Transcription
By now you are already familiar with the basics. You know how modes work and you have even played around with super mode and context capture. But now we're going to look at something that lets you take all that and make it your own. I am talking about custom modes. This is where Super Whisper gives you total control to build workflows that fit exactly what you need.
Let's go over to the modes tab in Super Whisperer settings and create a new mode. Pick custom from the list. You'll see a field called custom instructions. This is the core of your custom mode and it needs something here. If you leave it blank, the mode will not work properly. Keep in mind that something short and simple may work well depending on the required task. But when it comes to text cleanup or formatting your dictation, I want to give you a few tips that will help improve your outputs.
First, it's good idea to start by setting the role. See how I'm asking for text formatting on the user message specifically in Super Whisper. User message is what the AI gets from the recording. Here's the thing. The AI doesn't actually know it's a dictation tool. And this is a double-edged sword. On one hand, you get a lot of flexibility to use the AI for all sorts of text related tasks way beyond the built-in modes. But you also have to be very clear about what the AI should and should not do.
Second, I suggest you break down your tasks for the AI into clear actionable steps. It's like a checklist. Instead of saying, "Make my text better, be more precise." Here, I'm asking for selfcorrections, automatic list formatting, and a spelling and grammar check. You could ask for emojis, tone change, a specific format. Notice how I am using simple markdown headings like role and instructions to keep things clear for the AI. If you get into more complex instructions later, it's also a good idea to explore XML tags. But for something simple like this, markdown works perfectly.
Third, after telling it what to do, I also want to be clear about what not to do. The AI is designed to be helpful and answer questions and without guidance, it will do just that. Here I'm telling it not to answer or help and not to translate anything. This is useful if you want to dictate in multiple languages and want the output in the same language. I can open a note and dictate. The role is already set. My instructions are clear enough so it should detect. No, it should work correctly. What will happen here? The AI did not fix my self correction. If your instructions are not followed or the AI tries to answer, it just needs a bit more guidance.
Adding examples is essential to get the results you expect. You can add that within the instructions field itself or we also have a specific area to manage this in the sidebar. Let's add a couple. I'll add an example for list auto detection. I'm also including a selfcorrection here. As you can see, for custom modes, where you want the AI to format your dictation instead of answering questions, it's important to add examples specifically for that things the AI will normally try to respond to. So, I will add one for a writing request. Again, with a selfcorrection, bad grammar, and no punctuation. I am trying to cover several instructions in one example. Let's try that dictation again. Everything is set. My instructions are clear enough. So, it should detect. No, it should work correctly. You'll see this time my dictation turns out without issues.
Now, I'll dictate a list. I need to call John, go grocery shopping, pick up Alexandra, and send a meeting report to Jeremy. Okay, the automatic list detection works. Since I have only added two examples, there's still a high chance things will not be totally perfect. But here's another tip. If the AI doesn't behave as you expect, grab that dictation from your history, add it as an example, and next time the AI will know what to do. It's a great way to refine your prompts.
Now, here in the sidebar, you can see that the possibilities expand even more by using application context, selected text, or copy text. These are captured in the same way as they do in super mode. But in custom modes, you can be much more intentional about what you enable and how you instruct the AI to use each of them. Remember how I mentioned referring to your dictation as user message in your instructions? You should do the same thing with these context types. Instead of just turning them on and hoping the AI figures it out, tell it exactly what to do with this one.
Let me show you. I will enable application context. Now I will remove this examples I added. I will go back to the instruction and change this to something else. I want the AI to look at my application context and transform the user message so that it fits naturally. I'm actually giving it permission to change where I dictate in order to make it better and expand. Here I have a short text where I started reading about AI dictation. I'll record something. The useful thing here is that when you start providing more context, the results you get can significantly improve. This allows the AI to move beyond simple voicetoext conversion. The result will be improved to match what I have already written. The point here is that you could be answering an email, commenting on a discussion, and if you set a specific instruction for the AI to use that context, it will adapt to that specific situation.
My advice is that you don't try to make one single custom mode do everything. Trying to cram every instruction into a single mode can sometimes confuse the AI. And always be intentional about the context you decide to include. If you don't need application context, don't turn it on. It's all about giving the AI exactly what it needs to perform the required tasks. Now you're ready to start making your own custom modes.