Topic 2.3. What GenAI can generate
In the previous lesson, you learned how GenAI works under the hood: neural networks, layers, and word-by-word prediction. Now let’s look at what comes out the other end.
In this section, we’ll explore each content type through a single scenario: a small agritourism preparing an autumn marketing campaign.
The painter analogy
Before we dive in, here’s a useful way to think about what GenAI does across all content types. Imagine a painter who studies thousands of landscapes; not to copy any single one, but to internalise the rules of perspective, light, and colour. When this painter creates a new landscape, it follows those learned principles while being entirely original. GenAI works similarly, whether it’s producing a sentence, an image, or a melody. It has absorbed the “rules” from millions of examples and uses them to create something new, a fresh composition built on learned patterns.
Five content types
1. Text generation
Text generation is probably the GenAI capability you’ll encounter most often. These models can draft emails, write articles, summarise long documents, translate between languages, generate computer code, and power conversational chatbots.
Agritourism campaign example – The owner of a small agriturismo needs to write promotional descriptions for her off-season packages in Italian, English, Portuguese, and Lithuanian. She types a prompt into a text generator: ‘Write a warm, authentic description of a four-night autumn retreat at a rural guesthouse in the hills above Arcos de Valdevez. Include farm-to-table dinners, guided hikes through chestnut forests, and a wine tasting with a local producer. Target: travellers looking for slow, crowd-free experiences away from mass tourism.’ The tool produces a polished paragraph. She then asks it to translate the text, adjusting the tone for the target audience.
Stop and check – Exercise 1
Write a prompt
Access to a chatbot (ChatGPT, Copilot, or Gemini), try this:
- Write a prompt asking the AI to create a short promotional text for a hotel, hostel, or restaurant you know (or invent one). Include at least three specific details in your prompt (location, type of guest, a special feature).
- Rate the output. On a scale of 1–5, rate it for: accuracy (is everything factually correct?), tone (does it sound right for the audience?), and usefulness (could you actually use this with minor edits?).
2. Images generation
Image generation is one of the most visible applications of GenAI. Models like DALL-E, Midjourney, Adobe Firefly, and others can create realistic photographs, illustrations in diverse styles, 3D models, and edited versions of existing images. You can control aspects like colour palette, angle, lighting, and artistic style through your text instructions.
Agritourism campaign example – The agriturismo owner has plenty of photos taken on her phone, but they don’t always look professional enough for social media. She uploads a photo of the breakfast table on the terrace; the light is a bit flat, and the framing is off. Using an AI image-editing tool, she enhances the lighting, adjusts the colours to bring out warm tones, and crops the image to a more appealing composition. The result looks polished and inviting, and, unlike a fully AI-generated image, it’s an authentic photo of her agriturismo.
Stop and check – Exercise 2
Camera or AI?
Look at the images and decide: is it generated by AI or not?
3. Audio generation
GenAI can compose music, create soundscapes, generate realistic voiceovers, and synthesise speech in multiple languages. This ranges from full tracks with vocals and instrumentation to short ambient clips or spoken narration based on a text script.
Agritourism campaign example – The agriturismo owner wants to add an audio guide to her website, so guests can listen to a short welcome message and a description of the local area before arriving. Instead of hiring a voiceover artist for four language versions, she uses an AI voice tool to generate a warm, natural-sounding narration in English, Portuguese, Italian, and Lithuanian, all from the same written script. She also uses an AI music tool to create a soft background track that evokes the peaceful countryside atmosphere and layers it under the narration.
4. Video generation
GenAI can generate short video clips from text descriptions or still images, create realistic AI avatars for presentations and tutorials, and automate editing tasks like adding subtitles, trimming clips, or combining sequences.
Agritourism campaign example – The agriturismo owner wants to post a 30-second video reel on social media showing the property’s highlights. Using an AI video tool, she uploads a few photos taken on her phone and types a description of the desired sequence: ‘Start with the misty hills at dawn, transition to the breakfast table on the terrace, then the chestnut forest trail, ending with the fireplace in the evening.’ The tool generates a first draft with transitions and background music. She reviews it, adjusts the pace of a couple of transitions, replaces one clip that doesn’t feel right, and posts it.
Stop and check – Exercise 3
Text, image, audio, or video?
For each task below, decide which GenAI content type would be most useful: Text, Image, Audio, or Video. Some tasks might benefit from more than one — if so, explain which types and in what order.
5. Synthetic data
There’s one more type of GenAI output that’s less visible but increasingly important: synthetic data. This is artificially generated information designed to mimic real-world data without containing any actual personal or sensitive details.
Synthetic data is information that’s artificially generated rather than collected from real-world events. Crucially, it’s created to mimic the statistical properties and patterns of real data, meaning it looks and behaves like actual data without containing any sensitive or original information.
Think of it as a realistic “practice version” of data. For example, a hotel chain developing a new AI booking system needs thousands of sample reservations to test it, but using real guest data raises privacy concerns. Instead, they generate synthetic booking records: fictional guests, dates, room types, and preferences that follow the same statistical patterns as real bookings. The system can be trained and tested thoroughly without any real guests’ information being exposed.
This application is more behind-the-scenes than text or images, but it plays a critical role in building and improving the AI tools you’ll use in your future career.
Exercise 4
Plan a hotel campaign
You have the task of planning a short marketing campaign for one of the following scenarios. You have access to GenAI tools for text, image, audio, and video.
- A small boutique hotel launching a new “Local Experience Weekend” package.
- A city tourism board promoting a “Meet a Local” programme, where visitors are matched with residents who show them the city through their own eyes; no professional guides, just real people sharing their favourite spots.
- A coastal resort launching a “Beach for all” package, designed for guests with mobility challenges, including adapted water sports, barrier-free pathways, and staff trained in inclusive hospitality.
- A rural region promoting a “Slow Travel Week”: no cars, no schedules, no crowds. Guests walk between villages, eat with local families, and help with seasonal farm work like olive picking or grape harvesting.
- Plan the content. Decide what you need to create: at least one piece of text content, one visual, and one audio or video element. Write the AI prompt you would use for each.
- Identify the risks. For each piece of content, write down one thing that could go wrong if you used the AI output without reviewing it.
- Define the human role. For each piece, explain what you should check or adjust before publishing.
The Lesson In a Nutshell
Generative AI can create many types of content. In this lesson, we explored five of the most relevant for the tourism sector: text, images, audio, video, and synthetic data. The underlying mechanism is always the same because, as we saw in the previous lesson, the model applies the patterns it absorbed during training. Each type of content, however, has its own characteristics, strengths, and pitfalls. We saw them in action by following step by step the marketing campaign of a small agriturismo: an example you can adapt to your own context, taking inspiration from the exercises you’ve just completed.
What could this look like in my workday?
Imagine you work for a cultural association that organises walks exploring a lesser-known Turin. You want to promote a new route through Porta Palazzo and Borgo Dora. Using a text generator, you write a post describing the stops: the Porta Palatina and the Roman archaeological park, the Cortile del Maglio with its history as a military arsenal turned centre of solidarity, a stop among the multicultural stalls of Europe’s largest open-air market, and a bicerin at the historic café opposite the Consolata sanctuary. Using an image tool, you enhance photos taken during the scouting visit: correct the lighting under the arcades, bring out the colours of the market stalls. Using an audio tool, you generate a short narration in Italian and English for the promotional video. You review everything, check that names, addresses, and opening hours are up to date, and publish. Three types of content, three different tools, one workflow, and the important decisions remain yours.
Why does this matter in my future job?
Knowing what AI can do helps you identify your clients’ needs and choose the best type of service or content to meet those needs. It also allows you to use AI tools with confidence and create materials that reflect your personal style, and the profile and values of the company or organisation you’ll be working for.
Take a moment to reflect
Think about the five types of content you’ve just explored. Do you think they could really be used by a small restaurant, a cultural association, or a school like yours? What would it take? Time to learn how to use them? Someone who can identify the right needs? Awareness of what these tools can and cannot do? Which of these barriers seems easiest to overcome, and which would require a bigger change?