Prompting AI
A few tips for prompting AI and creating Gemini Gems to get things done.
I use AI at work to help create content. I never use the output verbatim, I tend to use what is generated and then write my own version. This may seem a little wasteful given that the generated text is generally pretty good, but I feel like it lacks authenticity if I just pass it off as my own, despite being encouraged to. The biggest benefit of using AI is that it can help the scary blank page problem.
So with this in mind, I've been thinking about strategies to make AI output closer to what we need. Gems are custom Gemini chatbots that can take on a persona. Most importantly, you can give a Gem context on how to do something with examples to help create better copy. So how to improve the output? Give Gemini a detailed writing guide for the topic. I wrote a Content Kit that would help both new members to my team, but also Gemini. I also added some of our best examples.
Gems need instructions as well as context, which is where I've spent a lot of time until I realised I could use a fairly standard template and work from there, even getting Gemini to write it's own instructions!
4 Prompting tips
- Purpose: The aim of the Gem and the output required
- Audience: Who the output is for - what do they need, what level of detail
- Style: The tone of voice, length, language, expressing numbers, etc
- Task: A step by step guide to creating the output
I explain I have added specific context it should reference, but not use as a source. I have also instructed the Gem to ask the user questions if something is missing.
The result has been positive. I've created several Gems to help with content outlines, animated video and simple knowledge checks.
One last point to make is that these things take a lot of testing and tweaking, so make sure the effort is worth it.