Text generative AIs are an exciting technology for content professionals. They can remove some of the drudgery associated with creating editorial pieces and copy, accelerate the creation of effective content, and assist in writing research pieces.
But how do you get the most out of text generative AIs?
Regardless of what technology you’re using, it’s not enough to make content. You have to make effective content. This goal can be difficult enough even before adding a technology as novel as generative AI.
We have tested several generative AIs in the past. Using what we’ve learned, we’re here to teach you how to squeeze every drop of efficiency and quality from a text generative AI to ensure your content stands out.
Text generative AIs are an emerging technology that has generated much buzz amongst content professionals. Essentially, they are computer algorithms trained on huge amounts of text that reproduce what they’ve been trained on. Text generative AIs can’t be “inspired” like human writers can, so what they can create and how they create it depends entirely on the materials used in their training.
This is why organizations like OpenAI trained Chat GPT with enormous libraries of written text. Chat GPT needed thousands, if not millions, of examples to draw on to be versatile and effective. Generative AIs can’t produce “original” content; they can only mimic what they’ve seen done.
Generative AIs can produce responses thanks to machine learning tools which train a computer to read, comprehend, and write language like a human does. The advancements in machine learning tools enable text generative AIs to handle a vast array of content processes and can provide value for more than just your dedicated content team.
A generative AI tool is like coffee, tea, or wine; you usually have to be mature enough to appreciate it. Before considering scenarios fit for a generative AI, ask yourself whether you’re ready to incorporate one into your content operations.
How mature are your content operations? What are your content needs? How much does your content team know about AI?
Performing a content audit and assessing the maturity of your content operations are two important, almost mandatory, steps to take before you start shopping around for AI tools. Remember, generative AI is just a tool, and a heavy one at that. If your content operations aren’t strong enough, overreliance on a generative AI tool will do more harm than good to your content operations.
Content Science’s AI Whitepaper is a great source to learn more about text generative AI and when to adopt it. Take some time to review the included maturity model and assess your content operation’s level of maturity.
Despite how impressive generative AIs are proving to be, they still have limitations. No generative AI can meet every one of your text content needs. However, text generative AIs have become so numerous and specialized that there are more appropriate scenarios than inappropriate scenarios.
Most generative AIs train on longform writing, like articles and books. Training a generative AI on longform writing has proven to be the best way to make the AI understand the nuance and style of language, producing a more natural-sounding response.
Generative AIs also excel at surface-level research from what’s available on the Internet. Most AI language models wouldn’t have access to documents and databases unless they came included in the AI’s training materials. Therefore, a text generative AI won’t be able to do as much in-depth research as a skilled human researcher.
However, a text generative AI can skim publicly available text online to learn about a subject. In a test we conducted with AskWriter, a text generative AI built for businesses, we asked, “To what extent was Jimi Hendrix responsible for the popularization of rock n’ roll?” It’s safe to assume that Writer didn’t train their AI on Hendrix’s biographies, but it still produced a valid, fact-based response.
This is because others have published works on Jimi Hendrix and his role in popularizing the genre. AskWriter read those works, took relevant facts and arguments from each, and created a new piece based on those writings.
Given what we’ve found, text generative AIs are equipped for longform writing pieces like:
To produce any of these pieces, the information has to already be on the Internet, and it has to be publicly available. Text hidden behind subscription walls, hidden in PDF documents, or incorporated into graphics won’t be discoverable to the AI.
And even if the text is available and discoverable, there are certain formats text generative AIs are poorly-equipped to handle.
What can’t generative AI do as easily? It depends on what AI you’re using, but common types of content that most generative AI can’t produce work for include:
As of now, no generative AI can produce an opinion piece. Remember, generative AIs produce results based on what they read and train on. Like your least favorite uncle on Facebook, a generative AI can only regurgitate an opinion it’s already read.
While generative AI can produce a current event article, it encounters the same problems experienced with opinion pieces. A generative AI won’t beat the pavement to learn what happened, interviewing eyewitnesses and asking officials for statements. It will recycle what’s already been written on the subject, and that creates ineffective content.
Generative AIs aren’t necessarily bad at shortform explainers, but most struggle to adhere to character and word limits. Generative AIs are more focused on giving comprehensive answers than succinct ones. However, an editor can always trim a response down to adhere to length limitations on a written piece.
When it comes to marketing and communication copy, an open AI does not work well. The AI does not know your products, services, brand, and the like. But closed AI can work well because it’s trained by your organization. For example, AskWriter gives you an API to their platform trained off your organization’s content. This allows AskWriter to adopt your organization’s voice, a vital aspect of any marketing and communication copy.
One of the reasons only mature content operations can effectively use generative AI is that mature content operations already have skilled, full-time staff.
Adopting a generative AI tool doesn’t replace this staff; it only makes them more important. Just because you have a generative AI doesn’t mean you can put Frank from accounting in charge of editorial.
Editors, copywriters, and other text-focused content professionals are critical to effectively using generative AI. Any product from a generative AI will require substantial editing and proofing, and these tasks are best left to professionals.
Marketers and technical communicators can also reap benefits from a text generative AI. Text generative AIs can help these groups identify keywords and effective written structures for their subject, reducing research time and streamlining their writing processes.
Now that we’ve covered the capabilities of generative AI and who will use them, let’s get on to the fun stuff: how to use a generative AI.
Using a text generative AI is simple but requires more extra steps than a human writer.
As we’ve discussed, different AIs excel at different tasks. When assessing your content operations, try to identify gaps or growth opportunities. Use these findings to determine your needs and then set off to learn more about different AIs. Content Science’s content technology infographic includes many of the best text generative and AI writing tools available.
Experimentation can help determine what AI you can benefit the most from. Many AI tools offer free trials and demos, and we recommend capitalizing on these offerings during your search.
It feels like every day new AIs are being developed to address new problems. A field that began with chatbots has evolved to encompass everything from producing product descriptions to generating content plans.
If you need something to cut down on research time, then a chatbot like Chat GPT could be the best solution. If you need help fact-checking blog posts, something like LongShot might be more appropriate. This article by Zapier lists some of the most popular text generative AI available for both individuals and organizations.
Here are five of the best examples included in Zapier’s article:
Don’t jump the gun on AI. Take the time to evaluate your needs, research offerings, and meet with your content teams to determine the best option for your organization.
Different AIs require different inputs to direct their processes. Most text generative AIs go off prompts like “What are the elements of effective content?” or “Create a list of things for tourists to do in New York.” However, some only need keywords or goals to produce text.
Regardless of what form it takes, your input is the most important aspect of using a text generative AI. Remember, these are machines, not people. They need clear, specific instructions to operate, and “Figure it out” does not count.
When curating a prompt, keep it brief. Don’t include unnecessary words and try to avoid modifiers unless they are important to your piece. Adjectives and verbs can go a long way in helping the AI stay focused on your goals, but too many will only confuse the algorithm.
Most text generative AIs love questions. When using an AI to generate a longform article, don’t just input “Article on 2023 inflation.” Instead, ask, “What are the main contributors to inflation in 2023?” The “What” in the prompt will prod the AI to explain and inform rather than just make a list. The “Main contributors” in the prompt will give the AI something to focus on and build an argument.
Text generative AIs operate better when given full sentences or questions as an input. This helps keep variation in the AI’s responses to a minimum. Fragmented sentences and ideas like “article on the American Revolution” broaden the scope of what the AI thinks you want and may cause it to go off topic.
Avoid yes and no questions at all costs. Inputting something like “Did Grover Cleveland wear pink socks” will prompt the AI to search the Internet for sources claiming Grover Cleveland did wear pink socks. Similar to Google searches, the AI will pick out what it thinks are keywords and find sources affirming the input rather than researching it.
Finally, tell the AI what format you want. Some AIs have built-in features to specify a template or type of response. Others need to be told the format of the prompt. Telling the AI beforehand what format you want the response in will reduce the amount of editing required afterward.
Text generative AIs do not operate randomly. They operate off significant training, rock-solid algorithms, and what information is available to them. However, there is a finite amount you can do to narrow down the responses they will give.
No matter how great and to the point your prompt is, there will always be some degree of variation in the AI’s responses. That’s why “averaging” these responses is the best way to produce effective content with a text generative AI.
This is more necessary for some prompts than others. Prompts based on statistics and facts like “What are the top 10 best-selling cars in America in 2020?” produce very little variation because the AI uses universally agreed-upon information. But something like “What’s the best Chinese restaurant in Atlanta?” is so broad and opinion-based that the AI will produce something different each time.
In cases like this, use the averaging method. Find what you like from each piece and sew them together until you have a textual Frankenstein of an article. This method takes the strengths from each generation and incorporates them into one rough, but workable, whole.
Your averaged-out text will probably remind you of Frankenstein: It will look rough and sound funny. Enter: the editor.
At this stage, skilled editors are instrumental in transforming AI-generated text into effective content. The average out method is effective at synthesizing the strengths of each generated response, but its major weakness is the amount of time required to fine-tune it to fit your organization’s voice.
We suggest following four steps to make your approach efficient and thorough.
Before addressing the structure of your piece, analyze its substance. Fact-checking is required for any piece of text, no matter who or what makes it, and is a critical function for the editor of a content team.
Generative AIs pull from the Internet to make a claim. They are accurate most of the time and know to avoid unreliable sources like small blogs or social media posts, but they are still prone to accidentally spreading misinformation.
Be sure to evaluate if your AI is cherry-picking evidence, using unreliable sources, or citing fake sources. Remember, text generative AIs are not skilled researchers. While they can search the Internet, they are not well-versed in how to do it effectively and reliably. If you cannot find a proper source for a generated fact, statement, or citation, it’s best to remove it and avoid potentially spreading misinformation.
After fact-checking your piece, it’s now time to make it flow. This is the time to check for spelling and grammar mistakes and evaluate the piece’s overall structure.
Your editor should look for ways to make the article sound more “human.” To do this, keep an eye out for the following opportunities:
After fact-checking and polishing, your piece is halfway done.
Using a text generative AI and skipping this step puts you at risk of losing your brand voice. This is a very real danger of relying too much on a new technology.
This step is less important if you use a closed AI trained on your organization’s content. However, evaluating opportunities to insert more brand voice into your work is never a bad idea.
Your organizational voice is why your readers return to your site to hear what you have to say. Without it, it doesn’t matter how much content you produce. No one will want to read it. While it may seem like an unimportant step, this is vital to incorporate a text generative AI into your content production process.
Using an AI-detecting program is one way to evaluate if you have effectively inserted organizational voice into a piece. Full disclosure, these can be unreliable. In one notable case, ZeroGPT, a detection software by the creators of Chat GPT, said AI wrote 59% of the US Constitution.
There has been significantly less investment in AI detection software than there has been in generative AI. Also, there is less transparency in how they operate.
AI-detection tools are a reasonable tool for determining if you’ve inserted your organization’s voice into a generated piece, but they are only one tool.
If your content operations are mature enough to incorporate text generative AI tools, then you’ve probably invested time into aligning and scaling your content voice. Writing assistance tools like Acrolinx, Writer, and Grammarly are useful for inserting organizational voice into your AI-generated content and ensuring the time you’ve spent on developing your voice doesn’t go to waste.
Use your writing assistant tool of choice to check for quality, style, structure, and language. What kind of tone has your organization developed? What are your primary audiences? What are you trying to accomplish with this piece? These are all relevant questions to consider when applying organizational voice.
We recommend using a combination of AI detectors, your writing assistant tools of choice, and human judgment to assess how much voice is present in a piece.
What? But the editor just went through three different steps to polish the article. Why do I need another editor to review?
When using a text generative AI, the person who executed the previous three steps is the actual “writer.” They are the ones who assembled and polished the text into a fully formed piece of text content.
It may seem redundant to have this piece edited, but the person who assembled it is prone to all the same mistakes a writer makes when they write a piece themselves. After all, they are creating text content, just not writing most of it.
Having an extra pair of expert eyes on the text before it’s published is a smart decision when using text generative AI. Especially in the early stages of adoption while your organization is still transitioning. There’s always a chance that someone focused on assembling text will miss some crucial structural element or have grammar mistakes in a transition statement they inserted.
Text generative AIs are useful tools but not substitutes for skilled writers and editors. In fact, writers and editors are just as crucial to operating a text generative AI as they are for producing text in-house.
The range of applications for this technology is huge, varied, and fascinating. To capitalize on all a text generative AI has to offer, you need to have skilled individuals on standby, ready and trained for your organization’s needs.
While some steps mentioned above may seem time-consuming, an experienced team can execute them in less time than it would take to produce an original piece. Incorporating text generative AI into your content production process may seem intimidating, but it could be one of your organization’s most valuable business decisions.
Learn how to bring out the full potential of text generative AI to create impactful content.
Discover the power, benefits, and opportunities of an end-to-end content strategy with this comprehensive white paper.
Celebrate the fifth-year anniversary of The Content Advantage, the top-rated book empowering professionals to better their content.