Tone of Voice Cloning.

Introduction

Another important thing that came out of interviews with end users was the tone of voice. It would be best if the tone of voice of a company or person could be cloned so that the generated pages are in the same style. This would make it more consistent and make the content feel more like it belogs to the brand. To see the possibilities of achieving this I did experimentation using GPT-4.

Abstract

To try and achieve this cloning of a brand's tone of voice, there weren't many straightforward methods besides giving the AI content written by a person or a company and requesting it to try and keep the same tone in the content it generates.

So I tried to clone the tone of voice of Prominent which is one of the iO clients and interviewees. I visited their website and copied the content from one of their published pages. I then gave this content to GPT-4 and asked it to try and mimic the tone of voice of that content using the prompt: "Try and copy the tone of voice of the following content and use it for the content you generate from now on."

Unfortunately the results were not as expected, and the AI did a bad job of accurately copying the company's tone of voice. Realizing that cloning the tone of voice was not a realistic option, I looked for a logical alternative: predefined tone of voices. So instead of cloning I created a list of predefined tones of voice that the end user can choose from. This approach still gives flexibility in the type of content the AI outputs. The list includes the following tones of voice:

• Formal
• Informal
• Technical
• Humorous
• Persuasive
• Analytical
• Creative

Testing these predefined tones of voice I found that the AI produced very different results depending on which one was selected which makes it a more effective solution.

Conclusion

Unfortunately it wasn't possible to properly clone a tone of voice. However the alternative solution of having predefined tones of voice still gives the user flexibility in the type of content the AI generates. Even though cloning would have been the best, giving the user a variety of tones to choose from that best fits their needs makes sure that the generated content remains adaptable to different contexts.