Have you heard of GPT-3? I’ve been seeing a lot of the tech community in Silicon Valley buzzing about it online, it’s the latest innovation in artificial intelligence developed by OpenAI.
One of the companies that’s utilizing this technology in a remarkable way is Copysmith, an Artificial Intelligence-powered startup that's automating the generation of creative copy for marketing. Their system helps businesses create ads for Google, Facebook, Instagram, product taglines and descriptions, all with the help of AI. I sat with Jasmine Wang the founder and CEO of Copysmith to discuss GPT-3 and how they managed to utilize it so remarkably well to automate marketing copywriting.
Q: Tell me about your background? What led you to start Copysmith?
A: Jasmine comes with deep knowledge of Machine Learning (ML), Natural Language Processing (NLP) and AI (Artificial Intelligence). She studied computer science in college and then did NLP research at MILA the Montreal Institute for Learning Algorithms, which is the largest Deep Learning Lab in the world, it's headed by Yoshua Bengio who’s one of the three godfathers of deep learning. Jasmine worked on research at Open AI, the company that developed the GPT-3 model, and she was a research fellow at Microsoft.
Jasmine says that the idea for Copysmith came to her when she started an eCommerce store last year, she saw how difficult it was for an entrepreneur to do copywriting for marketing. She searched and couldn’t find affordable options to automate the process. Finding freelancers or hiring an agency would have been too expensive, and doing it herself was taking too much time “I had to write a product description for every single item I listed, I had to write ads for different products, and Facebook ads and the meta tag for the site and my Instagram captions, it’s so much copywriting”.
Copywriting is the backbone of every website and every marketing campaign, Jasmine says that experiencing those pain points made her think how come this is not solved with NLP (Natural Language Processing)? With her background in AI research, NLP, and her experience with GPT-3, she started working on Copysmith the AI solution to solve that problem.
Jasmine notes that the AI copywriting is designed to also collaborate seamlessly with professional copywriters, to make it easier and faster for them to create high quality copy variations for A/B testing and experimenting.
Q: Tell me more about GPT-3, how does it work in general and within Copysmith?
A: GPT-3 was created by Open AI, a San Francisco based AI research laboratory. It stands for Generative Pre-Trained Transformer 3, it’s an autoregressive language model that uses deep learning to produce human-like text.
Jasmine says “What makes GPT-3 unique is that it utilizes the largest data set that any language model has had access to, it also uses transformer blocks, which is a recent breakthrough in deep learning adding to its increase in performance.”
The way Copysmith is using the GPT-3 API to improve accuracy and relevance for copyright generations is by incorporating user behavior into the process. For example, if a user likes a generation, they understand that for that user, that type of result was good, they then use that information to inform what the AI generates for you in the future.
They also are working on industry specific models to train the AI. For example, industry specific models for fashion, like fast fashion or high fashion which are very different. As they collect more data, they will be able to be more granular for different industries. She always thinks about “how can we improve our machine learning to handle different variants of user input”?
Jasmine says that to create a good AI solution it’s also important to take into consideration the UI and UX, in order to guide the user in giving good information for the AI to generate quality results. The basic information that the user inputs to generate the text is very important, for example if the user’s input has spelling mistakes, it's more likely that the output will also have spelling mistakes.
Copysmith works by selecting the content template you desire to create. There are a variety of options to choose from such as Google ads, Facebook ads, Instagram ads, as well as product descriptions and taglines. Once you select the desired content template, you begin by entering basic information such as your company name, your audience, a short description, a few keywords and the tone you want your creation to have.
Copysmith is designed to be a daily tool in a marketer’s arsenal. Jasmine says “We’re really trying to make Copysmith into a tool that marketers can utilize daily for all of their copywriting needs”.
Q: When it comes to marketing, how can Copysmith help with SEO writing for example?
A: Jasmine says “We are planning on incorporating SEO principles by getting examples of great SEO copywriting, and training our Machine Learning on that.”
“We are also planning to work with clients who hire in-house SEO expert copywriters who will be able to look over articles that we generate and incorporate best practices, that learning will then be added into our ML data set. It’s a type of Machine Learning called Active Learning, where you show the model examples which update the data sets even during the inference process. It means that you can quickly improve the results by learning from pattern matching in human best practices.”
Q: You have great templates in Copysmith for ads on different platforms like Google, Facebook and Instagram, how did you go about creating those templates?
A: Jasmine describes “We have basically found the best examples for different types of ads for each type of copy. We use that as examples for the Machine Learning model.
When it comes to actual performance when running the ads one thing to note is that only Google and Facebook actually know which ads really performed well. So in the future we will have the ability to directly launch the ads to Google, Facebook, Instagram, etc. to understand performance and enhance the AI models further to be able to tell our customers what we expect will work well for them”.
Q: How did you go about launching Copysmith?
A: Jasmine says “launching on Product Hunt has brought in a lot of customers. Since then we doubled and now we are at 3K plus users.”
When it comes to ICP (Ideal Customer Profile) she found that there are a few types of customers. People who run Shopify stores and have hundreds of new products for which they have to hire a copywriter to add a description for every single one of those products.
She says that they also see a lot of love from ad agencies who have to do a high volume of creative work for their clients.
As well as adoption with founders and marketing teams at seed and series A startups, because at the earlier stage you're doing a lot of the content in house or you're maybe contracting to a freelancer, so just being able to generate copy and not having to outsource it or go back and forth on email saves a lot of time and money. “It’s very valuable for them to be super agile and just be able to deploy experiments quickly” Jasmine says.
Q: How did you go about assembling your team? How did you find the right people?
A: Jasmine began by describing the foundation for her team which starts with her bringing the AI expertise and her sister who is an amazing full stack engineer and the CTO of the company. As well as their talented designer who’s leading the design for the entire site and the product, they met their designer through mutual friends.
As for the rest of her team, she said “I think like most things in life, a lot of good things come from warm intros, everyone who has been working with us I’ve either worked with them before in some capacity, or came through a warm introduction”.
Q: What do you think will be the future of Copysmith?
A: Jasmine says “I want Copysmith to be a one-stop-shop system where users can also run ads directly with one click of a button.”
“We prioritize what to build based on user feedback. We have a Slack group of early adopters that has over 100 people in it. And they've been really vocal about the sorts of things that they want to see. We basically iterate on designs and product with those early customers. They jump into design interviews with our designer and really help us test and quickly iterate while building what resonates most.”
I enjoyed the interview with Jasmine and started using Copysmith for AllFactors. It works remarkably well! I was blown away by the creative copy it generated, it’s basically professional copywriting that I probably wouldn’t come up with myself. Super awesome.
To experience Copysmith for yourself visit copysmith.ai, you can get started with a free trial. You can also follow Jasmine and Copysmith on Twitter for the latest and greatest in AI copywriting for marketing @j_asminewang @copysmith_ai