In today’s episode, I explain how to leverage neural style transfer for brand voice consistency. When training multiple AI personas like “boss” and “conference,” drift can occur. Condense distinctive writing styles into prompts with second-person imperatives. Join me to see how generating a centralized style guide maintains tone across all your AI content.
Can’t see anything? Watch it on YouTube here.
Listen to the audio here:
- Got a question for You Ask, I'll Answer? Submit it here!
- Subscribe to my weekly newsletter for more useful marketing tips.
- Subscribe to Inbox Insights, the Trust Insights newsletter for weekly fresh takes and data.
- Find older episodes of You Ask, I Answer on my YouTube channel.
- Need help with your company's data and analytics? Let me know!
- Join my free Slack group for marketers interested in analytics!
What follows is an AI-generated transcript. The transcript may contain errors and is not a substitute for watching the video.
Today’s episode of you ask I answer was recorded in front of a live studio audience at the digital now conference in Denver, Colorado, in November 2023.
The session title was appropriately you ask I answer live generative AI q&a.
Yes, this gentleman here and then you it.
Do you think help create sentiment consensus and sentiment around issues, for example, return to office and and trying to serve him one that’s an issue that our staff are thinking about.
They feel disavow.
What is, you know, around connecting the community, we have all that information what people posting in our in our organization and what I’d like to be able to determine what our members are talking about and are concerned about before they.
Yeah, all the language models can do that very, very easily.
So though there’s two ways of handling it, you could just do an aggregate.
So for something like Claude, you would load assuming that you’ve done the identified, you take out personal identifying information, you would write a prompt along the lines of, you know, here’s however many hundreds of responses.
I want you to do an aggregate scoring of sentiment minus 10 to plus 10 and then give me a roundup or secondarily, if you want precision, you would take that data set, you would write a prompt just to identify, just to solve for sentiment.
You’d say to a language model, your task is to identify sentiment of this passage and then you wrap some code around that process, the data set one response at a time and build a table of the sentiment scores and then you can perform, you know, load that in Excel and say on average, the sentiment was this and if you have other data that you can do quantitative stuff.
I would personally lean towards the latter, but I also code a lot so I would be comfortable with that outcome.
For the not as fancy version, let me see if I have one in here that, I don’t have anything that isn’t customer data, but I would say I would take a good selection of maybe 500 responses sampled, put it in a text file, feed it to Claude and say, give me percentages of sentiment, positive, negative, neutral, etc.
and then identify the top 10 topics within this conversation cluster and it will do that.
This is very specific to Chai GPT-4.
Since what I intend to use, we call her Gertrude, she’s super fiery and easy to train.
I have a photo if you want to see it.
Anyway, in training it, we’ve kind of gotten into a habit of having several different like chats going so like I have one that’s trained to sound like my boss so because I do a lot of ghost writing for him.
So I’ll put it in there and see how I make this sound like him and I’ve fed it things so I’ve got like my Chris chat, like a main chat over here and then up here like I have something about particular type of educational offering that we have and once for our conference.
Is that in maintaining like brand consistency and teaching it about our brand and our association? It’s not going to span through each of those individual chats, I don’t think.
But is there like a better way to do that instead of just having like all of these disparate chats going on? Or should we be building all online, all in one stream? It makes sense.
I get what you’re saying.
What you’ll run into is you’ll run into context window issues.
Yeah, you’ll start running to drift where it forgets stuff.
So what I would suggest doing there is what’s called neural style transfer and essentially it is to build a prompt.
If I have a neural style transfer prompt handy.
I should probably sort this.
There we go.
Neural style transfer.
So I have here it says you will act as a literary expert.
You’re an expert in style transfer neural style transfer writing.
Your first task is to read the following text and learn the author’s writing style.
Read the text then describe the author’s writing style in a bullet point format appropriate for use in a large language model prompt.
Use a second person imperative tone of voice and style.
So if I take let’s take a recent email.
Let’s take this one here.
Go back into my style prompts.
I’ll paste this in here and I’m going to copy that whole chunk.
Let’s do this.
ChatGPT looks like it’s having a bad day.
So we’ll just paste this in here into Claude.
So this is taking my CEO’s writing style and now it is creating essentially a second person imperative which is a writing prompt.
So now you would say you might do 10 or 15 or 20 pieces of content that your CEO or your boss writes in one big chunk because Claude can handle lots of large documents and say write me a writing style for this.
And the next time you go to use this you would say go back to my writing prompt here.
You will write with the following style.
You paste your bullet point list in and now it’s going to replicate that writing style without you having to remind it all the time.
It’s a great way to condense down that style into specific commands.
So here use an opening hook to draw the reader in establish credibility by mentioning experience and expertise.
Make your point clearly and directly.
Don’t dance around the main message invite feedback and input from the reader.
That’s all my CEO style.
That’s how she writes.
If you enjoyed this video, please hit the like button.
Subscribe to my channel if you haven’t already.
And if you want to know when new videos are available, hit the bell button to be notified as soon as new content is live.
You might also enjoy:
- B2B Email Marketers: Stop Blocking Personal Emails
- Is Social Listening Useful?
- Marketing Data Science: Introduction to Data Blending
- What Is The Difference Between Analysis and Insight?
- How I Think About NFTs
Want to read more like this from Christopher Penn? Get updates here:
Get your copy of AI For Marketers