An Introduction To AI-Powered Content Marketing With GPT-2
The text below could have been part of a blog post, at a first glance it looks insightful and includes proper keywords. It has been created with OpenAI's GPT-2 Model, on my MacBook Pro.
GPT-2 is OpenAI's large language model with 1.5 billion parameters, trained on a dataset of 8 million web pages. The objective of the model is simple, it aims to predict the next word, given all of the previous words within some text.
For our experiment, I decided to give the algorithm a seed phrase such as "Our 3-Step SEO Process for Higher Rankings in 2019. SEO without a strategy can be a deceptively long process, and you might find yourself going in circles.", (borrowed from our previous blog post) and see what the algorithm would come up with. Here is the result:
You might not realize it, but you're setting yourself up for failure.
If you need a more direct route to SEO success, then I've compiled 5 best strategies to bring your rankings up. If you just need a solid start, we can start with this list. I know there are plenty of other top SEO tips out there, and you can always give feedback, but this is my list so far. I'll keep updating this guide.
The Importance of the Big SEO Tools
There are thousands of tools out there, all of which have a significant impact on ranking. But I'm going to focus on one tool that's absolutely crucial – the web site rank engine (SHE). When you're considering a tool to improve your rankings, you want everything to be perfectly aligned.
In short, you want all the best tools.
That said, a few guidelines and suggestions, to make sure that SHE is your tool of choice. As soon as all of these factors align, I'd recommend buying it.
The big three are:
As you can see, the output is not bad at all. It sounds like an informative and natural introduction to the topic. The model might not be able to provide us with content that offers strategic perspective (for now), but sure is capable (for the most part) of filling the gaps with words. The bits and bytes that don't fit the text can be easily edited.
It might take some tweaking until you find the best parameters to initialize the model for your use case (my parameters here are hardly the ideal ones), but overall it's a simple process of trial and error until you get it right.
Is AI-Powered Programmatic Content Marketing Worth The Attention?
Being able to programmatically create large numbers of pages filled with ai-generated content, based on real time SEO data, will have a profound impact on search.
These pages might lack the strategic human perspective, but Google's algorithm isn't human either. It's a model like this one. The technology is out there and it's only a mater of time until someone fill in the gaps.
The process would look something like this:
- Make use of an API to fetch real-time SEO data, such as the Mozscape® API.
- Utilize a machine learning model such as OpenAI's GPT-2, to dynamically create pages with optimized titles and content for a large number of search queries, using the SEO API's data as seed content.
Welcome to the possibly dystopian future of self-published AI-powered websites.
The technology implementation is not perfect yet, the content lacks insight and doesn't sound well written, thus there is a need for editing before publishing it. But when OpenAI releases their larger 1542M model the later will probably no longer be a problem. Our intuition is that that programmatically generated blog posts and other content-oriented pages will do well on search. Programmatic SEO is a thing of the present, mostly used by e-shops and sites like Yelp.
Now imagine PBNs (private blog networks) hooked up to SEO APIs (there are literally hundreds) and neural nets. It's gonna happen sooner than later. The one thing we can be certain of, is that this technology will be exploited and pass under the radar until it becomes mainstream and ai-generated websites that look informative but don't offer insight get penalized by Google.
How To Get Started With AI-Generated Content
First of all, you would need to get the model (OpenAI's GPT-2) running. For that there are differrent options (such as Docker containers), the one I used is forking it directly from Github and configuring a Python environment on my computer to run in from.
Finally, if you just want to try it out fast and without the installation procedure, you can use Talk to Transformer. This is a web app that allows you to play with OpenAI's new machine learning model, but unfortunately without being able to tweak the model's parameters for best results.
At Studio for Digital Growth we help businesses to grow online with a mix of proven digital marketing strategies. If we could help you out on this, we offer free strategic calls without strings attached. See Availability