There’s a slew of new AI tools out there that make life easier for content creators.
Now, here at LNGFRM, we know basically nothing about artificial intelligence. We could talk to you all day about 8-bit NES soundtracks and how they rank against each other, but something as useful as AI isn’t exactly up our alley.
And yet, over the course of nearly four years, our site was unwittingly a participant in what may go down in history as one of the craziest AI experiments ever.
This is a story of how we got dosed by the creator of Juice, and ended up turning a random side project into a small business without even realizing what was happening.
The beginning
Back in 2018, I was approached by the owner of a small PR agency called Echo Park Marketing. Its owner, Derek Sturman, wanted us to interview one of their clients.
Seeing as our site was less than 4 months old and we weren’t exactly getting many interview opportunities, we happily accepted.
This happened a few more times through 2018. A couple of days after the new year, Sturman reached out to us via email and asked if we wanted help with LNGFRM’s content development.
We asked him exactly what ‘content development’ meant, and he gladly explained:
“We’re learning a little bit about SEO [search engine optimization] and may have just discovered a formula for helping readers find articles on your website. Since LNGFRM is basically brand new, this is a good opportunity to see if what we’ve come up with is any good.”
At that point, LNGFRM had almost been around a full year. We were getting a grand total of… drumroll please… TEN visitors per month. We didn’t exactly have anything to lose, so we agreed to write four or five articles per week that followed a specific formula.
The formula was pretty basic. In a quick call, Sturman explained a little about keywords, search engines, and a few other basic things that I said I understood but probably didn’t.
The game plan for these ‘enhanced’ experimental articles was simple:
- Pick one of the ‘magical’ keyword phrases that Sturman vetted for us. Every month we’d get a list of about 100. We’d pick the 20 or so that we’d like and write an article based on them
- Go through a basic checklist of what seemed like obscure considerations. Things like total article word count, headings, length of sections between headings, number of images, etc.
- Publish the articles consistently.
It seemed easy enough. But where was it getting us?
Getting on the Juice
After about 6 months of following this procedure, something magical finally happened. For the first time ever, we got 100 visitors to our site in a month. It seemed like an oddly big jump, given that our all-time monthly high until that point was like, 35 people.
We didn’t think much of it. But from that point, as each month passed, we saw our traffic grow about 10% month over month (over month, over month…).
By the end of 2018, we were getting a few hundred visitors every month.
Some fun things started happening at that point. We’d get an occasional comment on one of our posts. Our contact box started getting more interview requests and people asking us to write about their [insert business/website/weird idea]. We weren’t able to honor all the requests we got, but we did appreciate getting them.
Unfortunately, around this time, fellow LNGFRM co-founder Mike Malone changed jobs and found his blogging time eaten up by his new gig. I lost a lot of motivation and stopped contributing as frequently, too.
In order to monitor LNGFRM’s traffic growth (or lack thereof), we’d given Sturman access to our Google Analytics. Not too long after LNGFRM became static, we got a message from him asking why we weren’t posting anymore.
We explained that the team just didn’t have it in us anymore to contribute as frequently as we had been. We could tell that Sturman was disappointed, but he understood.
So alas, for about a year, LNGFRM was relatively inactive.
Then, something crazy happened.
Starting back up
I got a message around Christmas 2019 from Sturman, asking how things were going with the site. We were posting maybe once or twice a month since we’d last heard from him.
We explained to him that, despite the infrequent posts, our traffic numbers were still growing, albeit slowly. We were up to about 400 uniques per month, and as far as we were concerned, that was pretty good.
He asked us if we were interested in resuming the ‘experiment’ we had started over a year ago. Only this time, there was going to be a twist – OpenAI’s GPT-2 was now widely accessible. Sturman’s company, Panda, had started messing with a custom-trained version of GPT-2 that was designed specifically to help bloggers like me come up with first drafts within seconds.
On a Zoom call, he showed us how it worked: simply input the title of your article, click a button, and voila – a full 1,200-word article. Granted, it took almost 5 full minutes to get the finished article. But that was still a hell of a lot faster than it would take me to make a first draft.
We must’ve spent north of two hours in that meeting, trying out all sorts of hilarious prompts and seeing what the AI would come up with. Everything from “how to pick up girls at a bar” to “why is the sky blue” yielded fascinating results.
While impressive, it seemed like most of the time, the AI output was just fake. The intros were hairbrained, and the articles went completely off topic by the time you got halfway through them.
But at the same time, we were mesmerized. Occasionally, the AI would just nail a random section. Like, to the point where we thought to ourselves, “Wow, that’s actually a really interesting take.” It’s pretty creepy when you realize you’re giving a computer props because it kind of just did your job for you.
After the demo, Sturman asked if we wanted to use the tool. The hope was, by making it easier for us to create articles, we could continue the experiment even with limited input from me and Mike.
We politely declined and told him that the AI seemed to have potential. But given that it took 5 minutes to get an article that may only have one or two useful sections, we figured we’d be better off just writing the articles from scratch.
To our surprise, Derek elevated his offer one more time. Instead of using the AI ourselves, he asked if we’d be open to having him spend time generating a bunch of content, which we could then review and pick the best parts of.
That seemed like it’d be much faster than waiting around for AI outputs ourselves, and we accepted.
So, we revved back up to posting 4 to 5 times per week.
AI to the rescue
Compared to before, it felt like a breeze. While the AI-generated articles Derek delivered lacked our signature LNGFRM sarcasm, we’d punch them up, add a few images and a meta description, and schedule them for publishing.
This was February or March of 2020, right before things went south with the Covid-19 pandemic.
Throughout the pandemic, we stuck to the script of publishing 4 or 5 articles per week, every week, without exception. It was actually nice to have something to work on.
Within months, our traffic growth was exploding. As in we started getting thousands of visitors in a week.
Some of my fellow blogosphere friends couldn’t believe their eyes when I sent over screenshots of our Google Analytics.
A couple of them had been pouring their heart and soul into their site for years and hadn’t seen the type of results that we were getting from content that, in all fairness, a lot of times we were half-assing.
Realizing that LNGFRM had the type of traffic to start pulling in a little bit of extra cheddar for us every month, we joined the Amazon affiliate program and dropped a few affiliate links in some of our most popular pages.
I know what some of you veterans may be thinking – we got in on the whole Amazon game just a little too late.
And it’s true, we did. But when your side project goes from making zero to making enough to pay not only your hosting bill, but a couple other non-project related bills, the feeling is nothing short of magical.
That feeling persists to this day. At least a little bit. These days, we don’t really update LNGFRM like we used to. But luckily for us, the traffic continues to roll in.
A call in the night
One day earlier this year, I got a message from Derek. I’d messaged some weeks earlier how well the site was going some months earlier, but I’m not sure he ever read it.
He wanted to make a quick call and discuss a new project he was working on. It involved LNGFRM, or at least it did indirectly, he said.
“How’s your traffic been?” he asked. I gave him some numbers and said that we were very, very happy.
“Seriously? 12,000 monthly uniques is amazing!”
I concurred, of course.
And that’s when we found out that LNGFRM was a lab rat, the subject of an experiment
Over the past two years, Sturman had been providing AI-generated content to a handful of small sites just like ours. The goal was to see how they performed for the sites and determine whether or not his supposed ‘top secret formula’ was going to pay off.
In the event that the blogs started to experience meaningful traffic growth, Sturman said, there was only one more step required to build the ultimate AI for SEO. All that was needed was to, with the permission of each site, export/use their Google Analytics data to fine-tune the AI.
We agreed to give him access to our data, under one condition.
“What’s going to happen with your AI?” We asked Sturman.
He paused for a moment, and then replied proudly.
“Juice.”
Is AI going to kill us all?
As a stock millennial now in my late 30s, in a lot of ways, I feel pretty obsolete. My remote copywriting job for our local auto parts store is steady.
But even now, I see the (AI-generated) writing on the wall. I convinced my boss to let me use an AI copywriter to help me generate our latest Veteran’s Day local paper ad.
It was one of our better V-Day sales in recent years. And while I do credit the AI tool I used somewhat, it’s not hard to imagine a day in which I may not even need to direct the AI for it to create a compelling newspaper ad.
The same day I got a note from my boss congratulating me about the V-Day numbers, I got an email from Sturman about the launch of Juice.ai. His app, which had taken so many years to create, was finally in a paid beta and available for all to try out.
At first, I was a little saddened. I had been given an insider look at an impressive new technology. Now, I’d become an early adopter relegated to the confines of obscurity. Thousands of new blogs will go on to get the same “Juice treatment” that LNGFRM had for a couple years.
And while it’s scary to see machines get closer and closer to doing my job, as time goes by, it’s getting harder and harder for me to feel sorry for myself.
After all, if I, without much effort, could turn LNGFRM into something that paid for groceries and my car insurance, why should I feel bad about other people getting a similar opportunity?
Who cares if Juice, or any other AI, makes my career choice look old-fashioned. Highly-capable AI isn’t going anywhere, and I think as a society, we’ll need to make peace with that.
Recently, I sent a message to Derek thanking him for helping out with LNGFRM. It’s been about a week, and I haven’t heard back from him. That’s OK.
In the meantime, there are plenty of new readers coming our way.
As much as I’d love to be the hero in some I, Robot fantasy, it’s hard for me to see a downside. Maybe if the traffic stops coming in I’ll feel otherwise.