How our AI chatbot handles questions it can’t answer
We love sharing real user tests of our AI chatbot (powered by ChatGPT 5.4) — especially the kind designed to push it to its limits.
Here’s what a clever tester asked on our 500 pages demo cooking blog 👇
Q:
4 students, sausage and potatoes.
Fabio AI Chatbot:
I can help, but I need a quick clarification: the website content only mentions potatoes and ingredients like onions, tomatoes, chicken breast, canned tuna, spinach, lentils, and butter. It does not provide any recipe or method using sausage.
Do you want a potatoes-based idea for 4 people using only the ingredients listed on the site?
In that case, the demo blog doesn't have any sausage and potatoes recipe (again, it's a demo blog 🙂). So The AI (for this one the setting wans on ChatGPt 5.4) didn’t have a ready-made answer.
Instead of:
hallucinating a recipe
or stopping with “no results”
It did 3 things:
Grounded the answer in available data
Kept the conversation alive with a clarification
Proposed a relevant alternative
👉 It’s not just answering — it’s guiding the user despite missing data.
So here’s the real question:
If your AI can’t give the exact answer…
is this kind of response enough to keep users engaged?
Or would users still drop off?
In the mean time, test Fabio AI Chatbot on our 1000 products E-commerce demo store and on our 500 pages demo cooking blog.
