What I actually ask AI

I started using an AI chatbot the way a lot of people do, for conversations, experiments, and a few questions. After a while I wanted it to do the work I used to give Google.

Buying a dishwasher or a TV where I live meant tabs and comparison sites written for another country, listing shops I cannot buy from. Building a trip meant the same five famous landmarks, then a long hunt through forums for anything actually local. I did not want to spend five hours on that every time. I wanted to give the assistant some context about me and get a recommendation back.

So I started doing that. Sometimes the recommendations were useful. Sometimes they were wrong, in different ways.

It named a place nearby that had already closed. It described a normal price as a discount. When I asked, the assistant said it had made a mistake. The voice had sounded sure either way.

That is when I started paying attention. What is going on? How does this actually work? Can I even trust it?

I write this down for people who ask the same way, and for people who run an online shop, a café, a restaurant, a barber shop, or a travel agency that did or did not make that list.

The questions I actually ask

Four patterns keep coming back.

Products that fit a real person. I am in my forties. I train a lot. I do not ask “what are the best supplements.” I say what I actually do, including sleep or health constraints when they matter, and I want options I can actually buy here, not a US drugstore default. The same habit covers a dishwasher or a TV: a real option at a price I can pay, not a list of famous brands.

Travel that is not a brochure. I do not want the tourist loop. I want where locals eat, walk, and sit, or a nearby city with that vibe instead of the famous one. Who I travel with changes the answer: alone, with friends, or with my parents. A vague “best things in Milan” is how you get the brochure.

A specific offer, judged. I am not a wine expert. I came from beer. When I see a mixed case of seven or ten bottles from various wineries at a “good” price, I want a verdict: is the wine any good, are the wineries known, and is the price actually cheap or is the discount marketing? I do not want to price-check ten bottles by hand.

A stop on the way. Driving from here to there, I ask for coffee and cheap bites on that road. That is a different question from “best things to do near me” in a city I am not standing in. It fails in a different way when the listing is out of date.

I usually type in English. Sometimes the place names are in the local language. That mix still forces the assistant to guess whether the list is for a tourist or for someone who lives here.

The rest of my work is on my personal site, duskyai.com. This site is the one about those recommendations.

What I am trying to find out

I want to know which options come back, how they got there, and whether I should use them.

Deeper dives: Without your context, the answer is for someone else · Say what you want the answer to do · “Best things to do near me” is how you get the tourist list · AI place answers come from different directories · Why AI won’t just say it doesn’t know.

I started on ChatGPT, then used Gemini, and often Grok because the free tier is usable. The same specific prompt does not return the same options. I want to know why, and whether paying for a “better” model is worth it for this kind of ask. Ask the same thing in two tools is how to compare those answers across tools. Free, Plus, and Pro can also be different models inside ChatGPT. I have not run a paid-versus-free log yet.

Sometimes the miss is worse than a wasted afternoon. Google once told people to put glue in pizza sauce. In August 2026, three hikers on Mount Shasta packed what Gemini told them to bring, and they had to be rescued. I ask these tools where to stop, what to buy, whether a price is real. I need to know when they are simply wrong, and when someone can get hurt.

The cases are here: AI can give stupid or dangerous advice.

What that means here

This site is about those recommendations. When I ask with real constraints, which options come back? Are the facts right: the shop, the price, the hours, whether the discount is real?

If you run the restaurant that is still recommended after it closed, or the local place that never beats the tourist list, those are the same questions from the other side. Supplements, a wine case, and a weekend drive are not the same kind of ask. The way a recommendation gets chosen is still the same.

That small set of options is the shortlist. How that works: How AI chatbots build a shortlist.

Words this site uses live on the Glossary. Papers and official docs live on Resources.

Read next if: How AI chatbots build a shortlist · Why AI won’t just say it doesn’t know