If you haven’t heard of AI writing tools, and you’re a content marketer or writer, then we need to have a chat.
It seems like every marketer on social media has an opinion about this topic. These opinions range from outright rage and fear of losing their jobs to “holy moly, I can publish 10x the amount of content now.”
The reality is certainly somewhere in the middle.
It’s important for marketers and writers to have a realistic expectation of what these tools can and can’t do in order to use them properly. We are going to see huge improvements over the next few years, so let’s all get some perspective on where we are today.
As the underlying technology is updated and improved, the public sentiment is becoming more and more positive and we are seeing real use cases pop up as time goes by.
Why are there so many AI writing tools?
All AI writing tools are built on top of the same technology, with a few exceptions. The underlying technology went through a transformational (if you’re an AI person, you’ll appreciate how clever I am being here) shift in 2020. OpenAI released its original research in 2018 on “generative pre-trained models” (GPT) which laid the groundwork for the rapid advancement in the field over the next few years.
The ability to create high-quality writing tools that are actually useful quickly went from “possible with a lot of effort” to “very doable with minimal effort.”
These new AI writing tools are popping up everywhere like Starbucks stores in the early 2000’s— Dominique Jackson (@djthewriter) August 4, 2022
Once this technology was generally available to any developer who wanted it, the flood gates opened and an enormous number of almost identical tools hit the market. The companies that had a head start by building on top of this platform during the beta launch had a huge advantage, and they rightly invested heavily in marketing and community to brace for the onslaught of predictable copycats.
Since the AI language generation is now commoditized and most platforms are the same with different front end user interfaces. The winners among the cheap AI writers will be defined by a strong community and a successful marketing effort in most cases (especially the low cost options). However, there is a lot to still be added on top of new AI language platforms to go beyond what is currently available on the market, such as customized AI models for specific use cases.
OpenAI & GPT-3
OpenAI was the first to change the game for AI language generation with its beta launch of GPT-3 and then its transition to general availability.
There have been AI language generation tools available for many years. The original GPT was released by OpenAI in 2018, GPT-2 in 2019, and finally GPT-3 in 2020. There had been other language generation models before GPT, but the critical development of transformer models (I’ll now refer you to the reference in the intro) was an absolute game changer in quality and reliability of these models.
However, it wasn’t until GPT-3 came on to the scene that people other than AI enthusiasts really started to notice.
Here’s a Twitter thread from that time that demonstrated a few early, impressive use cases:
Today the world changed forever.— Ed Leon Klinger (@edleonklinger) July 17, 2020
Why? Because a group of Artificial Intelligence researchers released something called “GPT-3”.
In short: it’s a computer program that can interact like a ridiculously intelligent human.
And it will blow your mind 👇
AI writing tool = GPT-3 + Basic UI
The leap in performance between the GPT-3 predecessors and GPT-3 created a very interesting environment for budding entrepreneurs and developers. AI language generation became a commodity overnight. The generation of realistic, human sounding text was no longer a technical hurdle to overcome. The challenge was now to build an application around it for a market that had a real problem to solve.
Tools that simply provide short-form copywriting suggestions were the first to come along, such as CopyAI and Jasper (formerly Jarvis, formerly conversion.ai). This quickly evolved into the ability to create long-form blog posts and combining the writing with SEO insights and other add-ons.
Automata helps users leverage GPT-3 and competitive intelligence to repurpose marketing content and generate marketing copy to help their content distribution efforts.
What are AI writing tools good at?
AI writing tools are good at a lot of things. They are also bad at a lot of things. Like any other tool that people use for business, it’s important to understand the limitations and have reasonable expectations of what a certain tool can do for you.
Among the many redeeming qualities of these platforms include their ability to understand patterns and the structure and common use of language, their tendency to push writers out of their comfort zone, and using large pieces of text with a lot of content and expanding along a predictable line of thought.
In general, the more text provided to the tools, the better the output will be. Just like the more context given to a freelancer in a content brief, the closer the finished product will be to what you are expecting.
AI writing tools are good at understanding patterns and language
Conceptually, machine learning and AI are very simple. The goal: represent a bunch of data mathematically and do your best to find a pattern within it. When a new, unknown piece of data arrives, you look at the data you have, and try to predict the output that is most likely to match the new input.
For example in the case of language generation, the task is to find the next most likely word in a sequence of words. This is done by analyzing the entire internet and creating a model that describes commonly observed relationships between words and phrases.
This is what AI language generation platforms are very good at: understanding structure and patterns in the way we use language. They can answer the question, what is the most likely word to come next given all the words that have come before it?
AI writing tools are good at pushing you out of your comfort zone
We all get stuck when writing.
We end up writing the same phrases over and over and producing blogs and social posts that sound very much the same. This isn’t all bad, since it’s important to have a well-recognized tone and brand voice, but sometimes it can become stale.
Having an AI language generation partner there with you that understands language that it has analyzed from all across the web is a great way to help you explore perpendicular lines of creativity.
Whether it’s phrasing a statement differently than you normally would or coming up with a clever headline that is outside of your hilarious sense of humor, it’s good to get another voice in the room, and an AI writing assistant can certainly do that for you.
AI writing tools are good at expanding on large pieces of text
The old saying in machine learning is “garbage in” leads to “garbage out.”
If you train your AI models on data that is poorly structured and full of “garbage”, the predictive power of your model will be garbage as well. This is manifested in the “amount of data” that you give your model as well.
If you ask a language generation model to complete this sentence, “I like _______,” the number of possibilities are enormous and it’s unlikely that it will complete the sentence in a way that you find usable for whatever task you are assigned with.
However, if you ask the language generation model to complete the sentence, “I like to eat good food at _____,” it has much more information to understand what you’re looking for. It might finish this sentence with something like “new and exciting restaurants,” or “the same old restaurant every day,” or “the local farmers market.” Some may be appropriate for your task and some might not, but the cross section of possible outputs is much smaller and it’s more likely to yield relevant results to your prompt.
The moral of the story is the more text you provide the language generation model, the more likely the output will be close to what you are seeking.
What are AI writing tools bad at?
Now for the challenges that AI writing tools haven’t quite conquered.
There’s no reason to believe that these tools won’t improve in these areas with more iterations and companies adding layers of complexity on top of the core platforms. But for now, the most common sentiment among marketers and writers is that AI struggles with fact-based writing, writing really long text, and they are nowhere near being able to replace real, smart, human writers.
AI writing tools are bad at being an encyclopedia
AI writing tools catch a lot of heat for putting out factually inaccurate statements.
This really isn’t how these platforms were designed!
In general, we aren’t trying to create a tool that spits back a correct answer to a question, we are trying to model the way language works in the most realistic way possible.
When a tool provides a text suggestions that has a person, place, company, or event that makes no sense, it’s important to view these as a placeholder that needs to be reviewed by the writer.
AI writing tools are bad at writing really long text
Much like myself, AI writing tools tend to ramble.
When you give them enough freedom and let it loose, it often has a hard time keeping a coherent line of thought without going off to a completely unrelated line of text or just repeating the same thing over and over.
However, there are several workarounds employed by AI writing platforms. If we combine short pieces of text and review each one sequentially before generating the next, we can generate a coherent long piece of text. However, the generation of an entire blog post or news article in a single shot is not yet a good use case.
AI writing tools are bad at replacing human writers
It’s pretty clear that AI writing tools are a bit away from being able to replace real writers. The part that writers and marketers are great at is not only just the act of putting words in a page.
They are critical to the process because they have a deep, human understanding of their audience, customers, and potential customers. They understand the ins and outs of the product and niche that they are focusing on. They have a personal knowledge of the problem their company solves through conversations and testimonials from real customers.
This information is not easily transferred to an AI language generation platform. However, at some point this is likely to happen.
Public Sentiment: Extremely Positive and Extremely Negative
Everyone seems to have an opinion about…everything. But more specifically, marketers seem to have strong opinions about AI and its role in marketing, writing, and content creation.
Although there is likely a group of apathetic marketers who are going through their day-to-day life not giving this topic the attention that it may deserve, the most vocal opinions on social media include the enthusiastic champions and stubborn critics. Both extremes have valid points of view, but in reality the correct view of this technology is likely somewhere in the middle.
The AI for Marketing Champions
There are super users and super champions of AI writing tools. They are borderline fanatics and are likely writing the type of content that is a great fit for these types of tools.
Generic SEO content is typically not very technical and usually follows very predictable patterns and has simple language. You just need to read the posts in the Jasper Facebook group to see some real fans of AI writing tools.
The AI for Marketing Reasonable-ists
It’s not all gloom and doom or sunshine and rainbows. There are marketers that have reasonable expectations and review AI writing platforms accordingly.
It’s important to understand that these tools are in fact tools with correct and incorrect ways to use them. If you’re looking for a way to press a single button for a quality blog post, you should keep looking. That blog post is going to be a rambling hot pile of garbage.
AI writing tools like @copy_ai and @heyjasperai are growing very popular.— Jared Smith (@jareditor) August 3, 2022
Used correctly, they can increase a writer’s output from 2000 to 4000 words per day.
As more content is written by AI to rank in Google, the SEO game is about to change in a big way, says @ctwtn pic.twitter.com/MdKUdEScMr
AI writing tools can be great for generating lots of semi-informed ideas quickly, and they might be helpful in evaluating which of those ideas are worth exploring, but they also require a new form of work to be done: determining which of the recommended texts are true or untrue. pic.twitter.com/L9r4KdkkGG— Marshall Kirkpatrick (@marshallk) June 20, 2022
The AI for Marketing Critics
Social media is not short on marketers and writers criticizing AI writing tools that provide factually inaccurate and rambling text. They see the “press a button for a blog post” outputs and make up their minds quickly. This is not unreasonable at all. The ability to produce intellectually stimulating, unique, research-based content is not something AI writing tools can do. If this is what you are looking for, keep looking! Or perhaps wait a few years and check back in to see how the technology has advanced.
An uncomfortable truth for many:— Mike (Niche Twins) (@NicheDown) August 8, 2022
If you just sat down and wrote, instead of wasting all your time messing around with the latest AI content generator, you’d probably be ranking by now.
AI copywriting tools are bad for the creativity and intelligence of our human species. Instead of thinking and writing creatively, writing becomes just another data point. With each new content piece generated by AI, I believe we are loosing one piece of what makes us human.— Marius Schober (@mariusschober) May 10, 2022
AI writing tools are here to stay. However, there are few points that should be kept in mind when considering them as part of your marketing toolset. In general AI writing tools are
- not the perfect fit for every task, but having a clear understanding of their limitations and areas where they excel will allow marketers and writers to use them to the best of their ability.
- not a magic solution that will let you fire every writer on your team.
- not a useless toy to be ignored as we move into the next few years.
- powerful tools that deserve to be considered as critical parts of most content creation pipelines for what they are: tools.