You’ve probably heard about the many ways artificial intelligence (AI) is changing how we work, including how emails are written, tested, and sent. You may be wondering if you can use AI to simplify your workload, and it’s possible – but it’s important to understand how AI tools work and what their current limitations are. In this article we will help you to understand how you can use AI to write emails, but without making mistakes that could damage your professional reputation.
How do AI writing tools work?
Most AI writing tools currently available use a Generative Pre-trained Transformer (GPT) model to generate text, and the tool itself acts as an interface between the GPT model and the user. This means that no matter which tool you use, the quality of the output is going to be similar, and so it will still need to be checked over by a person, to give it personality, and uniqueness, as well as a more natural style.
GPT models take text inputs as parameters, process them, and give text outputs based on the data they were trained on. In this case, training means giving the model an enormous amount of data for it to learn from, which it will use to match its response to the given parameters.
The most common GPT models in use today are GPT-3.5 and GPT-4, both released by OpenAI. GPT-3.5 is the currently released model, with GPT-4 as an updated version currently in a limited beta release.
AI writing tools are designed to streamline the process of using a GPT model to generate text for a specific task - like writing an email. They add the right parameters, so you don’t have to try as many iterations to get the results you want, and many of them are easier to use than ChatGPT’s interface.
However, the market is currently swarmed with tools that promise to write for you with AI, and new ones are being thrown together every day. Almost all of them require a paid subscription, and not all of them are worth the cost. Since this technology development is so new, a lot of people are trying to capitalize on it – and some of these tools don’t do much more than plugging your inputs into ChatGPT.
To figure out if a tool is worth your money, it’s important to understand the problems AI-generated writing can have. This will help you pick the best tool for your needs, so in the next section we will talk through some of the biggest concerns with AI text.
What problems do AI-written emails have?
AI generated text has some major pitfalls to be aware of, especially since it’s relatively new. These can include plagiarism, incorrect information, discriminatory language, poor grammar and spelling, tone mismatches, and more. Letting these problems slide will spread misinformation, damage your professional reputation, and cost you time and money instead of saving it.
Let’s break down why some of these problems happen and how they can be mitigated.
If you want to do a quick experiment, ask ChatGPT to generate a few paragraphs of text and then run it through a couple of plagiarism checkers. A good portion of the results will be flagged. This is because GPT models are essentially taking bits and pieces of the data samples they were trained on and piecing them together, not writing something new.
The quickest way to remove plagiarism from your AI-generated writing is to do something similar to the experiment above. Find a plagiarism checker you like, run your text through it, and edit any snippets it flags. You might want to test multiple checkers since they rely on different sources to see if something was plagiarized or not, and depending on the subject matter, something that shows as all clear on one checker might have a lot of flags on another.
The other way to deal with plagiarism in writing, especially if you have more experience with writing, is to use AI generators for inspiration and do the actual writing yourself. AI-written text tends to be formulaic, so if quality and originality is your biggest priority, it’s usually still much faster to write it yourself.
Incorrect information and discriminatory language
The data ChatGPT is trained on is a massive sample of text scraped from all over the internet. The sample is so large it’s impossible for humans to go through it all and remove pieces that are untrue, illegal, offensive, or otherwise inappropriate. ChatGPT has no idea if something is true or not, it just knows that people have written something, and it was included in its training sample.
ChatGPT has tried to get around the inappropriate content problem by training a smaller model on exclusively inappropriate content so it can identify and flag the worst of it in the larger model. This is one of the major improvements GPT-4 has over GPT-3.5, but it’s not a complete fail-safe.
The best way to take care of this problem in your AI-written emails is to read them over with a careful eye. This has to be more than just scanning the text quickly the way you would with something written by a human. The errors it makes might not stand out in the same way since it’s designed to mimic human writing.
Fact-checking is even more critical than usual. There shouldn’t be an expectation that the results are factually true, even if they often are. Again, most AI text generators just spit out things they have seen written elsewhere, and don’t currently have a method for identifying valuable and reliable sources for information.
Another problem area for AI-generated writing is the quality. This includes things like grammar and spelling choices, tone mismatches, structural problems, and a general loss of nuance in writing.
Correct grammar, spelling, and word choice depends on who your audience is. When writing emails, you generally want to match the conventions your email uses to the people reading it. Some conventions are dependent on geographic location, education, topic, and other factors, and these differences aren’t always reflected in AI writing.
In addition, AI-written emails are going to sound formulaic to an extent no matter the prompts you put in. They can be a little bland. They repeat sentence structures. Their outputs are geared towards the most general audience possible. There are details that go missing, they skip strategies that make writing pop. This can be more subtle, but it’s something to keep in mind as AI tools continue to grow – people will learn to recognize what AI writing looks like.
The easiest way to fix both of these problems is to edit and proofread your emails. A quick edit can improve the quality of the writing, and proofreading can help you tailor your email to your audience.
The next section will go over some tools that might be useful for your needs, and also how to use ChatGPT to generate email text directly.
How do I use AI to write emails?
When looking for AI help with writing emails, it’s important to determine what kind of help you’re looking for so you can pick the best tools. For example, if you want a tool that generates text, you might want to find one that incorporates a plagiarism checker as well to save you some time.
We’ll go over some different tools and options below.
Using ChatGPT to write emails
Let’s talk about the most basic option, which is simply using ChatGPT to generate the written portions of your email without any bells and whistles. You’ll need an account with OpenAI, which has a limited amount of free uses. After that, you’ll have to pay for more access to ChatGPT. In addition, if you want access to GPT-4, you’ll need to get on the waitlist for it.
Once you have an account, you can type in some prompts to the interface asking it to write an email about your topic. You might find it helpful to specify the email’s audience and the tone you would like it written in. An example would be something like: “Write an email about Mailosaur’s market leading email testing software geared towards a general audience with a casual tone.” Then, ChatGPT will generate the text. You can ask for a different result if you’re not happy, but if you are, you can fact-check the text if necessary, add it to your email template of choice, proofread it, and test it.
Grammarly is renowned for helping with grammar and editing, but they’ve developed a generative text assistant as well called GrammarlyGO. This tool has access to 100 prompts a month for free, and paid plans for power users and businesses. You’ll need to set up a Grammarly account if you don’t have one.
GrammarlyGO has an option to set your preferred voice and tone, and also makes it easy to pick and choose between tones. It works in a similar way to ChatGPT, but the interface is more user-friendly. The biggest advantage is how easy it is to refine the generated text results. Once you’ve asked it to produce the message you want, there’s options to simplify the text, shorten it, change the tone, and more that only take a single click of a button.
Jasper is one of the oldest AI writing tools out there, so it’s more polished and reliable than a lot of the more recent tools developed. It combines outputs from GPT-4 with a couple other models, including one developed by Google. It’s mostly geared towards businesses and marketers and is priced accordingly starting at $49 a month for their lowest tier plan. They offer a 7 day free trial for you to test out their tools and see if they work for you.
Get help testing emails written by AI
Mailosaur has software tools and experts on testing email and SMS functionality so you can be sure your emails are working how they’re supposed to, regardless of if they were written with AI or not. Reach out to us today with any questions you have about email testing, we’re happy to help!