AI and copyright for hotels: who owns generated text and images, and what you may publish

AI copyright determines who holds rights to text, images or music produced by a machine. Turkish Law No. 5846 protects works carrying their author's characteristic and names the creator as author. Its text says nothing about AI, and no Turkish court decision or official guidance on AI output was found. The answer depends on four specific decisions, not just the law.

A marketing manager generates a lobby image on Tuesday, publishes it on Friday, and three weeks later the agency asks who owns it. Nobody in the room can answer, because neither the tool's terms nor the contract was ever read that closely. After reading this article you will know which four decisions actually sit behind the question, what the law says and does not say, and what to record before the next asset goes live.

What does a hotel actually need to decide about AI copyright?#

Four decisions, not one: whether you may publish the output, whether you can stop a competitor copying it, what happens if it resembles someone else's work, and whether AI-made music changes your licence position.

Publishing is a commercial-use question, and the first document that answers it is the tool's terms of service, not the law. Stopping a competitor from reusing your image depends on whether the output counts as a work at all, and that is the open question. Resemblance is the risk that actually arrives, usually as a letter from a photographer or a brand, not as a court case about authorship. Music is a licence question with a new counterparty: the tool that generated the track, plus whoever holds rights in anything the track resembles.

Three layers answer different parts of the problem. The law, Law No. 5846, sets what a work and an author are. The tool's contract sets what you may do with the output. Your contract with the agency or freelancer who delivered the asset sets what happens between you and them.

This article describes the law in general terms. It is not legal advice; confirm your own case with a lawyer before you publish or sign.

What does Law No. 5846 say about a work and an author?#

It defines a work as an intellectual or artistic product that bears its owner's characteristic, and the author as the person who creates it. The consolidated text contains no provision on artificial intelligence.

Article 1/B, added by Law No. 4630 in 2001, sets the vocabulary: work, author, adaptation and compilation, and places a work in one of four groups, science and literature, music, fine arts or cinema. The characteristic test is the hinge. The product must carry something that comes from its owner, not merely be produced.

Article 8 repeats that the author is the person who creates the work, and adds rules for adaptations and compilations, where the adapter or compiler is author provided the original author's rights are preserved. For cinematographic works the law names co-authors: the director, the composer of the original music, the scriptwriter and the dialogue writer, and in animated works the animator as well.

A footnote in the consolidated text records that Article 28 of Law No. 5101 (2004) deleted the word "gerçek" (real, natural) from the author definition. That removed a word; it did not create machine authorship.

Adaptation matters commercially. If you feed an existing photograph or a competitor's room description into a tool and rework it, the original author's rights survive the rework.

Source: the consolidated Law No. 5846 on mevzuat.gov.tr (opens in a new tab), Articles 1/B and 8 (7 November 2026).

Why is AI output debated, and what is not settled?#

Because both definitions assume a human: a characteristic that comes from a creator, and a creator who is a person. No Turkish authority has said how they apply to a machine's output.

The characteristic test becomes a question of evidence: how much of this image or paragraph came from a person, and can you show where. No official Turkish statement or court decision on the copyright status of AI output was found in a search dated 7 November 2026; the words artificial intelligence do not appear in the consolidated Law No. 5846.

What the search did return was academic articles and law-firm commentary. Those are opinions and arguments, not rulings, and they should not be read as a green light. Foreign decisions are sometimes quoted in that commentary. This article cites none, because a ruling from another jurisdiction needs its primary text before it means anything for a hotel in Türkiye.

The practical consequence: you cannot plan on either answer, so build a marketing practice that holds up whether AI output is eventually treated as a work or not.

What do the tool's own terms give you?#

A contract right, not a copyright. The terms decide what you may do with the output commercially; they do not decide who owns it in law.

Find the commercial-use clause and check it against the plan you actually pay for: free and trial tiers often restrict commercial publication. Check what licence the provider takes over your inputs and outputs, and whether your uploads or generated files may be used to train or improve models. Look for an indemnity. It is usually absent, which means the resemblance risk sits with the hotel.

Note any disclosure duty in the terms, and any restriction on using output in a way that suggests a real person or a real place. Record every asset at the moment of generation: tool name and version, date, the prompt, the seed or settings, any reference images, and the output file exactly as delivered.

Check who holds the account. Output is usually granted to the account holder, so work generated inside a freelancer's personal account is a problem you inherit. Keep the record with the asset, not in someone's inbox: the record is what turns an argument into a file you can hand to a lawyer.

The same discipline applies to words as to images. If you are using generated text for room descriptions or newsletters, the questions about terms and records are the ones covered in AI for hotel marketing copy.

What if the output looks like someone else's work?#

Resemblance is the risk that actually reaches you, and the prompt log is the evidence you will need when it does.

A prompt naming a living artist, a photographer, a brand or a named hotel is a request for resemblance, and the result is judged on the output, not on your intention. Output can carry protected elements without asking: a logo, a character, a signature, a watermark fragment, or a building distinctive enough to be recognized. Uploading your own photographs as reference is usually fine on ownership and risky on terms: check whether the provider may keep and use what you uploaded.

Run a reverse image search before publishing, then look at the file itself for text artifacts, signatures, logos, recognizable architecture and faces. Keep the prompt and the generation record. Independent creation is a defence you can only make if you can show the path from prompt to file.

Ask a second person to look at the image next to the room it claims to show. A guest who books on that image expects that room, which is the separate honesty question covered in the AI-generated images checklist.

Does AI-made background music remove the licence question?#

No. It changes who you ask, not whether you ask.

Playing music in a public space generally needs permission from whoever holds the rights, and a lobby, a spa, a restaurant and a pool bar are public performance settings. The tool's licence covers the track it generated. It does not clear a melody, a riff or an arrangement that the track reproduces. A generated track can land close to an existing song, and the claim arrives against the hotel that played it, not against the tool. Confirm your own case with a lawyer.

Record every track the way you record an image: tool, date, prompt, reference material supplied, and the file as delivered. If you generated from a reference track or a hummed melody, write down what you supplied, because that is the first thing anyone will ask. Treat AI music as a new licence counterparty with its own paperwork, not as a way around the old one.

Who owns what an agency delivers, and who used AI to make it?#

Your contract decides, and most hotel marketing contracts were written before anyone thought to ask.

Ownership of delivered work normally has to be assigned in writing. A contract without such a clause leaves the question open, however much you paid, so confirm your own case with a lawyer. Add an AI disclosure clause: which deliverables used AI, with which tool, and whether any reference material belonging to a third party was supplied. Ask for a warranty that the deliverables are original and do not reproduce a third party's work, and for an indemnity that says who answers if someone objects.

List the files you want handed over: source files, layered files, prompts and generation records, so the asset is usable after the relationship ends. Write one rule for marketing and keep it: no AI asset goes live without a named owner, a recorded tool and a resemblance check. Read the contract before the campaign, not after the invoice.

The same reading discipline applies to your software agreements, not only marketing contracts. See hotel software contracts: clauses to read before you sign and AI clauses in hotel software contracts.

What to do next#

Before the next campaign, open two documents side by side, the tool's terms and your agency contract, and write one line in each: who owns the output, and who answers if someone objects. The law is unsettled, the terms decide what you may do, and the contract decides what happens between you and your supplier. Those three lines are the whole practice.