Robots Writing Your Content: An Interview with Aki Balogh of MarketMuse

March 23, 2021
James Scherer

The robots are taking our jobs.

They’ve been building our cars and responding to our customer service concerns for years now, but (until 2021) they haven’t been creatives.

That’s all about to change.

For this interview, we reached out to Aki Balogh of MarketMuse — a content optimization platform getting ready to launch one of those actually “game-changing” pieces of software.

AI-written content.

Let’s see what he has to say for himself, and (even more importantly) take a look at how good the robot’s article is.

Instantly Export Google Docs Into WordPress

Login with your Google Account now for a 3-export free trial.

An introduction to Aki and MarketMuse (the “why the hell should I care what this guy says?” section)

aki balogh headshot

Aki has been sharpening the cutting edge of content marketing since 2015. As the president and founder of MarketMuse, he’s responsible for a system that analyzes your competitor content and offers insight into how you can create something better.

And, since 2019, they’ve been pushing the limits of content with AI. Using the latest advancements in artificial intelligence, MarketMuse’s First Draft creates a preliminary article for a chosen topic based on its corresponding content brief. The draft strives to attain all the important content metrics while requiring minimal editing.

It’s that last piece which piqued our interest, and drove us to reach out.

Here’s how he describes himself:

Aki Balogh is the President of MarketMuse. Aki co-founded MarketMuse in 2013 and served as its CEO through 2020.

Prior to MarketMuse, he worked with the CEO of InfiniDB, a Series B startup that made an analytic database for Big Data engagements.

Before joining InfiniDB, he was an Associate at OpenView Venture Partners where he looked at Series B investments in Big Data & Machine Learning. Prior to that, Aki was a software developer and a management consultant in the Data Science practice of Diamond Management & Technology Consultants (now PwC Diamond Advisory).

So, a pretty smart dude.

Let’s dive into our interview with Aki, covering the current state of AI-written content, the future of it, and the role he thinks it’ll play in the content marketing industry.

Question #1: So how does it work?

Answer: We take a two-step approach to article generation. First, the user builds a content brief in our platform.

Given a topic, our AI engine generates a detailed outline describing that topic, including potential subheadings (which the user has to select and lock in), and relevant questions and topics that the user selects from.

The process takes about 5-10 minutes per brief and requires the user to think through the goal of the article.

AI engine

When the brief has been finalized, it is sent off for generation.

In short, we generate and re-generate the article a number of times against the topics selected in the brief.

At each generation, our system evaluates the brief for a number of quality measures:

  • Lexical richness (the quality of vocabulary used)
  • Uniqueness
  • Similarity to other sections of the brief
  • Topical comprehensiveness.

This process can take 30 minutes or an hour, depending on the number of generations that have to be executed to find an article that meets all criteria.

Once the generation has been validated, it goes to a human for editing. AI systems have a few shortcomings. They cannot provide analysis and advice, and cannot evaluate factual correctness.

A human editor looks through the draft and makes any changes that are necessary. Once that QA step is complete, the article is sent to the client for their evaluation and publishing.

Question #2: How good is it?

Hard to say! The best test is really a human test. Take a look at the sample (below), try it, and see!

Question #3: How good will it get? When?

This is an interesting question. We’re at the very early stages of this technology, so the quality we’re seeing today is just a starting point.

I estimate that the technology will speed up publishing by about 3-5X in its current state. But over time, the technology will improve to surpass what we ever believed possible.

Question #4: Will Google’s algorithm be able to tell the difference?

Our goal is to create human-generated content at scale, with the help of AI.

Google surfaces the best content on a topic, using AI algorithms that evaluate quality, so it wouldn’t be surprising if most content in the future was written by human-assisted AI, given that the hybrid approach is far more effective than humans alone.

Question #5: Do you think we’re looking at “the future of blogging” here? If not, where are the limits?

Yes. We are absolutely looking at the future of blogging.

The biggest limit I see is the form of media. Once most text is AI-generated, we’ll want to move to AI-generated art, music, film, programming code, education and lifestyle suggestions. The sky’s the limit.

Question #6: What do you think the ramifications are for our reader and our industry?

As with any new technology, one must be open-minded to try it. It is the wave of the future, which is always unsettling, but it’s not going to erase our jobs.

For every job title destroyed by automation, a new job is created. We’ll just adapt accordingly, to higher and higher levels of productivity.

The moment of truth: 500 words written by the MarketMuse AI

Now the moment we’ve all been waiting for. Here’s 500 words on a very specific blog topic: “What is Glucagon?

AI-written content: “What is Glucagon”?

Glucagon is a hormone produced by the beta cells of the pancreas. When blood sugar levels are high in a diabetic person it causes the pancreas to release insulin, which stimulates the breakdown of glycogen to glucose in the blood. Glucagon also releases insulin slowly, so that it is not influenced by the blood sugar levels. Once blood sugar levels fall, the pancreas releases glucagon, which stimulates glycogen breakdown and glycogenolysis of glucose in the liver and skeletal muscle.

When a person with diabetes is unable to keep up with the amount of insulin they are given, they are forced to take insulin injections, which must be taken as soon as possible. At this point glucagon is released from the pancreas and into the bloodstream, which is where it is able to boost the levels of glucose in the blood.

As the blood glucose levels fall, the pancreas releases glucagon, causing the liver and pancreas to release stored glycogen into the blood. This causes the liver to convert excess glucose into glycogen, which is released into the blood. When blood sugar levels are low, it causes the pancreas to release glucagon, which stimulates glycogen breakdown, promotes breakdown of glycogen to glucose and promotes glycogen to glucose. When blood sugar levels are high in a diabetic person it causes the pancreas to release more insulin, which stimulates gluconeogenesis and glycogenolysis of glucose in the liver and skeletal muscle.

In people with type 1 diabetes, the pancreas has one primary function: producing and releasing glucagon. Glucagon acts as a carrier for insulin and regulates its use; it carries the insulin and glucagon together. Insulin and glucagon work together to allow for the conversion of glucose to glycogen and vice versa, and they work together in conjunction.

Glucagon produces two types of insulin.

Type 1 diabetes is caused when cells in the pancreas are destroyed by the body’s attacking the pancreatic beta cells. This leads to insulin resistance and diabetes mellitus. In type 1 diabetes, the cells that produce insulin are not necessarily the ones that are destroyed. The cells in type 1 diabetes are the ones that release glucagon, and these cells are only able to make insulin when they are destroyed in the body. In type 2 diabetes, the cells in type 1 diabetes develop insulin resistance and diabetes mellitus, both of which can be prevented. Type 1 diabetes can be reversed by reducing the amount of insulin injected and by raising the amount of insulin injected.

Type 2 diabetes is caused when cells in the pancreas have been destroyed, but not destroyed by the body’s attacking the pancreatic beta cells. This is called insulin resistance. In type 2 diabetes, the cells in the pancreas are not necessarily the ones that are destroyed, although they may still be able to release glucagon and release it when they are destroyed in the body.

What did we have to do to get this?

Step 1: Enter the target topic that you’re looking to write an article on.

Step 2: If you’re looking to optimize an existing page, drop in the URL. The platform will analyze that as well.

Step 3: Order the First Draft (2 credits in MarketMuse).

Step 4: You’re all set!

MarketMuse will automatically build a content brief on the topic, and will build a First Draft generation according to the topics in the brief. You then drag-and-drop the generated text components to assemble your article.

MarketMuse

My take

So there you have it — they’re coming for your job.

Or are they?

The fact of the matter here is that, even when AI-written content hits its peak in quality, there’s still a need for thought leadership — cutting edge stuff, rather than research-based pseudo-regurgitation of what’s already out there.

There will always be a need for content which is heavily branded — a brand-specific style or tone will be difficult for AI to work within.

And there will always be a need for editing — tweaking of what’s there to make it work for your business and your objectives.

My takeaway here is that, at least for another few years, the value of AI-written content, like AI-written content outlines and briefs (the stuff Content Harmony and Frase are doing), will be in giving your business and mine a leg up in content production.

Just that: a leg up.

Not replacing us, not yet.

But still… Watch this space.

And keep an eye on your microwave.

James Scherer
James Scherer is Director of Editorial at Codeless. He has been a writer, editor, and inbound marketer since 2013, and loves to see brands grow through content.

About the Author

James Scherer
James Scherer is Director of Editorial at Codeless. He has been a writer, editor, and inbound marketer since 2013, and loves to see brands grow through content.