the record as valid but may show the signer as unknown or not trusted. Anyone can inspect a file at contentcredentials.org/verify.
• If signing fails. The work does not stop: the file is published without the record, we log the failure, and the network AI label is still sent.
• After publishing. Networks may keep or remove this data when they process an upload. SemPost does not control that. YouTube says it may add an AI label itself when a file carries C2PA data, and that such a label cannot be removed by the creator.
• Providers' own marks. Some providers embed their own invisible watermark, for example Google's SynthID in Google models' output. SemPost does not remove these on purpose, but normal editing (cropping, resizing, re-encoding, adding text, a logo or an end card, mixing voice and music) can weaken them. When SemPost re-encodes a file, the provider's own Content Credentials record is not carried over; SemPost's record takes its place.
• What SemPost does not add. SemPost does not add an invisible watermark of its own, and does not draw a visible "AI" notice on pictures or videos. Where a visible notice is needed, add it yourself (section 6.2).
• How long. SemPost keeps no separate record of how a file was made. The Content Credentials inside a file and our generation records are deleted together with your workspace, like the rest of its content.

4.3 Network AI labels

When a post carries a picture or a video with AI content, including pictures added to a collage, SemPost turns on the network's AI label when it publishes the post, wherever the network and our gateway support it. This applies to posts made in Generate from products, in the Studio and to posts you write yourself. It also applies to videos with a SemPost Music track made with an AI model and to files you marked as "Made with AI" (section 4.1). You cannot switch this label off in SemPost.

Text written by the AI writer is not labelled on any network.

5. The law on AI disclosure (summary)

5.1 EU AI Act, Article 50

Article 50 of the EU AI Act (Regulation (EU) 2024/1689) has applied since 2 August 2026.

• Providers (Article 50(2)). A provider of an AI system that generates synthetic audio, image, video or text must make sure the output is marked in a machine-readable format and can be detected as AI-generated or manipulated. This does not apply where the system only assists with standard editing or does not substantially change the input.
• Deployers (Article 50(4)). A business that uses an AI system (a "deployer") must disclose that a deep fake image, audio or video was generated or manipulated. For evidently artistic, creative, satirical or fictional work, the duty is limited to disclosing that such content exists, in a way that does not spoil the work. AI-generated text published to inform the public on matters of public interest must also be disclosed, unless a person reviewed it and someone holds editorial responsibility for it. Personal, non-professional use is outside the definition of a deployer.
• How (Article 50(5)). The information must be clear and distinguishable, given at the latest when people first see or hear the content, and accessible.
• SemPost's role. SemPost acts as the provider for machine-readable marking (section 4), and you act as the deployer for disclosures to your audience (section 6).
• Transition period. Regulation (EU) 2026/1744 gives extra time until 2 December 2026 only for the marking duty and only for systems placed on the market before 2 August 2026. SemPost does not rely on it.
• Guidance. The European Commission has published draft guidelines on Article 50 and a final code of practice on marking and labelling AI-generated content.

5.2 Turkey: Commercial Advertising Regulation (in force since 1 August 2026)

• Article 18(8). If an advertisement uses AI or software in a way that can significantly affect consumers' economic decisions, or shows digital characters that cannot be told apart from humans, this must be stated in a clear, understandable and distinguishable way.
• Article 27(12). Advertisements must not give the false impression that an AI-made digital copy of a real person experiences, uses or recommends a product.
• Who is responsible. Under Turkish consumer law, advertisers must be able to prove the claims in their advertisements (Law No. 6502, Article 61). As the advertiser, you add a visible notice where SemPost cannot set a network label (section 6.2).

5.3 The networks' own rules

• Meta (Facebook, Instagram, Threads): you must use Meta's AI disclosure for organic posts with photorealistic video or realistic-sounding audio that was digitally created or altered. Meta may add its own label, and may apply penalties if you do not disclose. Meta Community Standards
• TikTok: you must label AI-generated or significantly edited content that shows realistic scenes or people. TikTok may remove, restrict or label unlabelled content. TikTok Community Guidelines: Integrity and Authenticity
• YouTube: you must disclose realistic altered or synthetic content that viewers could mistake for a real person, place or event. Minor edits do not need it. YouTube may add a label itself, for example when a file carries C2PA data. YouTube help: disclosing altered or synthetic content
• LinkedIn: synthetic or manipulated media that shows people saying or doing things they did not do must be clearly disclosed. LinkedIn Professional Community Policies
• Pinterest: its community guidelines apply to synthetically generated content, and its misinformation rules cover fabricated or manipulated media. Pinterest Community Guidelines

5.4 Other countries

Other countries and US states have their own rules on AI disclosure in advertising and on using a person's likeness or voice. The rules of the places where you sell and where your audience lives apply to you.

6. What you must do

6.1 Review before you publish

• Check every AI result before it is published: the product's shape, colour, size and label, every fact and claim in the text, prices and discount codes, people in the picture, spelling and the network's rules.
• SemPost's default approval setting is "Only content made by AI and automation waits for approval". If you choose "No approval", AI content is scheduled without a person looking at it, and you remain responsible for it.
• Keep your original product photos and product information, so that you can prove your claims if asked.
• Do not publish a result that shows the product differently from what a buyer will receive.

6.2 Disclose AI content where required

• Where a law or a network requires a disclosure and SemPost cannot set a network label (Facebook, Threads, LinkedIn, Bluesky, Google Business Profile, and TikTok drafts until you add the label in the app), add a clear notice yourself, for example "Created with AI" in the post text or on the picture.
• SemPost never labels text written with AI. Where a law or a network requires a disclosure for such text, add it yourself.
• Give the notice at the latest when people first see or hear the content, and make it easy to notice and to read.
• If you promote a post as a paid ad, or reuse a file in an ad, the ad platform's own AI disclosure settings and rules also apply.
• A wording you can use: "This image was created with AI." In Turkish: "Bu görsel yapay zekâ ile oluşturuldu."

6.3 Synthetic people and voices

• Do not present a synthetic person as a real customer, employee, expert or reviewer.
• Do not make a synthetic person or a synthetic voice say or show that they recommend, review or rate your product. Fake reviews and testimonials are also forbidden by the Acceptable Use Policy.
• Disclose synthetic people and synthetic voices where the law or a network requires it.
• If a synthetic person happens to look like a real, identifiable person, do not publish the result.

6.4 No deep fakes of real people

• Do not use SemPost to create or edit pictures, video or audio that show or imitate a real, identifiable person, including celebrities, influencers, your customers or your staff, in a way that people could take as real. The only exception: the person has given prior written consent and you disclose it clearly; even then, never as an endorsement.
• Never show an AI copy of a real person as using, experiencing or recommending a product.
• Do not upload photos of people you have no permission to use.

6.5 Keep labels and Content Credentials

• Do not remove, alter or hide Content Credentials, providers' watermarks or network AI labels, and do not re-encode or crop a file to get rid of them.
• When you download a video from the Video Studio (its "Download" button gives you the file as SemPost made it) and publish it elsewhere, keep its record and add the disclosures that apply there.
• Do not switch off an AI disclosure in a network's app after SemPost has set it.

6.6 Follow the networks' rules

Each network's terms and policies apply to what you publish there. The Platform Terms page lists them.

7. People, hands and voices: what SemPost does by default

• Scenes in the Studio and the Video Studio. The "People" choice has four options:

  • None (default): the model is told that there are no people in the picture.
  • Hands only: hands and forearms hold or use the product; no face.
  • Person using the product: one adult, seen from the side or slightly from behind, with the face mostly turned away, not posing and not looking at the camera.
  • Worn or held: the picture is framed on the product (a hand, wrist, neck or body); the face is out of the frame or turned away.

  The page tells you that the person is made by AI and must not be presented as recommending or reviewing the product.

• Generate from products. The setting of each scene is written by the language model in one sentence. Scenes contain no people by default: the model is told that there are no people in the picture. People appear only when you choose them yourself in the Studio or the Video Studio.

• Motion clips in Generate from products and the Studio. The model is told to keep the product's exact shape, colours, materials and proportions, to move only what is already in the photograph, and to show only what the photograph shows.

• Scene actions in the Video Studio. You describe in your own words what happens in a scene. The suggestions include hands and a person. The model animates what you describe.

• Voices. A voiceover is added only when you write the text and choose a voice. SemPost offers preset synthetic voices only. There is no voice cloning.

• No avatars. SemPost does not offer avatars, talking presenters, lip-sync or face swapping.

• Labels. A video with a synthetic voiceover or AI ambient sound is AI composite content: it carries Content Credentials and the network AI label (section 4). So is a video with a SemPost Music track made with an AI model.

8. Ownership and copyright of results

• Between you and us. We assign to you whatever rights we have in the results made for your workspace. We keep only the permission we need to run the service for you (to store, process and publish), as described in the Terms of Service.
• Copyright may not protect AI results. The US Copyright Office's report "Copyright and Artificial Intelligence, Part 2: Copyrightability" (January 2025) says that prompts alone do not make you the author of an AI output. Your own photos and texts, your creative selection and arrangement, and meaningful changes you make yourself can be protected. Under Turkish law (Law No. 5846, FSEK) a work must carry the personal characteristics of a human author; whether AI results are protected is unclear. Other countries have their own rules.
• Results are not unique. Other users can receive similar results, and you may not be able to stop them from using similar content.
• Provider terms. Who owns an output, and whether attribution, a mark or a watermark is required, can differ from one model to another under the provider's terms. Limits in those terms also apply to you.
• Your inputs. You must have the rights to everything you give SemPost: photos, logos, product texts, page addresses you import, voiceover texts and instructions.
• Other people's rights in results. A result can resemble another person's work, trademark or appearance. Check before you publish, and do not publish such results. The Copyright and Notice-and-Takedown Policy explains how rights holders can complain.

9. Your content and AI training

• SemPost does not use your content to train or fine-tune any AI model. This covers your product data, photos, posts, brand settings, voiceover texts, analytics and media.
• AI providers process the data we send them to produce the result you asked for, under their own terms. Under Anthropic's commercial terms and Google's terms for the paid Gemini API, they may not use this data to train their models. We use the Gemini API on the paid tier with storage of requests switched off; Google keeps logs of requests for a limited time, only to detect misuse and to meet legal requirements. fal.ai and the models it runs, such as Mirelo's, process the data under fal's terms for business customers. Stability AI receives none of your content.
• When you use "Write in the style of my best posts", your posts are sent as examples inside that one request. They are not used to train a model.

10. Music

10.1 The networks' music libraries

• Instagram Reels: in the Studio you can add a track from Instagram's own music library to a Reel. Only accounts connected through Facebook can do this. SemPost does not copy or store the music: it sends only the track's ID, and Instagram adds the sound. Meta shows only tracks that it allows for third-party use.
• TikTok: when you publish directly from the Studio, you can choose a track from TikTok's Commercial Music Library. When a TikTok post goes as a draft, you choose music in the TikTok app.
• Facebook Reels: music cannot be added through the API.
• Licence. Music from a network's library is licensed between that network and the account holder, under the network's terms. The network decides which tracks are available and may remove or mute them. If a track is no longer available when the post goes out, the post fails and SemPost tells you.
• Ads. A track from a network's library may not be cleared for paid ads. Check the network's rules before you promote such a post as an ad.

10.2 SemPost Music

• What it is. A library of background tracks made by SemPost, either with Stability AI's Stable Audio (whose terms assign the output to us, non-exclusively) or with SemPost's own rule-based composer, which uses no AI and plays freely licensed (CC0) instrument recordings. SemPost staff review every track before it enters the library. For every track we keep a record of how it was made and copies of the source's licence documents.
• AI label. A video that carries a track made with Stable Audio counts as AI composite content: it carries Content Credentials and the network AI label (section 4). A track from the rule-based composer does not make a video AI content.
• Where you can use it. In the Studio's "Make video" and in the Video Studio. Videos made with Generate from products have no music today.
• Licence. You may use a SemPost Music track only as part of a video made in SemPost and published on networks where the library is available; videos you published while you were a customer stay licensed after your subscription ends. You may not take the track out of the video, use it on its own, register it with Content ID or any other fingerprinting service, or claim it as your own.
• Availability. We can switch the library on or off by region, network and workspace. A video that carries a library track can be published only to networks where the library is available.
• Records. SemPost records which video used which track. If a track is withdrawn, new videos cannot use it, and we can find every video that used it (see the Copyright and Notice-and-Takedown Policy).

10.3 Your own music

You cannot upload your own music to SemPost. This feature is switched off. If a video you upload yourself already contains music, you must hold the rights to it.

11. Reporting misuse

If you see AI content made with SemPost that breaks this policy (for example a deep fake of a real person, a fake review or a removed AI label), report it to [email protected]. The Acceptable Use Policy explains what to include and how we handle reports. Copyright and trademark complaints follow the Copyright and Notice-and-Takedown Policy. Security issues go to [email protected] (see the Security page).

Content that is already published on a network can only be removed from that network by the account holder or by the network. You can also report it to the network directly.

Changes