6 Predictions About AI in Content Marketing


I shared a cryptic LinkedIn post last week, so in the spirit of elaboration, here are six predictions about how AI is going to change content marketing.

1. AI Will Dominate Simple SEO Content

We’re already pretty comfortable with the idea of AI writing advertising material, website copy, product descriptions, and even social media content. But AI can (and will) be used for bigger, more complex parts of content marketing — particularly SEO.

GPT-3 is very good at writing things that are short and creative. Task it with something long and factual, and it struggles: it can’t sustain an argument over multiple paragraphs, and it routinely generates made-up information that sounds like fact. (This is where human writers tend to breathe out a sigh of relief — we’re safe.)

But that is less of a problem than it sounds. Most human-written SEO content already has little-to-no narrative, reading more like a wiki page than the New York Times. We already assume the reader has a fish-like attention span and have chosen to optimize our content for “scannability,” writing content like What is X? and How to do Y articles — so does AI’s inability to write narrative prose really matter?

Many human-written SEO articles also contain factual inaccuracies, a product of the less-than-rigorous research that goes into them. In many cases, this does not hinder the article’s ability to rank: these inaccuracies often sound coherent, buried as they are in paragraphs of fluid, flowing prose (that few readers will even read). Google is not sophisticated enough to vet the accuracy of every claim and idea in every article, choosing to use other heuristics of quality (like backlinks). So again — does it matter if similar issues plague AI-generated content?

These issues can also be avoided outright with a small amount of process input. Asking GPT-3 for a 2,000-word article will result in a sprawling, incomprehensible mess, but ask GPT-3 to generate 10 200-word paragraphs and stitch them together yourself, and the result is far more compelling. Add in a minimal amount of human oversight — someone to reorder sections, add keywords into H2s, regenerate wonky phrases, and fact-check — and functional, human-like SEO content can be created in minutes.

Case in point: here’s an article I generated in <five minutes on a pretty esoteric topic:

This is not good content, but it is human-like. It is as good as the best effort of a bad writer, and in a non-competitive SERP on a website with decent domain authority, it will rank.

This is not to say that this is a desirable outcome; simply that it is a likely one. GPT-3 will do a passing job with any content where the narrative doesn’t matter (say, “skimmable” SEO content), where the subject matter is well-represented in its training dataset and where minimal human input is used to overcome its greatest flaws. We will see more AI-generated content, and SEO will be where it lands first.

Read more: The Search Singularity: How to Win in the Era of Infinite Content

2. Google Will Do Very Little to Counter It

One common argument against this idea is that Google will simply not allow it. They have already ruled that AI-generated content is against their search guidelines; they will then find a way to penalize AI-generated content. 

But there are problems with this thinking.

For one, it’s going to be hard to separate out AI-generated content from human-generated content. The nearest parallel we have is “spun content,” something that Google has devoted numerous algorithm updates to combating (very effectively). But spun content contains obvious hallmarks of automation: it’s essentially a fancy rephrasing process, and the underlying structure of the article remains the same as the article that was plagiarized. It’s easy for Google to spot synonyms.

AI content does not have these same hallmarks. Natural language generation is so remarkable because it looks and feels like human-written content — and it is original, the result of the model learning the rules of language and not simply copying existing information. It will be much, much harder for Google to systematically differentiate between AI-generated and human-written content.

As a non-engineer, I may have a blindspot here. There are efforts to use technology to detect AI-generated content. But even if we assume these efforts will be successful, we still run into issues:

  • No detection tool will be 100% accurate. Without perfect accuracy, detection will cause lots of false positives (and negatives). If the detection model is, say, 72% accurate, 28% of all the legitimate, human-written content analyzed by the model will be incorrectly flagged as spam. Is it fair to action that data in any meaningful way? (Imagine being one of those false positives and having your entire site hit with a manual penalty.)
  • Technology is an arms race. SEO has always served as an arms race, and as new detection methods improve, new ways of hiding those hallmarks will appear. Natural language generation is in its infancy: what about LaMDA or GPT-4? Or aftermarket additions to these models from companies like Copy.ai or Jasper.ai?
  • There is a blurred line between human-written and AI-generated content. If a human writes an outline, AI generates the first draft, and a human performs minimal editing, should Google penalize that content for being AI-generated? Where is the crossover point between AI-generated and human-generated?

3. Non-SEO Content Will Matter More

SEO has long been the primary form of content marketing, thanks to its ability to generate predictable, compounding traffic. But as many content marketers have observed, the marginal return on SEO content is trending down as more and more companies begin to compete. AI will vastly accelerate this process.

But there are still types of content that natural language generation is not equipped to tackle. It can’t reveal anything new about the state of the world. It can’t interview people, conduct original research, share personal experiences, or analyze data. It can’t, in a nutshell, write thought leadership content.

When every company is able to publish functional SEO content, the field of competition will shift. Utilitarian SEO content (What is X? and How to do Y articles) will offer less and less benefit. Companies will be incentivized to invest more in other forms of content, like media and thought leadership.

The SEO content that remains worthwhile will need to find ways to stand out from an SEO of AI-generated copycat content, increasing the importance of “information gain” and differentiators like personal opinions, original research, and contrarian stances.

4. The Provenance of Writing Will Become More Important

There’s an interesting ethical twist to AI content: if an article generated by AI is accurate and helpful, does it matter that it’s AI-generated? I think yes, for two reasons:

  • The provenance of information matters. If I were to share advice with you about winning a Formula One race, my advice wouldn’t be credible: I’m a poor driver at the best of times, and I’m certainly not an F1-caliber race driver. This is true even if the advice shared is technically accurate: it isn’t enough to get good advice from a bad source. To that end, will we trust the advice generated by AI, even if it is technically true?
  • We need people to hold to account. If a human writer shares something scandalous or harmful, it’s relatively easy to hold them to account — but what happens if AI does the same? Who is responsible for the harm that ensues? The person clicking the button? The company hiring the person clicking the button? The developers of the tool used?

These problems already existed in a pre-AI world, as did one possible solution: Google’s EAT guidelines. Google realizes that topics like medical and financial advice benefit from greater scrutiny of the authors who are tackling them, so they encourage websites to have real, expert, credentialed people author this content. In a world where human-sounding content can be written without any human involvement, this will become more and more important.

We are likely to see both readers and Google place more credence on the authors of online content, placing more trust in people and institutions and becoming more skeptical of anything authored by brands, teams, or pseudonymous people. Having real, credible (perhaps even verified) people author and edit content will become a competitive differentiator.

5. AI Writing Tools Will “Land-and-Expand” in Most Organizations

Even in the parts of content marketing that AI is currently unequipped to take over, it will still become a routine part of our processes. AI is simply too good at the hard parts of writing, like brainstorming titles, adhering to style guides, or writing introductions and conclusions, to be overlooked.

It’s also incredibly cheap. Most AI writing tools operate on either freemium models or very consumer-friendly price points. There are low barriers to use: non-technical people can “talk” to the tool in plain language and get workable results in seconds.

Huge productivity gains and freemium pricing mean one thing: we’re going to watch a bottom-up, land-and-expand-style adoption of AI writing tools within organizations. Copywriters will speed up their brainstorming process. Sales teams will generate custom proposals in a fraction of the time. Legal and finance teams will expedite SOWs and MSAs. AI writing tools will spread quietly and thoroughly throughout most organizations.

Regardless of the type of written work you’re conducting, AI will become a part of the process, either as the engine generating the bulk of your written work under your guidance or as the creative sparring partner helping you work through creative blocks.

Read more: GPT-3 is the Sparring Partner You Didn’t Know You Needed

6. We Will All Become Strategists

As content marketers, our careers have been predicated on our ability to write well and better than other people. It’s impossible for writers to compete — at least mechanically, in terms of sheer output — with an AI that can generate thousands of coherent words in a matter of seconds. Seeing lines of code come pretty darn close to emulating this is pretty scary — but it’s not necessarily a bad thing.

There are parts of content marketing that aren’t very fun. Writing long, sprawling listicles. Rewriting your article title a dozen times. Creating 50 variations of ad copy. It’s these parts of content marketing that AI can — and probably should — be unleashed upon, freeing up our capacity to focus on higher-leverage, more skilled things.

Our usefulness also extends beyond rote content production. There is still human skill required in AI content generation: priming the model, refining the output, seeding core facts or product information, structural editing, fact-checking, keyword research and optimization, reporting and analysis, and fitting each article into a broader, purposeful business strategy.

In an AI-augmented world, it will make sense to think of content marketers as the pilots of this technology, shaping and bounding the direction it takes. The role of content marketers could easily shift away from content generation toward curation, optimization, fact-checking, and directional input. We will all become strategists.

Content Marketer, V2

When confronted with natural language models, the urge to protect our hard-earned roles and responsibilities is a natural one — but it’s important to remember that the role of a “content marketer” was itself only made possible by new technology, like the internet.

While I think that there’s a good chance that today’s iteration of the role will change because of AI — who says that has to be a bad thing? What if the V2 of “content marketer” is more skilled, more fun, and more rewarding? What if AI writing tools become the best thing to happen to our role?