AI tools may be fairly new, but marketers are adopting them quickly. Research shows 84% of marketers use AI tools daily, and of them, 87% use it for content creation. Whether your team taps AI to write social posts, generate images, craft emails, or produce articles, quality control is a concern.
Generative AI platforms pull information from existing content, synthesize it, and transform it into fresh content. But what if AI pulls information from an unreliable source? Or writes something that doesn’t seem truthful? It happens. Marketers are quickly being introduced to “AI hallucinations,” which means AI is making things up.
As a result, you need to fact-check every piece of AI-generated content.
Steps to fact-check AI-generated content
No matter what kind of written content AI has produced, you should comb through it to ensure everything is accurate. If you don’t, you could publish “fake content” that can hurt your brand’s reputation and disengage customers.
To fact-check, follow these steps:
- Highlight facts
Once a piece of content is generated, proofread it and highlight the facts. Turn each sentence yellow so you can go back through each item and confirm it is, indeed, true.
Keep in mind that “facts” go beyond numbers. Facts include:
- Statistics
- Claims or statements of truth
- Events, dates, timeframes
- Places
- Studies, reports, or research
- Quotes
- Check facts against multiple sources
With the facts highlighted, you need to find multiple sources online to ensure its validity.
Copy the facts into Google’s search box and see what comes up. If search results return a list of sources that support it, move on to the next. If you can’t find the statistic or the person quoted, for example, do some more digging. If you can’t find supporting sources, highlight them in red with plans to remove or rewrite them.
- Review the sources, if possible
Some AI platforms provide a list of sources used. While AI won’t source each sentence, it gives you a good starting point. Look for credible sites, ones you recognize simply by name.
If you don’t recognize a site, research it. Check out the company’s About Us page, review the blog, and click on a few bylines to ensure credibility. You’ll need to edit the content if it’s not a site you want associated with your company.
Use fact-checking tools
Consider bookmarking a handful of fact-checking tools that you can turn to, like Google Fact Check Explorer, Claimbuster, Microsoft’s Natural Language Processing (NLP), and IBM Watson Discovery Service.
To use them, you’ll copy each fact into its search bar, and it will offer a rating of either “true” or “false” and provide a source that supports it.
- Look at publishing dates
Sometimes, statements made by AI writing tools were true but aren’t anymore. For example, if AI references a 2015 article on cell phone use, that statistic isn’t going to hold up today. Using a dated reference isn’t a good look for your brand, so it should be changed to something more current.
- Question quotes
One of the more common AI hallucinations that marketers have noticed is related to quotes. AI often pulls quotes from subject matter experts to provide supporting evidence or statistics. Sounds great, right? Except many marketers have noticed that quotes are attributed to non-existent people or that AI adds real quotes but attributes them to the wrong person.
Help AI create more accurate content through specific prompts
Generative AI content is only as good as the prompts you give it. The more general the prompt, the less accurate the content. It’s why marketers are educating themselves on “prompt engineering,” which helps you create specific prompts that result in better content.
To write more effective prompts, you should:
- Define what AI is writing, such as an article, email, or social post, and provide a word count.
- Describe the audience, tone, and writing style, like conversational or technical.
- Give AI keywords, which help AI find relevant content and aid your SEO.
- Specify formatting by asking for a certain number of paragraphs or bulleted lists.
- Suggest sources that AI can pull from as guidance.
- Ask AI to hyperlink sources in-text (FYI, this isn’t a common feature, and it may only source statistics).
By producing more effective prompts, you can elevate the final product. However, some platforms are better than others when it comes to citing sources. And, no matter which platform you use, you’ll still need to proofread and fact-check before releasing it.
If you proof content and the majority of it is flagged as hallucinations, it’s probably more time-effective to start from scratch with new prompts and guidance.
Remember, AI is still in its infant stages. It has the potential to help marketers with their ever-growing to-do list, but you can’t sacrifice quality control in the name of efficiency. Use AI to improve your content strategy, with the understanding that humans must play a role in its output, too.
