As content professionals, we often hear the question: 'Do we really need to improve our content? Can’t an AI agent or chatbot just handle it for us?’ Yet as Julie Odner, content designer at Sopra Steria EGGS Norway, explains the reality is a little more complicated.
One of my children believes that dragons exist because an assistant at their preschool said they had dragon meat in the freezer.
It’s not always easy to prove that something doesn’t exist, especially when we get incorrect information from someone we trust. Both humans and artificial intelligence rely on available information and without breadth or critical thinking this can lead to misconceptions. To get good answers from AI, you need good data. But do you actually have it?
Technology needs quality content
For new technology to deliver value, it must be properly implemented and supported by high-quality content. If your customers frequently ask for help with something you want them to handle themselves, you have two choices: Create accessible content they can use to solve their issue or use a service like an AI chatbot to give the impression of personalised assistance.
Regardless of the choice, you need good content. AI cannot answer everything straight out of the box. I recently asked ChatGPT about the symptoms of chickenpox and requested the service to compare it with other childhood diseases. The guidance was impressive.
But facts about childhood illnesses are general knowledge. If I had asked an AI solution about the procedure for applying for parental leave, it wouldn’t have been able to provide an answer unless my employer had supplied the relevant rules.
If AI doesn’t know the answer, it makes one up
This is what we call AI hallucination. The answer is meant to appear legitimate but is entirely fabricated. The question is what do we humans trust more? Information we find in a text ourselves, or information we get by asking tailored questions? A personalised response from an AI chatbot can give a false sense of security. We must be especially cautious when using AI to help users understand not only what they are entitled to, but also what they are not entitled to.
Who is responsible for the error if the AI chatbot invents a service or source that doesn’t exist?
Having control over data quality is crucial
It is absolutely essential to know what type of information you have, its value to the company, and what you are allowed to use it for. Many directorates hold important domain knowledge and with that comes the responsibility to ensure the data is accurate and up to date. Even with skilled content designers, the content quickly becomes outdated if professionals don’t report changes.
Updating news articles doesn’t help much if the main text remains unchanged. That’s why you can’t base AI solutions on simple rules like "last modified." Without structured and updated knowledge, it’s risky to build AI services. The biggest problem isn’t missing answers, it’s incorrect answers.
Is it okay to implement the service?
Even though AI agents can evaluate applications or grant employees vacations, leave, or promotions, that doesn’t necessarily mean they should be used in that way.
The AI product you purchase might have been developed in a country with different laws or ethical values that exists in the country where the service will be used. As an example in Norway the National Strategy for AI says that humans must be "involved in decision-making processes to ensure quality and provide feedback at all stages of the process" (regjeringen.no).
Have you considered what tasks should be handled by humans and what the service itself can manage? Before launching a major AI initiative, it’s essential to think through what you truly want to achieve, what you are allowed to do, and what it will require from you.
A good AI service starts with quality content
The more information AI has to handle, the more energy it requires. Therefore, clear and updated content is both more efficient and more environmentally friendly. Even an advanced language model will struggle to provide good answers if the data is messy, outdated, or poorly written.
Here are four good tips for succeeding with quality content:
- Write simply and structured
- Have content that users seek and that you want your service to provide guidance on.
- Update content you already have
- Delete old, outdated content and duplicates
If you use AI on poor data, users might get odd responses, such as dragons existing because someone said they had dragon meat in the freezer. A trivial example, yes. But that’s because we know dragons don’t exist.
Don’t expect your users to know the answer to what the AI agent or AI chatbot is supposed to help them with. Do everything you can to ensure your AI solution provides the best possible answers. Quality content is the key first step toward achieving your goal.