A contribution by Sophie Hundertmark.
Sophie Hundertmark is an expert in the practical use of artificial intelligence with a focus on chatbots, AI strategies and responsible technology integration. She is a researcher and lecturer at the Lucerne University of Applied Sciences and Arts and is currently writing her dissertation in the field of Conversational AI at the University of Fribourg. As a consultant, she supports companies, administrations and educational institutions in the introduction of effective AI solutions. Find out more about Sophie Hundertmark on Linkedin.
Do you know the feeling when a message seems strange to you? That’s exactly what happened to me once again – on the Tutti.ch platform, which many of you may recognize as being similar to eBay classifieds. And this is exactly where I have used AI to build a practical tool that protects me (and from now on you too) from such scams.
The trigger: suspicious messages in seconds
I had created an advertisement on Tutti.ch and had only just finished – when four messages arrived within a few seconds. They all sounded similar. Too similar. Partly unformatted texts, conspicuously polite language, immediate contact request via WhatsApp. It quickly became clear to me that these were not normal interested parties – this was a scam.
I asked myself: Could ChatGPT see the same warning signs as I do – perhaps even faster and more systematically?
Step by step: How I trained the AI to detect fraud
1. feed in raw data: The original messages
I first pasted the four suspicious messages completely unformatted into ChatGPT. It was important not to make any assumptions – simply: “Here are four messages. Find similarities and possible warning signs of fraud.”
2. have ChatGPT analyzed
ChatGPT analyzed the messages for patterns in a matter of seconds and created a list of typical characteristics for me, e.g:
- Message arrives a few seconds after publication
- No direct reference to the specific advertisement (often only “the article” or “the product” is mentioned)
- Contact via WhatsApp desired
- Unusual formatting of telephone numbers
- Excessively polite or formal tone
- Repeated text modules across several messages
Additional added value: ChatGPT has provided an example from the messages for each feature. I checked the result and realized: This is exactly what I would have written down myself – but it was much faster and was formulated in a standardized and structured way so that it can be used by an AI in the further course.
3. develop promptly for reuse
The next step was to ask ChatGPT:
“Create a prompt that I can use in the future to automatically analyze such fraud messages.”
ChatGPT then provided me with a suggestion as to how I can structure the input – i.e. which instruction and which features should be taken into account in future.
I then manually adapted and supplemented this prompt, e.g. by adding the information that many messages appear immediately after publication – this was a feature that ChatGPT could not recognize from the existing data alone.
4. build custom GPT
Then came the decisive step: I converted the customized prompt into a custom GPT. This means that I have created my own GPT wizard on the chat.openai.com platform – with exactly the rules and instructions that I had previously defined together with ChatGPT.
Here’s what I did:
- Define name and role: The GPT should be a security assistant for sellers on Tutti.ch
- Insert guidelines: I have inserted all recognized characteristics of fraudulent messages as a list
- Define rules of conduct: The GPT should warn the user for each individual feature and cite the location in the text
- Include a disclaimer: Clear indication that this GPT is not officially from Tutti
- Add data protection notice: Do not enter any personal data, as everything is processed by OpenAI
You can also find more details on creating CustomGPTs in one of my last posts.
5. test phase with real examples
As with all AI projects, sufficient testing is essential. That’s how the test phase started for me before I was able to share the CustomGPT with others.
I then tested the GPT with real messages – even with non-suspicious requests. A message formulated in Italian, for example, was correctly recognized as unproblematic. The indications were comprehensible and the assessment matched my own feelings.
6. release for all
Finally, I have made the GPT publicly available in the Custom GPT Store so that you can use it too. You can find the AI bot for fraud detection publicly in the GPT Store.
What you can learn from this – and why the example goes far beyond Tutti
In my case, fraud detection was only one specific reason. But if you look at what’s behind it, it becomes clear that it’s not about this one solution – it’s about a whole principle that you can apply to many other situations.
The real potential lies in the fact that an AI like ChatGPT recognizes patterns, repetitions or anomalies at lightning speed, which we humans often either don’t notice at all – or only after many repetitions. This is precisely where the added value lies.
Here are a few concrete lessons and application ideas that you can take away with you:
1 AI can support you in pattern recognition – faster and more objectively than your gut feeling
Many decisions in everyday life or at work are based on intuition. But what if your gut feeling is wrong? Or when you’re so stressed at work that you don’t even look closely anymore?
With good AI support, you can:
- Check form letters, applications or inquiries for recurring formulations or copy-paste texts
- Have comments, emails or support tickets analyzed for mood, urgency or seriousness
- Automatically recognize repetitions, risky clauses or missing wording in contracts
- Have invoices checked: Are the format, amounts, payments, deadlines correct? – without having to go through every line manually
There’s no question that sometimes a human gut feeling is still better than AI, or at least that’s exactly what distinguishes a human from an AI in a given situation.
2. you can define your own rules – and train the AI to follow them exactly
The biggest advantage: You don’t have to wait for ready-made tools. You can feed ChatGPT with your examples and logic – and create something that fits your case exactly.
A few ideas:
- A custom GPT that automatically classifies new customer inquiries into categories such as “standard question”, “urgent”, “special case”
- An assistant that scans social media comments and filters out all critical voices, for example, so that you can react more quickly
- A helper for HR that checks applications for minimum criteria, tonality and comprehensibility
- A reminder tool that warns you in texts or presentations if important information is missing or the wording is too vague
3. you can automate routine tasks – and thus gain time for the essentials
It gets really exciting when you realize how much time is wasted every day on small, repetitive tasks:
- Do you keep comparing the same Excel spreadsheets?
- Do you check emails every day to see whether they contain an invoice or a letter of application?
- Do you read through feedback to see “between the lines” whether someone is satisfied?
A well-configured GPT can pre-filter or pre-analyze all of this – so all you have to do is check and decide.
4. you can increase security – without external tools or IT projects
What I’m showing you with the Tutti example is also a kind of cybersecurity for non-technicians. You don’t have to buy big security solutions – you can:
- Have emails checked to see if they sound like phishing or fake invoices
- Check inquiries in the web store or via contact forms to see whether they contain spam patterns or fake names
- Have internal messages scanned for potentially unsafe content or data breaches
These simple applications are of great added value, especially for small companies, the self-employed or associations – because there is often no budget for professional solutions.
5. you can train yourself further – and use the AI as a sparring partner
What I experienced when setting up my GPT: You learn an incredible amount in the process! Because as you establish rules, give examples and refine prompts, you yourself become more sensitive to language, patterns and logic.
You can use ChatGPT not only as a tool, but also as a co-thinker, trainer and feedback provider:
- Formulate a set of rules and ask GPT to apply them to your own texts
- Get feedback on emails, posts or offers – in terms of content, language and style
- Practise changing perspective with GPT: “How does this text affect a customer who has little prior knowledge?”
Conclusion: The real added value is not in the tool, but in the principle behind it
The trick is not that you simply have a “cheating GPT” or an AI to detect cheating.
The real progress is that you solve a concrete problem with AI – and thus understand how powerful, flexible and adaptive this technology is.
Whether in the private sector, customer service, human resources or marketing – wherever language, data and decisions play a role, you can get massive support with ChatGPT systems.
And you don’t need any coding, programming experience or a big budget. Just a good idea, a few examples – and the desire to try it out a little.
Any further questions?
Do you have any questions? I am happy to support you, act as a sparring partner and answer your questions. I am always happy to receive your messages, preferably by WhatsApp message or e-mail.
This article is also available as a podcast episode
Attention! The podcast was created entirely by my AI-Assistant based on my contribution – no guarantee for incorrect content.
