fragRoger · Workshop with Roger Basler de Roca
This page is for quick reference after the workshop. Every prompt is ready to use here, each with a button to copy it. You paste it into ChatGPT or Perplexity and adjust the details in square brackets.
No course, no sign-up, no account. Use the search field above to find a prompt in seconds.
To get started
Before we get to the prompts: three short explanations, without technical jargon.
Artificial intelligence, or AI for short, is a computer program that has learned from enormous amounts of text which words usually follow one another. From this, it writes new sentences. It does not think, believe or know anything in the human sense — it calculates very well what fits next.
That is why the texts are often surprisingly good, and yet sometimes wrong. You remain the person responsible for the result.
A chatbot is the interface for this: a window like on WhatsApp. You type in an instruction, called a prompt, and get a reply. You can ask follow-up questions, correct and refine it, until the result fits.
Within one conversation, the chatbot remembers what was said. A new conversation starts again from scratch.
ChatGPT is the writer. It turns your keywords into texts, creates images and summarises content. It mainly draws on what it has learned and usually does not name a source.
Perplexity is the researcher. It searches the web first, then answers, and shows the link each statement comes from. In exchange, it writes less elegantly and does not create images.
In short: write with ChatGPT, source with Perplexity, verify with Crapcheck.ai.
First things first
Is it about a specific, named person? Then don't enter it. Is it about a general pattern, without names or details? Then there's no problem.
Section 1
You provide the keywords, the AI turns them into a text. Always read the result through carefully before sending it.
You supply the keywords, the AI does the writing. Nothing is invented that you haven't provided.
Turn these keywords into a friendly invitation: [keywords]. Tone: warm, clear, not too formal. Plain, natural English.
For notes that are correct in content but still rough in wording. Remove names beforehand.
Here are my notes from a meeting: [notes]. Turn them into clean minutes with agenda items, decisions and open points. Factual, plain English, no embellishments.
Useful when a text is fine in content but sounds too strict or too casual.
Rewrite this text without changing its content: [text]. Tone: friendly and respectful, short sentences, no sales language. Plain English.
Section 2
The first result is rarely the best one. Expect two to three rounds of tweaks — that's normal and quick.
The more specific the details, the more usable the image.
Create a warm, welcoming illustration for [event, e.g. harvest festival]: [2-3 concrete details, e.g. laid table, candles, autumn leaves]. Calm, dignified style, not kitschy.
The first result is rarely the best. Simply say what should be different.
Same image, but: [change, e.g. fewer people, brighter colours, more space at the top for text]. Keep everything else the same.
Section 3
Here text, image and research work together. This is exactly how it happens in the office.
First the text, then a matching image. Two steps, one result.
I'm planning a flyer for [event] on [date] in [place]. Write me a short invitation text (max. 60 words) and then suggest a matching image idea. Plain English, warm and clear.
Tap the microphone icon, speak freely, then give this instruction.
Listen to my voice message and turn it into a print-ready email to [recipients]. Tone: grateful, clear, not promotional.
Always useful when the information must be correct: figures, dates, regulations.
What is currently the rule in [diocese / regional church] regarding [topic]? Please include the source.
Section 4
That's all you need: ChatGPT writes, Perplexity sources, Crapcheck checks. All links open the respective tool directly.
Writes texts, creates images, understands dictation. For invitations, newsletters, minutes and illustrations.
Caution: ChatGPT can invent details that sound plausible. Don't rely on it for facts.
Open chatgpt.comSearches the web and shows the source directly. For diocesan information, figures, dates — anything that must be correct.
Rule of thumb: write with ChatGPT, verify with Perplexity.
Open perplexity.aiChecks a statement, a passage of text or a link for soundness and shows how solid the sources are.
Ideal for anything that comes from an AI and later ends up in the parish bulletin or newsletter.
Open crapcheck.aiLicences
Free and paid versions — and what nonprofit organisations pay.
The free versions are enough for everyday parish office work. Paid versions are faster, allow more images and longer documents.
Try the free version first and only pay once the limits genuinely get in your way.
See ChatGPT pricingOpenAI offers ChatGPT Team at a reduced price for nonprofit organisations, and in some regions discounted ChatGPT Enterprise as well. Eligibility is confirmed through a verification service.
Depending on their legal form, parishes may qualify as nonprofit organisations — eligibility is checked during sign-up.
Open OpenAI for NonprofitsImportant: a paid licence does not change the privacy rules. Never enter personal data from registers, pastoral care or accounting.
Important
AI answers sound confident, even when they are wrong. So briefly check everything you publish or send out. Three minutes is usually enough.
Mark every specific detail in the AI text: date, time, name, figure, quote, statement of church law. These are exactly the spots at risk.
Copy the statement or link into crapcheck.ai. You'll get an assessment of how solid the claim is and where the weak points are.
Ask Perplexity about the detail and request the source. Actually open the link, rather than just reading the summary.
For parish and diocesan information, in the end only the official website of the parish, diocese or municipality counts — or a phone call.
A detail you can't back up doesn't belong on the notice board. Better one line less than a correction later.
Crapcheck.ai is our standard tool for a quick fact check: paste in the statement or link, run the check, read the assessment and reasoning. Don't enter any names or personal data from the office there — only the factual claim to be checked.
Check a statement on crapcheck.aiImportant
Since 2 August 2026, the transparency obligations of the EU AI Act (Article 50) apply in full. In short: anyone who creates or edits images, videos or audio using AI must disclose this. For the parish office, this mainly means: label AI images in the parish bulletin, on the website or on notices.
Switzerland has no AI-specific law of its own. What applies is the revised Federal Act on Data Protection (FADP), copyright law, personality rights, and for church bodies additionally the relevant church data protection rules. The Federal Council has opted for a sector-by-sector approach and has initiated ratification of the Council of Europe's AI Convention. There is no explicit declaration duty here today comparable to the EU AI Act.
Nevertheless, the EU AI Act can still affect you as soon as your content is accessible in the EU or aimed at people there, for example with a bilingual website or parishes near the border. And regardless of the law: whoever fails to disclose and gets caught loses trust.
Basics
These ten points apply equally to ChatGPT and Perplexity. Following them means far fewer attempts are needed.
A prompt improves instantly if it states the role and the audience: "You are the secretary of a parish. Write for older parishioners." The AI then gets the tone and word choice much more accurately.
"Around 120 words, friendly, with a title and three paragraphs." Without such details, you often get texts that are too long and too sales-like.
Paste in an earlier invitation from your parish and write: "Write in the same style, but for this topic." The result will then sound like your parish, not an advertising agency.
Simply reply in the same chat: "Shorter." "Less flowery." "The date is wrong, it's 14 November." The AI remembers the course of the conversation.
One chat for the parish newsletter, another for confirmation. This way details don't get mixed up and you can find old results more easily.
Work with [name], [date], [place] and insert the real details by hand only at the end, in Word. That way no personal data leaves the office.
In the ChatGPT app you can use the microphone and simply talk through what should go in the text. For many people this is faster and feels more natural.
Figures, deadlines, addresses, diocesan regulations: ask Perplexity here and click through to the source shown. Only use the detail once the source is credible.
"Give me three versions: a factual one, a warm one, a short one." Choosing is easier than writing it yourself.
Always proofread names, dates, times and theological statements yourself. The AI is a writing aid, not the person responsible.
Reference
So that technical terms don't get in the way.
Finally
Tap a question to reveal the answer. Only one answer is open at a time.
Isolate the specific claim and check it on Crapcheck.ai (crapcheck.ai). In addition, look up the source in Perplexity and cross-check it against the official website of the parish, diocese or municipality. Leave out anything you can't verify.
Not for everyday office work. The free version writes texts and creates images. If you use it daily and often hit limits, the paid version can be worthwhile.
No. Names, addresses, dates of birth and death, details about baptism, marriage, illness or pastoral care do not belong in an AI tool. Work with placeholders and add the real details only in the finished document.
If you revise the text and add your own words, usually not. More important than that question: the content must be correct and fit your parish.
Tell the AI directly in the same chat: "The date is wrong, it's actually 14 November. Please correct the text." For factual questions, switch to Perplexity and check the source.
You may use the results for your parish, including in the parish bulletin. Be cautious with images of real people, logos or artworks. When in doubt, choose a neutral subject.
Hands, writing and faces are difficult for image AI. Describe the subject more simply, avoid text within the image, and have two or three variants generated.
Yes. Write in the prompt: "Plain British English, formal but friendly tone." That will then apply for the whole chat.
No. Within one chat, the context is retained. For a new chat, simply copy the matching prompt from this page.
Yes, do it. Since 2 August 2026, the EU AI Act (Article 50) requires that AI-generated or AI-edited images, videos and audio be recognisable as such. A short caption like "Image: created with AI" is enough.
Switzerland has no AI-specific law of its own; the FADP, copyright law, personality rights and church data protection rules apply. The EU AI Act can still affect you if your content is aimed at or accessible to people in the EU. When in doubt, declare it.
Send Roger Basler de Roca a short message via WhatsApp or email. Even weeks after the workshop.
Section 5
Write to me. A short question via WhatsApp is explicitly welcome, even weeks after the workshop.
Ask a question via WhatsAppThe workshop handout as a PDF, to read on screen or print out.
Download handout as PDFPDF, around 0.6 MB. More templates will be added here.
Contact
This page sets no tracking cookies and requires no sign-up. Prompts are working aids; always check content from AI tools yourself.
The legal notice and privacy policy are only available in German.
fragRoger · Roger Basler de Roca
Bonus
🎵 To finish: this song was made with Suno, from a few keywords about the workshop. That is what AI music sounds like — and exactly how it has to be labelled.
Listen to the song on Suno