The short answer
Since 2 August 2026, Art. 50(4) of the EU AI Act requires you to disclose AI images that deceptively realistically depict a real room or house. AI texts only need labelling if they inform the public about matters of public interest and no one reviews them editorially. If you label AI images, mark them visibly on the image and show rooms only with real photos.
Key takeaways
- Hotels and hosts that use image and text generators are, as a rule, operators; for them Art. 50(4) has applied since 2 August 2026.
- An AI image that resembles a real room or house and looks like a photo is a deepfake and must be disclosed.
- Room texts that your team checks and takes responsibility for do not need labelling under Art. 50.
- Embellished room images can be misleading regardless of the AI Act, under § 2 UWG (Austria) or § 5 UWG (Germany).
- Taking stock of all images and texts takes about four and a half hours for a 24-room house in our sample calculation.
For hotels and hosts, Art. 50 of the EU AI Act requires two things since 2 August 2026: AI images that deceptively realistically depict a real room, house or place must be disclosed as artificially generated, and AI texts on matters of public interest must be disclosed in the same way, as long as no one reviews and takes editorial responsibility for them. Standard room descriptions that your team reads and approves are not subject to the text obligation, but embellished room images can be misleading regardless of the AI Act.
Status: September 2026. For your specific case, consult a lawyer.
What does Art. 50 require for AI images and AI texts?
Art. 50 distinguishes between providers, who develop AI systems, and deployers, who use them professionally. A hotel is, in almost all cases, a deployer. For providers, paragraph 2 applies: the outputs of their image and text generators must be marked in a machine-readable way as artificially generated [Q1]. This obligation falls on the makers of the tools, not on your business.
For you as an operator, paragraph 4 is the relevant one. It contains two separate obligations [Q1]:
- Deepfakes: Anyone who uses AI to generate or manipulate image, audio or video content that is a deepfake must disclose that it has been artificially generated or manipulated.
- Texts: Anyone who publishes AI-generated text to inform the public about matters of public interest must also disclose this, unless the text has been reviewed or editorially controlled by a person and a person bears editorial responsibility for it.
Paragraph 5 governs timing: the information must be provided clearly and distinguishably at the latest at the first interaction or exposure, and it must be accessible [Q1]. According to the European Commission, the obligations apply from 2 August 2026. The transition period until December 2026 under the Digital Omnibus only concerns the machine-readable marking by providers of generative systems that were already on the market before that date [Q2].
The obligations for chatbots in paragraph 1 are covered in two separate articles: the overview of Art. 50 for hotel chatbots and the question of what a chatbot in a hotel needs, and from when. How we implement labelling technically is described on the page EU AI Act in hospitality.
When is an AI room photo a deepfake?
Under Art. 3 No. 60, a deepfake is 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 [Q1]. Places and entities are explicitly included in the definition, so a hotel room can fall under it.
In its questions and answers on Art. 50 (status 24 July 2026), the Commission names two touchstones: a high degree of resemblance to something that really exists, and the context in which the image is shown [Q3]. If the audience does not expect a real recording, the content may not falsely appear to be authentic. According to the Commission, tools that serve only a supporting function for standard editing are exempt from the providers' marking obligation [Q3].
For everyday use in your house, this gives a rough classification:
- Exposure, white balance, straightening: classic image editing, even if the program uses AI. As a rule, not a deepfake.
- Power cable or smoke detector retouched out: a grey area. The more the image changes the room, the more likely the definition applies.
- Room refurnished with AI, view of greenery instead of the car park: the image resembles your real room and looks like a photo. A deepfake is likely here, and disclosure is mandatory.
- Mood image with no link to your house, such as a drawn illustration of breakfast: nobody takes it for a recording, so the obligation probably does not apply. A label does no harm nonetheless.

Why embellished room images are a problem even without the AI Act
A room photo that shows fittings or a view that do not exist can be misleading under competition law, with or without an AI notice. In Austria, under § 2(1) UWG, a commercial practice is considered misleading if it is capable of deceiving market participants about the essential features of a product and thereby inducing a decision they would not otherwise have made [Q4]. In Germany, § 5 UWG governs this and, among other things, covers information about the nature of a service [Q5].
A small note such as "AI-generated" does not necessarily correct a false depiction, because the guest books the room they see. The safer rule: show rooms, bathrooms, views and the exterior only with real photos.
How much work is the inventory for a 24-room house?
If you want to label or replace AI images, you first need an inventory. It takes only a few hours, as the following calculation with sample values shows.
Hotel case, sample values:
- Rooms: 24, spread across 4 categories
- Photos per category: 6, so 24 room photos
- Further images (exterior, breakfast room, wellness, surroundings): 16
- Images in total: 40
- Review time per image: 3 minutes, so 120 minutes
- Texts (rooms, offers, blog): 30, at 5 minutes each, so 150 minutes
- Total: 270 minutes, around four and a half hours
Why the effort? At 60 percent occupancy and an average stay of three nights, this sample calculation gives roughly 1,750 arrivals a year. Each of these guests compares the room with the picture they saw when booking.
Holiday apartment case, sample values:
- Apartments: 4
- Photos per apartment: 12, so 48 images
- Review time per image: 3 minutes, so 144 minutes
- Descriptions and house guide: 8 texts at 5 minutes each, so 40 minutes
- Total: 184 minutes, a little over three hours
With holiday apartments, one more point applies: the same images often also reach the portals via the channel manager. An image you replace there is replaced on the portals too.

How to label AI images and disclose AI texts
A label works when the guest understands it at first glance without having to search for it. The Commission requires a clear, distinguishable disclosure that people can perceive, for example visibly on the image, but it does not prescribe a fixed position [Q3]. Since 10 June 2026 there has been a voluntary code of practice on marking and labelling AI-generated content, along with EU symbols that operators can use [Q6].
A practical sequence:
- Create a list: every image and every text, with its origin: photo, edited, AI-generated, or AI draft with review.
- Rooms and house only real: replace AI-altered room images rather than labelling them.
- Label AI images visibly: a short note directly on the image, such as "AI-generated image", plus the same note in the alt text for screen readers.
- Texts with approval: whoever reads an AI draft, corrects it and approves it has the approval entered in the list with a date.
- Transparency page: a page describing what your house uses AI for, with a contact for questions.
For questions about implementation in Austria, the RTR's AI service desk (KI-Servicestelle) is a point of contact. It gathers information on Art. 50, the Commission's guidelines and the code of practice [Q7].
How we handle this at HOHEIT
At HOHEIT, labelling is part of the system rather than manual work at the end. Our AI concierge for guest questions identifies itself as AI at the start of every conversation and hands over to your team when it has no verified answer. It only answers from the information you release.
The same approach applies to images and texts on the website: we show rooms, houses and apartments with your real photos. Where the AI concierge answers in several languages, the same rules apply, as described in the article on the multilingual AI concierge.
When labelling is not necessary
Not every use of AI triggers an obligation. As a rule, the following do not require a disclosure under Art. 50(4):
- Room and offer texts that your team takes over from the AI as a draft, reviews and takes responsibility for, insofar as they count as information on matters of public interest at all [Q3].
- Translations that a person proofreads and approves.
- Photos that have only been edited for brightness, colour or cropping.
Your transparency page may still mention voluntarily that texts are written with AI support and checked by people.
If you want to know which parts of your website and guest contact are affected, look at how our AI concierge with labelling works, and go through the list with us in the first consultation: book a call
Sources and further reading
[Q2] EU Commission: Quick Facts, Transparency rules for AI systems (as of 29.07.2026)
[Q3] EU Commission: Transparency obligations under Article 50 of the AI Act, FAQ (as of 24.07.2026)
[Q4] RIS: § 2 UWG, misleading commercial practices (Austria, retrieved September 2026)
[Q5] Gesetze im Internet: § 5 UWG, misleading business acts (Germany)
[Q7] RTR: KI-Servicestelle, the AI service desk (retrieved September 2026)
Häufige Fragen
Do I need to label photos that my smartphone automatically improves with AI?
As a rule, no. Automatic exposure, sharpness or noise reduction does not change the content and is more likely to count as standard editing. It only becomes critical when objects are removed or added, or rooms are altered.
Is a general note in the legal notice that the website contains AI content enough?
For deepfakes, probably not. The disclosure must be clearly recognisable at the latest on first contact with the content, that is, directly on the image. A general sentence on the transparency page supplements this but does not replace it.
What about AI images I post on Instagram or Facebook?
Art. 50 does not distinguish by channel. A deceptively realistic AI image of your house needs a disclosure on social networks too. Many platforms offer their own labelling features, which you can use in addition.
Who monitors compliance in Austria?
Member states designate market surveillance authorities. You can find out which body is responsible in Austria and what the current position is at the RTR's AI service desk (KI-Servicestelle).
Related service
EU AI Act
Chatbots, automatic emails, AI images: what the EU AI Act requires of hotels and hosts from August 2026, and how we implement it.
Scope, process and pricesTalk to us
Direct booking, without the commission
We build the booking site, connect it to your channel manager and keep it running.
Book a first call