E-Handbook: AI optimizes content management workflows, personalizes CX Article 4 of 4
Tip

Using AI in content creation can streamline publishing

Companies using AI in content creation can generate short-form material automatically, which frees up humans to handle more complicated tasks. But there can be drawbacks.

As an organization grows, it inevitably finds itself needing to create more content for a greater number of people, and AI is already proving an effective answer to this demand.

With content automation, it is now possible to generate stories with no human at the keyboard. Still, AI in content creation is a mixed blessing. In journalism, AI can generate crime reports, summaries of corporate finance, quick-hit sports stories and, unfortunately, fake news. In marketing, it can crank out product descriptions and summaries, recommendations, variations on advertisements and, unfortunately, artificial reviews of products and services.

A few years ago, Gartner predicted that 20% of content would be AI-generated by the end of 2018 -- a prediction that has come true. AI programs such as Heliograf, Wordsmith, Quill and Bertie do short-form journalism. Canva, Scoop.it, Storyboard and Alibaba spin up marketing content faster than ever before. Retailers use programs such as Alibaba's AI writer to generate up to 20,000 lines of ad copy per minute.

How AI generates short-form content

The use of AI in content creation requires three key pieces: natural language generation (NLG), structured data and templates for the desired formats.

The use of AI in content creation requires three key pieces: natural language generation, structured data and templates for the desired formats.

When using NLG, the data items to be used are known beforehand, and the process is planned around it. After interpreting this data to generate written content, AI applies words that work well conceptually with the data presentation and supplements them with referential expressions -- words that identify peripheral locations, events and objects. Then, it tweaks the syntax via a rule-based computer analysis until it comes out reading like it could carry a byline in a newspaper or magazine.

In short, structured data is the starting point, and templates streamline the NLG process for routinely generated content. The result of this process is highly accurate when AI culls data from reliable sources and the applied templates are well vetted for their intended purpose.

What AI produces

So what can AI produce?

Fresh content. Given a strong data set and a well-crafted template, short-form reporting is possible -- and becoming commonplace.

Types of content that AI produces chart

Repetitive content at scale. It's often necessary, especially in marketing, to take a single data set and rehash it dozens if not hundreds of times -- depending on audience and context. This means endless rewrites of the same information for different presentations. Strong templates simplify this drudgery, but AI in content creation makes it machine drudgery.

AI-enhanced search engine optimization (SEO) relevance. It's often useful to take content that's already been generated and optimize it for SEO. This is another simple task for AI in content creation.

Curated, previously published content. Content curation is the marketing practice of digging existing content out of the internet that supports content being used in a campaign and reposting it, but permission from the original content creator is often required. AI can take curated content and rewrite it.

What AI can't do -- yet

Currently, AI can handle short-form content, but long-form content is a completely different proposition.

Even a modest news article of more than a few paragraphs requires the use of multiple -- and variable -- data sets. While AI might render some longer sports stories and financial reporting as general templates, it cannot do the same for most news reporting.

The biggest difference between short-form and long-form content generation involves the gulf between NLG and its cognitive counterpart, natural language understanding. The former is about taking the input of established, anticipated concepts -- such as sports scores and game events -- and putting them into words. The latter is the opposite, taking words offered as input, extracting the concepts and rephrasing them. No commercial software yet exists that can do the latter accurately -- and in large volume.

AI-generated content won't fully replace human beings anytime soon. The point of using AI in content creation is to free up human writers for more complex assignments, which involve understanding and other human traits.

Dig Deeper on Enterprise search platforms and strategy

Business Analytics
Data Management
ERP
SearchOracle
SearchSAP
Close