GPT-3

GPT-3 And Beyond: How It Affects Content Publishers

Note: Listen to this post instead using the audio player below, and consider subscribing on your favorite podcast player!
[spp-player optin=”off” ctabuttons=”on” poweredby=”off” url=”http://api.goshortwave.com/shows/OVS74gI6VsijB2Ri/episodes/W7KWa8aSldklvaM1ZSY5l3KjUa78uFDm/play.mp3″]

GPT-3 may bring the single biggest shift to content publishing, since the advent of the internet.

Yet while this technology may radically change the publishing and SEO landscape, it also comes along with a huge host of benefits – benefits that you can take advantage of today.

Although the general public doesn’t know it yet, this is a transformative technology that will have far-reaching effects for everyone.

The most exciting part?

It’s just getting started.

What Is GPT-3?

If you haven’t heard of it before, you may be wondering – what the heck am I talking about? What is GPT-3, and why is it so important?

GPT-3 is an AI language model that was developed and released by a company called OpenAI. Essentially, it is a form of artificial intelligence that is capable of generating realistic text in an extremely powerful manner.

These types of language models have been around for a few years now. However, the models were a lot smaller, making them far less realistic, far less logical, and far less believable.

With GPT-3, that is no longer the case. As you’ll see later in this article, GPT-3 is fully capable of generating text that is indistinguishable from human-written text.

It is even able to pick up things like context, the writer’s tone and style, and the unique mannerisms of the person using it.

We’ll go into the practical applications of this for marketers in a minute. First, let me address everyone’s biggest concern.

Note: If you’ve published content online or participated on forums like Reddit, you helped to provide some of the data that GPT-3 was trained on! Isn’t that interesting?

GPT-3 Won’t Replace Writers… At Least Not Those Writing Articles

This is the thing that everybody’s talking about. If GPT-3 is here, will there be a need to continue writing articles? Won’t those with access to GPT-3 just generate thousands of them, leaving everyone else out of the water?

The answer is no, and there’s a few reasons why.

First things first, Google specifically addresses auto-generated content in their guidelines, stating that it’s not allowed and may be penalized. This alone is enough to avoid spitting out thousands of GPT-3 written articles.

The exact wording is this:

Automatically generated—or “auto-generated”—content is content that’s been generated programmatically. In cases where it is intended to manipulate search rankings and not help users, Google may take actions on such content. These include, but are not limited to:

  • Text generated through automated processes, such as Markov chains

It is believed that Google will be able to detect content written by GPT-3, if they aren’t able to already. Therefore, this is a very real risk for anybody hoping to use GPT-3 to replace content writers.

That being said, it is also unknown how Google’s position will change on this in the future, and they haven’t spoken about GPT-3 specifically. There is little reason to keep high-quality content out of the search results, even if it was in fact written (partially or in full) by NLP models like GPT-3.

Google’s primary concern is about the quality of the search results. They may end up only penalizing content that is low-quality or otherwise does not provide value to users.

An Example Article Written By GPT-3

This is not the only reason why GPT-3 won’t replace article writers. To further illustrate this point, let’s look at an example mini-article generated by GPT-3. The content in bold is what I wrote in order to prompt GPT-3, and everything in italics is what GPT-3 generated for me.

Short Article On SEO Best Practices:

In today’s day and age, SEO is more important than ever.

If your business fails to build a solid SEO strategy, you will fall behind and your target audience will not be able to find you. In this article, let’s look over some of the best SEO Practices in today’s day and age.

1. CONTENT IS KING!

I’m sure you’ve heard this many times before, but it’s true. Your content is the most important aspect of your SEO strategy. Building a solid website with high-quality content is the first step to SEO success. You have to be able to write content that is engaging, relevant and most of all, original. Blogging is an awesome way to improve your content quality.

2. LINK BUILDING A lot of people make the mistake of thinking that link building is dead, but it’s really not. Link Building is still a very important aspect of SEO. Not only does it help you improve your search engine rankings, it also improves your reputation online. It will also help enhance your visibility in the search engines as well as improve your website’s domain authority.

3. ON PAGE OPTIMIZATION On page optimization is another very important part of SEO. You need to be able to optimize your title tags and meta descriptions. You also need to be able to include relevant keywords in your main blog content. In addition to this, you will need to have a good internal linking strategy as well as strong navigation elements that will help keep your visitors on your site longer.

4. SOCIAL MEDIA MARKETING Social Media Marketing is another very important aspect of SEO today. Social media platforms are very powerful and if used in the right way can have a huge impact on your search engine rankings. The key here is to focus on building up as many followers as you can on social media platforms such as Facebook, Twitter, and Pinterest etc.. Make sure to post interesting and relevant content on a consistent basis in order to attract more followers and make them loyal customers for life!

If you would like to learn more about the best SEO practices in today’s day and age, please give us a call!

For the sake of the example, I used the first output that GPT-3 gave me. It is certainly capable of even higher quality content, and I could have asked it to regenerate the article several times until I was satisfied. We’ll look at even better examples of GPT-3 later.

Anyway, let’s address some of the concerns that I had.

Inaccuracies

The first and perhaps the most important issue is that certain parts of the text are simply not accurate. Most notably, the 4th point GPT-3 made about social media (which doesn’t help with SEO at all, yet GPT-3 claims is ‘very important’.)

If you are truly an expert in your field, this wouldn’t be as big of a deal. You would be able to recognize that the content is factually incorrect, and make adjustments accordingly.

If you weren’t an expert however, then you wouldn’t know. You would publish factually incorrect information, jeopardizing your blog’s credibility and misleading your audience. In my opinion, this is unacceptable and immediately makes GPT-3 unusable for large-scale content production. Content still needs to be edited, audited and verified prior to publishing.

This is particularly a concern for YMYL (your money or  your life) where incorrect information can have serious consequences on someone’s life.

An important note to keep in mind is that GPT-3’s data that it was trained off was cut off in October of 2019. Although ongoing updates to the training data are planned, this means that GPT-3 does not have any knowledge of events or changes that happened after October 2019. If you are working in an industry that changes quickly, language models in general are likely to struggle to remain accurate more frqeuently.

Limited Context

When generating with GPT-3, you are limited to 2048 ‘tokens’, with one token equating to about 4 characters on average.

This limitation is made up of both the prompt / training examples you give it, as well as the text it generates out.

What this means is that GPT-3 only has a limited memory. It works best when you can give it an example of the style of writing that you’re going for, but this limits how much it’s able to output.

If you’re publishing long articles, GPT-3 may not be able to generate the full article at once. Even if it does, it may go off track and transition to other related topics while doing so. See this example (click here to enlarge):

That being said, there’s no reason you couldn’t use GPT-3 to generate articles section-by-section. However, you would need to manually tie them together yourself.

Biased Output

GPT-3 was trained on data that was published online, by the world as a whole. Just like groups of humans tends to have biases, some of these have worked themselves into GPT-3 as well.

Outside of racism, sexism and other similar issues, the AI outputs tend to be quite repetitive surrounding the same topics. Although it may phrase things differently each time that you generate text, the AI will often default to many similar sub-topics for the same prompt.

This is true even if you change GPT-3’s ‘temperature’ settings, which affects the level of randomness in the output.

Lack Of Originality, Unique Expertise, Or Personal Connection

GPT-3 has an incredible amount of general knowledge, and allows people to instantly tap into research that would normally take hours to conduct.

However, part of what makes certain writers great is their ability to take conceptual ideas, unique expertise or interesting personal stories and weave them into the content.

Interestingly enough, GPT-3 is perfectly capable of generating fake stories to demonstrate a point. That being said, it’s just not the same as learning from a unique person who has actually lived what they’re talking about.

With certain types of content, this may not matter. For other types of content, it matters a lot.

Closing Thoughts On Article Generation

The more general and less personal the topic is, the better GPT-3 handles article generation.

While I would not ever publish GPT-3 written articles without first editing and revising them, that isn’t to say that GPT-3 does not have its uses.

GPT-3 can drastically cut down on research time as well as idea generation. It can give you the direction you may wish to take for your article, and even fledge out the main points you’ll want to address.

As a writing assistant rather than a writing replacement, it is a fantastic tool – one that I’ll certainly be paying for!

GPT-3 And Comment Generation

Let’s move on to the next topic I wanted to test out – comment generation.

As a blogger, I leave lots of comments on people’s articles. Blog commenting is a great way to generate traffic to your website, but it is even better as a networking and relationship building tool.

While I personally wouldn’t replace my comment writing with GPT-3 (due to the personal nature of it), it is still something I wanted to try out. And unsurprisingly, GPT-3 does very well with it!

Let’s look over a few example. Again, the bold is the prompt I gave it, and the italics is the generated output:

Comment on article about speeding up WordPress websites:

Hi! Wow, what a great list of tips here. My website has really gotten bogged down as I’ve added more plugins to it (needed to host my paid training) so I’ve been looking for ways to speed it up.

I’ve implemented a few of these already, but I still have some room to go further.
– James McAllister

Comment on article about internal linking and SEO:

Internal links can make a big difference and it’s good to get into a habit of interlinking posts as you write them.

I recently discovered a plugin called Link Whisper which can scan articles and provide suggestions, it’s saved a ton of time and I used it to generate over 800 new internal links with my sites!
– James McAllister

Comment on article about using Pinterest for traffic:

I’m so glad this article was written!

A lot of people are missing out on Pinterest and I don’t think they understand how powerful it can be for them.

I don’t know why I never thought to check out Pinterest before, but I did and it’s been amazing for my traffic and sales.

– James McAllister

Comment on article about using infographics to build backlinks:

I’ve actually been using infographics for a couple years now and it’s been a big help in boosting traffic to my blogs.

The only problem is that I don’t have the time or resources to create my own. I’ve been trying to use free ones and finding that it’s hard to find one that’s good and high quality.

I’m going to look into this service for sure. Thanks for the great tip! – James McAllister

 

Comment on article about the importance of morning routines:

I’ve been following a morning routine for a long time now and it’s made a huge difference in my productivity. I used to be the kind of person who would sleep until noon on weekends and then be tired all day, but now I get up early every day and feel great!

As you can see, these outputs are not bad at all! If they arrived on my site, I would have no idea that they were auto-generated and I would have approved them.

This brings up a huge concern for large-scale spam, which I’ll address more later.

GPT-3 And Copywriting

One thing that really interested me, was the ability for GPT-3 to write product descriptions for eCommerce stores.

Stores with large catalogs may have thousands of different SKUs, and every additional minute writing product descriptions really adds up. In this example, I provided GPT-3 with one product description, and asked it to generate another. The bold is what I provided, and the italics is what GPT-3 generated:

Product description for a mug about a hardworking wife and mother:

‘Super Mom, Super Wife, Super Tired’ Mug from Kinacle

It’s no secret that mothers are superheroes. Keeping up with everything being both a mother and a wife entails is hard work, and leaves you exhausted at the end of the day.

With the ‘Super Mom, Super Wife, Super Tired’ cup, you can display this message proudly – you are a woman who does it all.

Who Is The ‘Super Mom, Super Wife, Super Tired’ Suited For?

Women who are both a wife and a mother.
Mamas who need to make it clear how much work they’re doing.
Moms who are proud of everything they manage to get done each day (seriously, you rock!)
And also makes a great gift for…

Stay-at-home mothers.
Wives of husbands who recognize everything they do each day.
Friends you don’t get to see enough because they end each day exhausted

Product description for a mug about a mother needing more coffee to get through the day:

‘I Need More Coffee’ Mug from Kinacle

It’s no secret that moms need coffee to get through the day. In fact, if you’re a mom, you probably need coffee just to get out of bed in the morning.

With the ‘I Need More Coffee’ mug, you can proudly display this message…and ask for a refill while you’re at it.

Who Is The ‘I Need More Coffee’ Suited For?

Moms who need coffee to get through the day.
Lovers of mugs with funny, witty sayings.
Mothers who love their morning cup of coffee.
And also makes a great gift for…

Mothers who are always up early and need their cup of joe to get going.
Mothers who are always up late and need a cup of coffee to unwind.
Mothers who are always on the go and need a caffeine boost.
Friends who are always over at your house drinking your coffee. (This is fine because they’re friends, right?)

GPT-3 did surprisingly well, especially since the original product description wasn’t the best. Notice how it mirrored the exact same style as the original, understood the context behind the product, and generated a description to fit it.

That being said, GPT-3 is most useful for very large stores, where products would otherwise have no description at all. For more expensive products or actual advertisements, a human copywriter is still necessary – missing even one objection or concern can be detrimental to sales.

In this instance, GPT-3 should be a writing assistant, not a writing replacement.

GPT-3 And Meta Descriptions

This is another useful little thing that GPT-3 can take care of. While you may need to generate these multiple times in order to get one that you want to use, it only takes a few seconds. Here is an example – with my input in bold, and the generated response in italics.

Meta description on an article about mistakes to avoid when purchasing a domain name:

If you’re about to purchase a new domain name, you’ll want to be sure to avoid these common mistakes. Let’s look over some domain name best practices!

Meta description on an article about marketing tools:

The right marketing tools can help you grow your business even faster. Let’s look over some of the best digital marketing tools in 2021!

Meta description on traits of the successful:

All successful people have these same 10 traits in common. Develop these 10 traits, and success will surely be in your future.

Meta description on an article about AI generated content and its implications:

Artificial intelligence is becoming an increasingly important part of our lives. Here’s a look at how AI will affect content and marketing in 2021.

It is possible to use your post titles instead of the topic of the article, but I have found that leading with the topic gives better output.

GPT-3 And Email Marketing

Out of everything I’ve tried GPT-3 for so far, this one has been my favorite.

I have already used GPT-3 to help generate over 100 different email marketing messages that I will send out over the coming months or years. These tend to follow a similar format and run between 150 and 400 words.

If you read about the token limit earlier, this length is enough for GPT-3 both to learn from my training examples, as well as provide lengthy output.

Let’s look over an example. The bold is the input I gave it, and the italics is the generated output:

Email promo for an article about speeding up your website:

Hey there.

Look, I did something careless.

But I have a good reason to forgive myself – most people do the same thing.

I let my website get slow. Very slow.

Worst of all? Like most website owners, I didn’t even realize it was happening.

The truth is, unless you’re taking steps to combat it, most websites get slower as time goes on – and Google’s paying attention.

See, Google’s entire goal is to provide the best experience as possible for their searchers. One of the factors they use to determine website rankings is how quickly your website loads.

Therefore, if Google thinks your website is slower than it should be, you have a problem. Fortunately, it doesn’t take long to fix, nor does it cost any money.

I just published a new article sharing the five steps you can take to help your website load up to 3x faster. You can view it by clicking here.

Implement even one of these, and your users are sure to thank you. Implement all of them, and the difference will be enormous.

Hope to see you there!
– James McAllister

P.S: While your users will notice the improvements straight away, it can take a few weeks for Google rankings to improve.

 

Email promo for an article on why you need to develop profitable systems to scale your company:

Hey there…

They say in business, every step you take either requires you to invest money, or to invest time.

The key difference? One of those is scaleable, and one of those is not.

Can you guess which one it is?

If you said money, you’d be right.

All of us will only ever have 24 hours in a day to invest. Money on the other hand can be scaled infinitely.

Companies go from making nothing, to pulling in millions a year solely based on their ability to profitably reinvest money.

After all, if you can create a system that returns $2 for every dollar you invest, then the only thing limiting you is how quickly you can pour more money into it.

What do you do when money is tight, though? How do you get started on this when you don’t already have money to throw around?

I share my best advice in this post, which you can view by clicking here.

Remember, it doesn’t take a lot to get this working for you – but it does require you to be smart with the resources that you have.

See you there!
– James McAllister

Email promo for an article about how blogging can benefit your small business:

Hey there,

I get it. You’re busy.

You’re working long hours, and you have a family to take care of. You have so much on your plate that it feels like you barely have any time for yourself, let alone anything else.

Blogging is probably the last thing on your mind.

But here’s the thing – if you’ve got a small business, you should be blogging. Not just because it helps you rank higher in Google (though that’s a big part of it) – but because blogging is one of the most effective ways to build your brand, connect with your customers, and grow your business.

I just published a new post on my site sharing three reasons why blogging is so important for small businesses, and how you can get started right away. You can read it by clicking here.

Don’t worry – even if you don’t consider yourself a writer, it doesn’t matter. The truth is that anyone can blog, and anyone can benefit from doing so. Even if you only have five minutes a day to spare, you can put that to good use and make some progress!

Hope to see you there!

– James McAllister

 

Here’s what’s amazing to me – GPT-3 copied the formula almost exactly. The generated output sounds like it was written by me. It gracefully transitioned to the promo for the article, and concluded the email in a natural manner as well.

While most emails it generates will need to be edited or revised slightly, this one could be sent as-is. That’s pretty incredible.

Even when asking to generate emails on much more specific topics, GPT-3 handles it without problems.

Because this content is not public (and thus at risk of Google penalizing it), GPT-3 makes an excellent tool for email marketers, and can cut out as much as 80% of the time spent writing emails.

GPT-3 And Email Marketing – Drawbacks

The only issue I’ve experienced with using GPT-3 is that the style is highly dependent on the training data that you provide it. Garbage in, garbage out as they say.

Additionally, the emails lack variety, with the generated outputs all appearing similar in cadence, length, and style. This can be helped slightly by changing the command you give to GPT-3, though.

For example, the command above was , ‘Email promo for an article about how blogging can benefit your small business’. This could instead be written as ‘300-word email promo for an article about how blogging can benefit your small business, written in a fun, playful, yet descriptive style.’ This would yield a different output.

If using GPT-3 for ghostwriting (you want it to sound like you), then heavy use of examples is ideal.

The only other area where GPT-3 falls short is on traditional newsletter style emails. GPT-3 knows nothing about current events (since they’ve occurred after the model was trained.) It also doesn’t know the inner workings of your company, your deals or promotions, or anything specific to your business.

GPT-3: The Risks You Need To Be Aware Of

GPT-3 is one of the most powerful tools ever created – and more powerful language models are already being built.

Let’s look over some of the negative use cases that might pop up in the near future.

Article And Comment Spam

Anytime a tool can be used to generate large amounts of content, you just know that it’s going to be used for spamming.

The fact that we finally have something that can generate believable content from scratch makes this a very real concern.

As shown with the comment examples earlier, most webmasters will have no idea that their comments are being generated with GPT-3. In fact, a Reddit user was able to post for over a week using nothing but GPT-3 written content, and people were none-the-wiser.

The good news is that OpenAI plans to shut down any user breaking their terms of service, which includes issues like spam, harassment and astroturfing. However, a lot of damage can still be done before OpenAI ever catches on.

PBNs and Link Farms

While AI-generated content may not be ready for publication on your company’s blog, it will still be used (to much success) for private blog networks and link farms.

These sites are designed for the sole sake of providing backlinks to ‘real websites’. They don’t care if nobody reads their content – they are simply trying to appear legitimate to Google.

Currently, these sites tend to use cheap, bottom-of-the-barrel content – much of which is rewritten or spun from other websites.

If Google isn’t yet discounting GPT-3 written content, then it is a perfect tool to generate content for PBNs quickly – lowering the barrier of entry for people to start one.

Link Sellers

Although I tend to advocate against the purchasing of backlinks, there is even more to be concerned about now.

It now requires much less effort for people to buy expired domains with high domain authority, crank out some GPT-3 articles, and begin selling links to unsuspecting marketers.

Because the sites have high domain authority and appear to be legitimate with their relatively quality content, marketers may have trouble distinguishing legitimate sites from fake ones – leading to wasted money.

Always check these sites with tools like the Wayback Machine to see what the site used to be like – if it looked completely different last year, run!

Guest Posting / Freelance Writers

This will be more of a concern if / when Google discounts the value of GPT-3 written content.

If you are someone that is accepting guest posts on your blog, you will not be able to tell which content was written partially or in full by GPT-3. You would have to take the person’s word for it, and trust that they actually wrote the content themselves.

The same goes for freelance writers. Unless you’re using time tracking tools that take screenshots of their computer, who’s to say they simply didn’t put the prompt in GPT-3, and returned the article to you?

This is why I prefer working with full-time writers from the Philippines, whenever I can. You have much more control over the entire process, and the content quality is usually much greater too.

Click here if you want to learn more about how to outsource content writing.

Fake Review Farms

GPT-3 does a great job at writing realistic reviews in a variety of styles, which brings in a risk of malicious use.

eCommerce is already a space where there are major trust and credibility issues, and this may only worsen things in the years to come.

Because product reviews are such a narrow type of writing, existing smaller models like GPT-2 XL can be finetuned to write product reviews on its own. Although GPT-2 was originally released by OpenAI, anybody is able to finetune it for their specific use case. This means that OpenAI cannot shut off access to the model or prevent its misuse.

Conclusion

GPT-3 is extremely interesting, and I’m very excited about the future of this technology. I truly believe it will be one of those rare advancements that actually changes the world.

Larger and more powerful language models are already being built, and it’s only going to get better as time goes on.

Of course, there are more practical use cases than just those mentioned here, and I’ll be sure to update this article as time goes on. If you have anything that you’d like me to test out, please leave a comment and let me know.

I’m very interested in hearing your thoughts and ideas. And if there are any questions that I can answer for you about GPT-3, please don’t hesitate to ask.

To your success,

– James McAllister

GPT-3 Infographic

To summarize, here is an infographic about the GPT-3 language model from OpenAI. Webmasters, feel free to use this on your site (just please give credit!)

GPT-3 Infographic, OpenAI Language Model


Comments

8 responses to “GPT-3 And Beyond: How It Affects Content Publishers”

  1. There is always a limitation to using AI to help us to our job. LIke you have mentioned, they might be repetitive and the output might not be human-like after all.

    1. Right – although this area is rapidly advancing and it’ll be interesting to see how it is just a few years from now. Far more powerful language models are already being built, which is pretty incredible!

  2. Hi James

    I haven't heard of GPT-3 until this very moment. I can see a good use for it with emails because those have to be written all the time. But I wouldn't depend on only that, I would use it sparingly. Other than that, I rather read an article that has some personality in it and from what I understand the GPT-3 doesn't have that. I wouldn't use it myself for any article.

    I can see the use of it for those who are writing context about information purposes. That kind of writing can be cut and dry. It's like when I use some PLR's for an email…I don't copy and past, but rather take a few paragraphs and re-write it myself. That's how I would see this working.

    -Donna

    1. Hi Donna! Good to see you again.

      I think you’d be surprised with how powerful GPT-3 can be – it can certainly write engaging pieces with personality, even making up fake stories or jokes to demonstrate points. Of course, they aren’t real, and I’m not a fan of that sort of inauthenticity… but if you can provide it examples or prompt it a bit, it has no problem with humor, excitement or even irony – which are considered to be very complicated.

      You’re right though, looking it as a tool to be rewritten or edited (like PLR) is probably where it does the best!

  3. Finally, someone wrote an interesting post on GPT-3. I have seen too many new SAAS around it. It's an attractive tool for AD Copy ideas and blog post outline ideas. Still, I believe many unsuspecting buyers will be looking at them as a "Do-It-All" content writer.

    High usage of GPT-3 tools might eventually nudge Google to roll out an update that will affect thousands and millions of websites worldwide.

    Thanks, James, for writing this excellent post.

    1. Haha, no kidding! They’re popping up like crazy and honestly the thing that bugs me the most is none of them focus on the downsides. I think the ‘do-it-all’ expectations are largely due to the deceptive marketing from some of these SaaS companies.

      GPT-3 is incredible, no doubt about it. But it’s not a magic solution, you can’t just push a button and spit out a perfect article ready for publishing.

      That being said, I have trialed and seen good use cases for some of them. If OpenAI didn’t give me direct API access, I’d probably have invested in one of the services available today.

      You actually bring up an interesting point, about Google. Imagine if they suddenly started penalizing GPT-3 content – some people would surely have a hard time remembering which content they generated, and which they actually wrote!

      Anyway, thank you for reading and for your valuable comment. Much appreciated!

  4. This is crazy. I can see how GPT-3 potential lies. May I please share with my students?

Leave a Reply

Your email address will not be published. Required fields are marked *