Content Creation with Linguistic Intelligence and Generative AI
Time and again over the last five years, we have been asked about AI and how far off machines are from being able to create technical documentation on their own. We have always stated in our response that this would unlikely be conceivable in the short term. As we now all know, more is now possible than we had previously believed. In fact, so much is possible that every company where text creation is an issue has to deal with it. But is so much already possible with AI that it has to be used productively?
We'll go over the strengths and weaknesses of generative AI for copywriting. We will also explain how you can use generative AI as quickly and efficiently as possible and what you need to do so.
The strengths of AI in the copywriting process
Rewriting texts that have already been written
Generative AI is very effective at rewriting existing texts to meet specific requirements. It works well when you want to make an existing text more gender-neutral and non-discriminatory.
All the firemen can be contacted 24/7
Source text
All the firefighters can be contacted 24/7
Corrected by AI
Generative AI can also provide reliable support if you need to make content more friendly.
Do not enter! Private property!
Source text
This is private property, so please do not enter. Thank you!
Corrected by AI
However, it becomes particularly interesting when there are complex problems and errors in a text that previously had to be rewritten manually. In such cases, the AI cannot reliably understand what requirements editors have for the text in terms of style and terminology. But if the problems are known and have already been specified, rewriting is once again very reliable.
If, for example, there is an editorial requirement that restricts use of the passive voice, the AI can easily rewrite the sentence.
This problem can be solved easily by a mechanic.
Source text
A mechanic can easily solve this problem.
Corrected by AI
Make newly generated texts look human
Another strength of generative AI is its ability to make texts seem human. It is often linguistically impossible to tell whether the generated text was written by a human or a machine. However, there is no guarantee that the text is error-free or complete in terms of content. In the following you can see two answers to the question "Is it possible to sell electricity from photovoltaic power plants directly to tenants?". We have not verified the content.
Can you distinguish between what a human wrote and what the machine wrote?
Yes, you can sell the energy produced from your solar panels to your tenants. However, you will need to check with your local utility company to make sure that you are in compliance with all applicable laws and regulations. Additionally, you may need to install a net metering system to track the energy produced and sold.
Energy that is generated on your tenant's roof can be sold back to the tenant at a discounted rate. So instead of your tenant being charged say 30p from the grid, as a responsible landlord you can charge something like 20p, saving your tenant money while making some additional profit for you.
Correcting linguistic errors in texts
Generative AI is also very good at "surgically removing" errors from texts. If you ask the AI to rewrite a text, it is very often subsequently error-free. The exact errors are not specified and rewording may be more extensive than actually necessary.
Solvng the equacion was not easys.
Source text
The final result is good:
Solving the equation was not easy.
Text corrected by AI
What the AI cannot do
Consider style guide
If you take ChatGPT as an example, it is a trained model whose responses cannot be influenced to any great extent. It is possible to formulate requirements for the response text via "prompts". However, it is difficult or at least impractical to include a company's entire style guide in such a prompt every time.
Conversely, this leads to the AI for text generation possibly repeatedly adopting a different style guide, which then leads to unwanted variant diversity:
- Insert a coin to proceed
- Insert coin
- You have to insert a coin
- Insertion of coin required
- Could you please insert a coin?
Take terminology into account
Let's stick with ChatGPT. Similar to the style guide, in-house terminology poses significant challenges to the model. It is highly impractical to write a prompt that includes all of the terminology for every request. Thus, it is virtually impossible to avoid expressing the same concept using different terms even in successive sentences.
Create unique content
The AI is able to create content that has never been written before. To do this, it accesses a considerable amount of reference material. Unfortunately, however, there is no guarantee that the content is actually unique and not, at least in part, protected by copyright. This can happen especially when very specific and rare content is created for which little reference material is available. Whether the following AI-generated sentence is unique and can be directly attributed to the AI, I will leave up to you:
Life is unpredictable and full of surprises, just like a box of chocolates - you never know what you're going to get.
Be liable for content
Similar to autonomous driving, creating text raises the question of responsibility. Who is liable for the content? In professional copywriting, the answer is very simple: the company is liable for everything it publishes. Whether the text was written by a human or an AI is irrelevant. Due to this fact, it is currently advisable - similar to autonomous driving - to only use the system with attentive curiosity.
Comprehend or explain
You can ask the AI to compose a text with prompts. But you cannot ask the AI to show the sources or the approach used. For us humans, this is the infamous black box. However, technical documentation in particular requires traceability.
Otherwise, the result cannot be classified as reliable. The AI does not explain how the created text comes about.
How to integrate AI into text creation in a meaningful way
Overall, generative AI is a powerful tool that should not be disregarded.
When correcting content, it achieves a gain in efficiency that would have been unthinkable a year ago. It's true that you can't let the AI do the correcting on its own, because this may put the style, terminology and content at risk. But even when revisions to automatically created content are required, a huge amount of time is saved.
Care must be taken when creating content. Although AI is good at providing assistance when creating templates, they should never be relied upon and rarely adopted without changes.
The AI is impressive, but like the human counterpart, it is not infallible and proofreading is required. Both content and language must be checked to ensure that everything is coherent. In software development, there are already those that assess the work required to "clean up" and optimize generated code to be as high or higher than the actual programming itself. This may be somewhat exaggerated. Nevertheless, it shows that this impressive technology should not be overrated, but seen for what it is intended: as a tool for more efficient communication.
It still doesn't work without being controlled and, as with humans, it works better with rules and guidelines.
Technically, there will be many solutions in the future to help you integrate AI into your processes. This is especially easy if you are already using Congree, since Congree already offers AI integration as an option. This allows you can incorporate AI into all editor environments in which you also use Congree with no additional effort.
tl;dr
AI can be used to good effect to generate new human-looking text, rewrite existing sentences, and fix simple linguistic errors in text. What AI without linguistic intelligence cannot do is use consistent language and corporate terminology. In addition, content errors can creep into AI-generated texts, and there is some lack originality. Liability issues and traceability are also topics addressed in this article. With Congree, however, there are exciting synergies.
How Congree UCC integrates AI
Using generative AI will enable you to publish more content in shorter cycles. As described earlier, this will significantly increase the importance of editing in your company. In and of themselves, the tasks remain the same. Critical subject areas such as style guides and terminology will continue to be the focus of copy editing. But the volume of text to be edited will increase. First, because AI handles tasks involved in creating text, and second, because this speeds up the text creation process by humans.
If Congree is integrated into the copy editing process, a fully automated copy editing process in conjunction with AI might even be conceivable if required.
The primary role of Linguistic Intelligence in Congree is machine editing. Congree knows your company's style guide and terminology. It also knows all the important grammar and spelling rules. It is therefore virtually impossible to break any rules that apply in your company without Congree lifting a finger.
Congree UCC thus closes precisely the gap that the AI alone would leave. The style guide is taken into account. The wording is reliably adhered to. Spelling and grammar are observed.
When Linguistic Intelligence identifies a problem, it can be assumed with a very high degree of probability that it needs to be corrected. By integrating the AI, it is now possible to automatically correct any errors that occur. Congree then automatically ensures that the corrections do not contradict the terminology or the style guide.
Your next steps
Take the opportunity to look at what AI can do for your business. The first step doesn't have to be a major project. If you're already using Congree, talk to your contact and give it a try. If you're not using Congree yet, get in touch with us. We'd be happy to set up a test environment for you.