5 min read

How to scalably and rapidly create AI-generated content for a marketing blog

One standard growth marketing trick for startups is to generate a lot of blog content. This can do two things: (1) it can create an SEO trail for your website for Google to pick up; and (2) it can convert browsers to customers. This post presents one strategy for using AI tools to scalably generate marketing content.

First, some caveats:

  • Consumer AI tech is relatively new and is rapidly improving its capabilities. What is standard today will not be standard tomorrow. It's important to read this post with this dynamism in mind. Successfully using AI tools requires an adaptable mindset.
  • The output requires some human editing. How much editing you do depends on whether your intent is that your content be read by people and search engine bots or only by search engine bots. If it's the former, you will have to do some minor editing to make the writing appear more human-like. If it's the latter, a light review to ensure that no egregious or erroneous material slips by is all that is required.
  • Google's treatment of AI-generated content is unclear at present, and that could change. Right now, for SEO purposes, AI-generated content doesn't seem to be penalized by Google. However at least one site claims Google has said that AI-generated content is against its guidelines. To the extent that AI-generated content becomes impossible to differentiate from human-generated content, it is unclear how Google would enforce this.
  • The content we're interested in generating for this exercise is marketing copy for a blog. It's not literary or academic writing. It is meant to be transactional: its goal is to either convert a casual browser of your site to a customer, or else to help improve your company's SEO ranking.
  • You should identify a quantifiable and objective goal for your content, and track your progress on a weekly basis. Adjust your strategy given your results.

Those caveats out of the way, let's move on to the process. This is a fairly structured and logical process.

Identify the elevator pitch for your company's product.

Say you build software that complements Hadoop installations. "We make Hadoop fly" or some such pitch. OK you're selling a product that helps people make their Hadoop installations perform better, or you help Hadoop customers get more out of the software.

Find 500-1,000 keywords that relate to your company's elevator pitch.

Google's Keyword Planner is a free service. Ahrefs' Keyword Explorer is a paid alternative. There are a bunch of other free and paid keyword tools, but those are the two big ones.

500 to 1,000 keywords sounds like a lot but consider two things: (1) the world of SEO keywords is incredibly large and (2) you're using AI to eliminate a lot of the drudgery associated with writing marketing content for blogs. If you systematically work through this process, you'll find that it doesn't take as much time as you'd expect to seed your blog with your 500-1,000 keywords.

Some tips:

  • Download your list of keywords, and the associated data about their relative competitiveness into Excel (or Google Sheets).
  • Identify those keywords that are not very competitive. Since you're trying to improve your company's SEO rankings, you want to first generate content against non-competitive keywords in order to gain some traction. As you perfect the content generation process you can move up the SEO competitiveness curve.

Select your AI tool

Create an account with one of the GPT3-powered copywriting tools. Jasper.ai, Copy.ai and OpenAI's Playground are three popular options.

A quick explanatory side note: OpenAI develops the GPT3 model, and licenses it to Jasper.ai and Copy.ai among many others. Jasper.ai and Copy.ai provide a slick user interface that lies on top of their access to GPT3. OpenAI's Playground tool provides much of the same feature set as both Jasper.ai and Copy.ai, albeit with a less slick user interface.

Learn prompt engineering

Now you want to take a few of your non-competitive SEO terms and start prompt engineering! Prompt engineering is an important concept in AI, and it takes a bit of explanation to understand, so bear with me.

The conventional model for human-computer interaction is either the WYSIWYG interface or else inscrutable code. At one extreme, inscrutable code can be as obscure as Brainfuck:

>++++++++[<+++++++++>-]<.>++++[<+++++++>-]<+.+++++++..+++.>>++++++[<+++++++>-]<+
+.------------.>++++++[<+++++++++>-]<+.<.+++.------.--------.>>>++++[<++++++++>-
]<+.

But what if there were a middle ground? What if it were possible to give your computer natural language instructions, and have it spit out output in a WYSIWYG environment?

Consider the following statement: "Write a blog post about how Hadoop allows its users to use their data, and include the phrase "big data"". "Big data" is one of the Hadoop-related SEO keywords we found, and we want to ensure that it and "Hadoop" is in the content that we generate.

This, in a nutshell, is prompt engineering. It is the art of creating a natural language prompt which induces the AI to create some desired output. Think of prompt engineering as an iterative process. When you first start playing around with prompt engineering, you may find that your AI tool generates output that isn't great, or isn't what you expect. You need to refine your prompts, in order to maneuver the output closer to your goal.

In some sense, the feedback loop looks like negotiating with a toddler trying to learn something new. Prompt engineering is an iterative learning process, and the feedback is the output that the AI tool generates. Refine your prompt as you see what it induces the AI tool to generate. Over multiple iterations, your prompt becomes more refined and the output more salient. You are engineering a prompt in order to get software to more closely hew to your expectations.

Output the content to your blog

Whether you use Substack or Ghost or another blog provider, you want to get your content out of the AI tool and into your blog editor. Copying and pasting is the easiest method, though you could obviously write a custom script which automates this process.

Review the content and edit as necessary

Remember that one of the caveats to this process is that you have to determine whether you intend for your blog to be read by human and search engine bots, or only by search engine bots. If it's the former, then you need to spend a bit of time with each blog post to ensure that the phrasing sounds human-like. If it's the latter, then all you need to do is to perform a cursory review of the output to ensure that no egregious or erroneous information slips through.

Publish your post, measure its performance, and iterate

Once you're satisfied that your blog post meets whichever threshold you've decided upon, publish it. Track its performance through Google Analytics, or whatever other traffic analysis tool your company uses. Basically, you want to determine whether the post converted browsers to customers, or whether your site's SEO ranking improved.

It will take some time, and a large volume of content, to really know whether you're on the right track. This is especially true for SEO-related performance analysis.

Summary

This process entails the following steps:

  • Identify your company's elevator pitch.
  • Find 500-1,000 keywords associated with that elevator pitch.
  • Select an AI content generation tool.
  • Use prompt engineering to induce the AI tool to create content relevant to your keywords.
  • Iterate through your 500-1,000 keywords, refining the prompts as you move through the process.
  • Move the AI output from the AI tool to a blog.
  • Review the post for human or machine readability, as appropriate.
  • Publish the content.
  • Review the performance, and iterate.

Published by: