If you follow AI and NLP news, you may know about the newest development from OpenAI. If you’re scratching your head, read on.
This week’s news of GPT-3 came bursting through the SEO industry. While natural language processing (NLP) isn’t at all new, GPT-3 levels up the game. As an 80+ person agency focused on content marketing, it got us wondering how this will change our industry in the coming years.
Prefer to listen? Download our Content & Conversation podcast episode on GPT-3 instead.
But First, What Is GPT-3?
GPT-3 stands for Generative Pre-training Transformer and is the third iteration from OpenAI. GPT-3 is a version of natural language processing (or NLP). NLP such as GPT-3 and others is a way to build computers that read, decipher and understand human words.
NLP isn’t new. In fact, versions of NLP have been around for many years now, such as Google’s auto suggest in your search bar, email or Google Doc. See an example of that in action:
As you can see above, Gmail autofills portions of my email based on commonly used words and phrases. Keep in mind that GPT-3 is not associated with Gmail or Google, but the basic idea remains the same.
Why Is GPT-3 a Big Deal?
What makes GPT-3 a bigger deal than other previous versions of of GPT (and other NLP tools from other companies summarized below:
Total Parameters
Parameters means the amount of data GPT-3 has to base its predictions off of. The previous iteration of this (GPT-2) worked off of 1.5 billion parameters. This version has 175 billion. Overall, this makes it a much more powerful tool because of the larger dataset it has to work off of.
Task Agnostic
As mentioned here, GPT-3 is task agnostic. Meaning you can create content for any need or platform necessary. As shown in my email example above, that NLP system only functions within Gmail and Google docs. You can’t open it up in a Microsoft Word document or Facebook post and expect the same thing to happen. OpenAI’s GPT-3, though, allows you to apply this content to anywhere.
Minimal Tuning
Another big selling point is how good the final copy actually is. This goes back to the large parameters the tool is referencing. Once given an outline to follow, GPT-3 can produce a pretty solid first draft. In some studies, GPT-3 fooled up to 88% of people when they had to guess if the copy was written by a bot or a human.
If you want a more in-depth descriptions of the points above, I recommend watching this video by Aaron Jack:
Is GPT-3 Free?
Currently, no, and there is a good reason for that. In previous open source models there was concern about what would happen to the web if tools like this were accessible to people looking to do harm. You can read more from OpenAI about how they’re thinking about things like harmful bias and low quality content in their FAQ and the rationale behind a commercial release. For now, OpenAI API is remaining in beta and you can sign up via this waitlist.
What Does This Mean for SEO?
Now that you have a high level overview of GPT-3, time to answer what you really came here for. So, are we all out of a job?
The short answer for us is: no. But the longer answer is: you may want to reconsider what your core skill or service offering is.
For us, our content marketing service extends far beyond just writing blog copy. Our content marketing specialists receive training on SEO best practices, design and UX and outreach in addition to writing. In fact, writing is only 50% or less of a specialist’s role at Siege. We focus on building a set of many skills in order to provide our clients more value than just a blog post.
If 60% or more of your deliverables (either in house or as a freelancer or agency) is writing content, it may be worth expanding your skill set to be a more diverse marketer.
With that, the news of GPT-3 has us strategically thinking of the following:
Create short form content at scale
GPT-3 will really excel for any writing project where you have to write a series of “low” value content at high volume. While no one likes to think about any content of their site being considered “low value,” search engines have made it necessary to include copy on pages that most users will ignore.
Examples of this include:
- Footer copy on product and product category pages
- Meta descriptions for old content
- Alt text and similar missing SEO elements
- Unique product copy for companies with 10,000+ products
Draft evergreen, long form content
Outside of short form, GPT-3 also excels at longer form content (which is another unique feature). If there are topics in your industry that are evergreen and written about consistently (e.g. definition queries, standard processes, etc.) GPT-3 can work off a short outline and provide a working first draft. I say “first draft” because…
Writers become editors
The more sophisticated a tool like GPT-3 becomes, the less of a need there will be for pure “writers.” If tools like this can produce any type of content in a much faster timeline (think: seconds) than a copywriter, then the role of the copywriter will likely change.
To us, we see our industry shifting to more of an editor model where a tool like GPT-3 will get you a good working first draft while you focus on leveling up the content to fit your brand.
If I Use GPT-3, What Should I Be Careful Of?
GPT-3 isn’t a “set it and forget it” solution. Two big areas of concern when using this (and similar NLP tools):
Beware of content “sameness”
Internally, we follow common blog templates and formats in order to scale our production. We also use tools like Ahrefs Keyword Explorer and Clearvoice to help us incorporate appropriate keywords and keyword frequency. But relying solely on existing templates and tools creates a model where it’s easy to mirror what’s already out there (or make it easier for your competitors to copy your model).
For example, check out these two posts by Shopify and The Balance both ranking for “what is a SKU”:
At a quick glance, the headline, subheads and core points are almost identical. It’s likely the team behind both of these articles researched the appropriate keywords to use and the type of content Google is serving for this query in order to match their post against those expectations.
So while we still use these tools, our work doesn’t stop here. The real work goes into leveling up that content to make it unique to the industry. Some of my favorite examples of this include:
- Including unique customer data when answering a “what is” query like CleverTap’s “What are push notifications?” post
- Case studies with quotes from experts like Valpak’s social media guide
- Unique UX like Nextiva’s customer service statistics
All of the above produce a much richer experience for the end user and content that is unique to the SERP.
Biased, racist and sexist language
It’s hard news to swallow, but if NLP is basing its output from what’s inputted—and that language is already biased—then the end result will be the same. Max Woolf addresses this shortcoming in his post reviewing GPT-3.
You may be thinking “my industry isn’t biased so I have nothing to worry about,” but if you’re focused on YMYL topics such as finance, medicine and health & wellness, these tools aren’t meant for sensitivity. In these areas, it’s even more important to carefully edit the output you’re receiving.
Are the Bots Out for Your Job?
GPT-3 is an interesting development for content writers and SEOs. While how quickly this will change the industry is unknown, it’s obvious NLP is only going to grow in importance as the technology becomes more advanced.
If you want to read more about NLP in general, some of my favorite go to’s include:
- Britney Muller and Kevin Indig’s NLP discussion on Tech Bound
- Ruth Burr Reedy’s Moz Whiteboard Friday on NLP and content