Gemini 2.0, or specifically Gemini 2.0 Flash, is fast becoming one of my favorite AI models. It’s not just the fact it’s currently free, but also about speed and versatility. It’s significantly faster than a lot of the alternatives, and has a huge context size capability. This means it can process a large amount of information in one go. That’s great for experimenting with new uses.
I decided to put it the test by creating a shopping assistant to save me time and hassle when searching for a particular item. Mostly, of course, I use the excellent buying guides here at Tom’s Guide to help me choose what to purchase — but how could AI help me with that? Here’s what I found out.
Finding the right prompt
The first call was to ask Gemini to create me a prompt which would deliver exactly the functionality I needed. This is one of the main beauties of AI, you can use it to help you get the best out of itself. What it ended up producing was quite a lot of text, which is why the context size matters in this kind of use case.
If you’re interested, you can find the full prompt here, but essentially it consists of three parts. The first part of the prompt tells the model exactly what you’re aiming for and what you hope to achieve. “You are an expert shopping assistant…your goal is to provide.”
Part two gives the model the workflow it should follow to produce the best result, and part three covers other factors such as user interaction to ask additional questions and using web search.
AI in real time
It’s important to note that one of the key criteria is to have some sort of real time access to information on the web, so for that reason you’ll need to make sure you have a chatbot wrapper which provides web search functionality.
I would suggest either the PageAssist browser plugin (https://pageassist.xyz/), or TypingMind (www.typingmind.com). Once installed you’ll need to add the shopping prompt to the model’s system prompt or create a separate agent out of the prompt text, then ensure that web browsing is enabled (it’s usually a manual toggle of some sort).
Don’t be afraid to make your own custom tweaks to get exactly what you want, since this is a personal tool and you have the power.
In my early tests Gemini delivered almost exactly what I needed. However I did have to fine tune the prompt to polish it up a little, for example delivering live links to the actual products, so I could evaluate the recommendations with one click.
The prompt also allows me to add three key features to the search, so I can, for example, add “wet and dry”, “cheap” and “rotary” for a men’s shaver search. This is useful because it lets the assistant give me a scale of feature relevance in the recommendations to help me make a better choice.
Don’t be afraid to make your own custom tweaks to get exactly what you want, since this is a personal tool and you have the power.
The great thing about creating this kind of agentic prompt with your AI model is the fact that you can repeatedly use these tools for all sorts of things over time. Once you’ve got a good solid library of agent prompts like this, you can genuinely save a decent amount of time with many of your day-to-day tasks.
Note that if you’re using a chatbot wrapper like PageAssist, with access to lots of various different models, you can test with both free and commercial ones to see which work best for your needs and budget.
Finally, while today this is a manual process, tomorrow it’s extremely likely it will be automated to make it even easier to use.