Robot Revolution & the 'AI Sandwich'

I don't believe it's hyperbole to say, "this changes everything."

Robot Revolution & the 'AI Sandwich'
  • AI is already widely used in various industries, but AI-driven change will accelerate. We must guard against its abuse.
  • Ideation, prototyping, research and content generation are key client use cases for ChatGPT and AI tools.
  • AI will be a boon for digital marketers but quality requires active human oversight: the "AI Sandwich."

I have spent the last several weeks obsessively playing with AI content-generation tools.

In my 25+ years of working on the web, I can't recall having been this excited about a new technology – not even the single-pixel gif. I don't believe it's hyperbole to say, "this changes everything."

Source: Google Trends

Marketers, and in particular SEOs, are losing their minds about the potentially revolutionary nature of AI, Chatbots, and ChatGPT in SEO. And While ChatGPT is grabbing the most attention, there are many AI tools for marketers.

Before we get to those, it's important to look at the dystopian potential. Below are some fictional and non-fiction representations of AI.

AI In Fiction and IRL

Cory Doctorow in his 2003 novel "Down and Out in the Magic Kingdom," envisions a world where we can all live forever, in part due to AI. If we score enough "whuffie," the social credit of Doctorow's imagined future, we may upload our consciousness, be decanted at will into a new body, or even be entitled to drive away in someone else's car.

For Doctorow, AI is an enabling technology that allows people freedom and self-actualization – as long as they score enough social validation in the form of "whuffie."

There are many other literary and pop culture references to AI. Often they appear in the form of autonomous robots, where the outcomes range from funny and sweet to downright dangerous.

Among the earliest, in "2001: A Space Odyssey," the ship's computer, an AI named HAL 9000, becomes self-aware and threatens the crew of the spacecraft it inhabits. And everyone remembers Arnold Schwarzenegger's "The Terminator" character, but he's just an emissary of Skynet, which becomes self-aware and decides humanity is the problem.

Pixar's "Wall-E," the more benign one-robot cleaning crew, is a by-product of humanity evolving (devolving, maybe) to the point it can't care for itself or the planet.

Science fiction often becomes, years later, science fact. And the technologies now emerging should make us aware that the future is now. Very soon, these technologies will be part of our everyday routines.

In Healthcare, AI is already analyzing medical images and helping with diagnosis and treatment. It's helping to optimize the energy grid, arguably the most ubiquitous technology of all. AI is even helping to personalize education, and assists with grading and feedback. It's also driving (or helping drive) your car.

In a particularly foreshadowing turn, Tesla announced it would not offer the option to purchase certain cars at the expiration of their lease, as is common practice. "Please note, customers who choose leasing over owning will not have the option to purchase their car at the end of the lease, because with full autonomy coming in the future via an over-the-air software update, we plan to use those vehicles in the Tesla ride-hailing network. Customers can visit tesla.com/3 now to lease a Model 3."

Las Vegas was the first, but not the last, city in the U.S. to allow autonomous cars for hire. San Francisco is testing it; Los Angeles may be next. The robot uprising can't be far behind.

Do you trust Uber and Tesla with the future of humanity?

OK. We'll put that away for a minute. Now for the marketing discussion.

AI and ChatGPT for SEO

Scammers and spammers are often the first to take new technologies seriously. It's often said in SEO that if you want to know what's coming next, look at the "three Ps": Poker, Porn and Pills.

Though most of us are just being exposed to AI content tools with ChatGPT, there are a number of tools that have been in development that look a lot like the "content spinners" of old.

Content Spinning is the practice of taking an existing piece of content and using software tools to rewrite it to be unique. This practice has been around for more than a decade and was previously based on simply rearranging words or replacing synonyms. It's the sort of content you would see on a site that was hacked and taken over by a Viagra promoter.

Beyond this, we're already seeing reports of ChatGPT being used for ill.

Generating Code

One of the cool features of ChatGPT is that it can render programming code. Here's an example: my prompt was "php code to write the html for an ascii cat face."

Notice that ChatGPT went beyond my request, answering a question I didn't ask, how to style the ASCII cat face using CSS. Pretty cool and innocuous, right? What if I asked it to write me a virus or an exploit?

Turns out that's already been done.

Cybersecurity researchers are already finding instances of bad actors posting functional malware code to cybercrime forums. As reported, hackers and aspirants openly share code that can do all kinds of harm. Examples included malware which can be embedded in an email to encrypt all the files on a user's computer – potentially to be used as ransomware. Another example is searching the user's computer for files of a certain type to compress and send them to the hacker – again delivered by email.

ChatGPT even wrote the email to deliver it.

It's not all evil, right? No, of course not.

AI Does It Better

I've been playing with ChatGPT since the new year and have greatly benefited from it. Some of my experiments have been in SEO and website development, and some have just been free association.

For example, I was on a plane watching "Fast Five" – judge me if you must, but I'm a sucker for the "Fast & Furious" franchise – and I remembered hearing there was some kind of beef between Vin Diesel and Dwayne (The Rock) Johnson.

So I asked ChatGPT what that was about:

You would never get that kind of result from Google. You would get this instead:

The Google result offers a starting point, but it's not an answer. That lack of clarity and decisiveness is the big difference between an AI Chat-based answer and a conventional search result. (We'll see what Google Bard brings.)

Despite Google's seeming attempts to make sure you get what you need without ever leaving Google, AI does it better. ChatGPT and other AI-enabled tools shine in examples like the code above, where they're either completely doing the work of humans or significantly shortcutting the process.

(Note: simple prompts generate all the cartoony images in this article in Midjourney, an AI image creation tool using Prompts like "robot uprising in Disney World.")

And while you might think that the next frontier for knowledge work is writing great prompts, I'm sorry to tell you there's an app for that. You'll have to look elsewhere for career opportunities, sadly.

Multiple Use Cases

We can all agree there's too much crap on the internet. Unfortunately, I think that AI writing tools will amplify that 100-fold before the search engines get good at detecting it. So how can AI be used for "good" rather than for "evil."

Ideation

Imagine you're working on a social post for a topic you're unfamiliar with. You could ask ChatGPT to suggest some hashtags you might use.

The following was for a post on how the growth of AI is like the long-feared robot uprising. See above.

The next one was for an AI image of a pair of cuttlefish rendered in the style of the video game Minecraft.

I also tested was asking ChatGPT to recommend blog post topics.

Prototyping

Let's say you're building a website and you don't want to wait for the client to give you text and images. You might use ChatGPT or other AI writing tool to create sample content and Midjourney to create images you can replace later.

It's not ready for prime time, but it's a great placeholder while you work with the client to get the real content.

Confirming assumptions/'fact' checking

If you're wondering about what kind of work a client does, you can ask ChatGPT and get some ideas. Yet, as with all internet-trained models, take everything with a grain of salt. There are some well documented mishaps in the early days of AI Chatbots for general information – Microsoft's first attempt was reported to give racist and mysogenistic answers.

Whether it's correct or not, at least ChatGPT isn't offensive.

Rough drafts

The same content we identified as useful for prototyping might also serve as a good rough draft to be further developed by the humans in your employ. Here's an example:

Other AI content tools

Lest you think ChatGPT is the only game in town, it's not. There are a number of established AI writing tools, including:

Over the coming weeks and months, I expect we'll see an explosion of variations on this theme. And just in case you don't have humans at your disposal, Mike Blumenthal pointed me QuillBot, which has a rephraser tool that can help make your AI-generated content sound more human.

There are an equal number of AI image-generation tools available, too.

Many people are familiar with the original DALL-E, from the makers of ChatGPT, OpenAI. Others include:

Size matters

Sad but true, size matters in SEO content. As content creation has become easier, the length of authoritative articles on the web is getting longer. Helpful inclusions are things like FAQ, and People Also Ask (PAA in SEO speak).

The above would be very useful if you are building content for a Chicago limo company. They add value to the web page for a reader and could be used in structured markup (they could also help with voice queries). And, ChatGPT can write the Schema.org complaint markup as well.


It can also write HTML, PHP, Python, Curl, and many more languages.

You would not want to use the content above verbatim, because the output of AI tools is relatively easy to detect given they are following a predictive model to "write."

Agencies and AI: the Concerns

As a digital marketing agency that uses freelancers, we are already finding that some freelancers are starting to use these tools.

There are many AI detectors out there, but the one that appears to be the furthest along is Originality.ai. It's still in Beta (like Google was for a decade) but it seems pretty good so far. Unfortunately, in a couple of instances we found of heavy-handed AI content inclusions from our freelancers did not make their work better. The content was mediocre and the writer who produced it was not a keeper, regardless of the tools.

Google's Helpful Content Update

Some may worry that the use of AI tools will lead to content being deemed "unhelpful." I don't believe that's a concern if the tools are used appropriately.

It's unlikely that the first iteration of a blog post spit out by ChatGPT or Jasper would be truly helpful to a searcher on Google. They are helpful, however, to skilled content writers and strategists who can use them as a framework or inspiration for fleshing out a more complex piece of content, which would in fact be "helpful" to an end user.

E-E-A-T and Search Quality Rating

E-A-T, and now E-E-A-T are references to the Google Search Quality Rating program and a big topic for SEO professionals.

Search Quality Raters are human beings who review Google search results and give feedback on the quality of the results. Among the things they're looking for (in the original version) are Expertise, Authority, and Trust. The guidelines have been updated to add another E for Experience, hence E-E-A-T. Essentially, the graders are looking for signals that content that appears in the search results deserves to be there.

While a Large Language Model (LLM) Generative Pre-trained Transformer (GPT) has a lot of data, it doesn't "know" anything. Therefore, it fails the "expertise" test.

There are widely reported stories of ChatGPT passing the Bar Exam or business school exams. But this is not knowledge, it's fact-based answers pulled from a massive database.

Conclusion: AI Sandwich

A friend described the best use of AI tools as an "AI sandwich." In essence, humans write the prompts, AI does its thing and the human edits content on the back end.

I think that if we want to produce helpful, trustworthy content it will have to go much deeper than that. The output of the AI content generation tools will need to be like parts on the shelf that together make the whole. And, of course, a human has to edit them for grammar, clarity, and accuracy.

We are about to see an explosion in AI-generated content. The spammers will spin it out like pollen in the springtime. However, legitimate content creators will use it to create better content, develop great ideas and serve their organizations and clients more effectively.

As an old-school SEO,  I believe ChatGPT and its peers are a great opportunity for testing and tuning. As a client-serving marketing organization, we will use it to do better work and amplify the impact for our clients.

I'm really excited about the ways in which these tools and those still to come will help us produce better, more engaging content. And like many of you, I think we may see things get worse before they get better.

Scary that I have to say this: a human wrote this article. The images, however, were generated by machines.

Me, I'm hoping our new robot overlords will be gentle. See you in the Matrix.

Will Scott is the CEO and co-founder of Search Influence.