‘AI Boosted My Productivity 4x’: Man of Many’s Scott Purcell
Self-publishing tweets, SEO tricks and automation have transformed workflows at this innovative men’s lifestyle publisher, but GPT-generated articles are off the menu.
What becomes of the broken handed? Typically, they struggle to do up their shirts, say no to tennis matches, and waste hours awkwardly typing.
But Scott Purcell, co-founder of men’s lifestyle publisher Man of Many, says that despite nursing a broken hand, he’s still getting more things done than in the pre-AI era.
“I reckon I’d be four times as productive,” he says. “I’m working one-handed now, and I’m still doing more than I was doing before.”
“I can’t type long emails, so I’ll ask GPT to give me an email response, and it’s trained in my tone of voice. It’s a new way of thinking with a lot of tasks - can we automate this?”
Scott’s publishing business, Man of Many, is a rare success story from the media carnage of the 2010s. Focusing on men’s lifestyle and luxury content, the site has grown to a monthly audience of 2 million through its mix of high-quality content and tech experimentation, particularly with SEO.
Since the launch of ChatGPT, Scott and his team have been experimenting liberally with AI products and workflows. I got interested in Scott’s work after we both spoke at the recent humAIn conference in Sydney, where I was impressed by his passion for AI and his team’s breadth of experimentation.
This was done using off-the-shelf products, with no coding required. Scott chiefly uses OpenAI products, and combines this with process automation, to achieve results, although he is experimenting with a wide range of tools and models. If you don’t have a technical background, Scott’s work should give you confidence that you can still gain tangible benefits from experimenting with AI.
Bots trusted with tweets, but not stories
Man of Many writers are banned from using generative AI to generate copy, even for draft versions of stories.
“Even if you’re not intending to, you’ll probably plagiarize other people’s stories,” Scott says. “If you use AI to get a draft, you're bound to get something that's quite generic and similar to what other people are producing. And that's not the sort of content that Google wants to reward.”
But content transformation is a different story, and AI is used to generate tweets for Man of Many’s X account. When an article is published, it’s sent via an API to a GPT prompt that writes a tweet and tags the accounts of any brands mentioned. The prompt also includes sentiment detection – if a story takes a negative tone or is controversial, then the tweet will be escalated for human review and will not contain brand tags.
The majority of tweets publish without a human check. “I think with the way that we’ve stress tested it and gone a lot of iterations, we got it to a stage where I was comfortable with pushing it live because of the nature of what we're publishing,” Scott says. “Ninety percent of the time, it works pretty perfectly.”
So what sorts of things go wrong the other 10 percent of the time?
One issue is that it might fail to tag a brand, or otherwise tag it incorrectly. Another issue is character limits. “These models are really good at generating text and words, but if you give it a character limit, it’s incredibly stupid,” Scott says. When a tweet exceeds the character limit, the team gets a notification about the failure to post, and manually fixes it up.
So far, Scott says hallucination has not been a problem. A key step here is setting the “temperature” of the prompt to zero. Temperature is a parameter that adjusts model creativity and setting it low reduces the risk of the model making things up. As mentioned before, controversial tweets are already escalated to humans for review.
To editorialise for a moment: Man of Many’s focus on lifestyle content and positive stories reduces its legal and reputational risk. But I think a news publisher reporting on crime or corruption stories should definitely maintain closer human oversight on publishing - I have seen summarisation tools make mistakes like naming the victim of a crime as the perpetrator, for instance. The editorial processes that you use to guarantee quality of consumer-facing AI need to fit with your risk profile.
Scott says that Man of Many is actively considering adding a human to its X workflow, despite not seeing any critical issues so far. “We have strong editorial standards, so having a human checker makes sense and is easy to implement in our scheduling software.”
AI tricks for old pics
Man of Many focuses heavily on SEO to acquire an audience and recently used AI to add missing alt tags to images. For anyone who doesn’t know, alt tags are text descriptions of pictures that help visually impaired people using screen readers and boost SEO page rank.
“We started 10 years ago, before I knew what SEO was,” Scott says. This meant that that thousands of images lacked alt tags, potentially dragging down the site’s page rank. So the team identified the images and stories that were missing alt tags, ran them through GPT and populated the blank fields with generated descriptions.
“AI was able to generate alt text that is probably better than what I could write myself,” Scott says. “We were able to give it the context of the article, and it would look at both the article and the image to describe the contents of the image.”
What didn’t work: automated emails drafts
While Man of Many bans use of AI to generate editorial copy, it does use it for some other types of content creation, such as press releases.
But one experiment that didn’t work out was auto-generating emails to speed up work for the sales team when responding to clients.
“GPT would draft an email based on an advertising campaign for the sales team to use as a follow-up to the brand manager, but it was quite generic. It was a pretty good response but it’s not smart enough to know the full context of our historical relationship with that brand.
“So it created more work for the sales team and we canned it.”
Advice for AI newcomers
For a smallish publisher, Man of Many is well ahead on the AI curve. Scott says that the most important thing to do is experiment.
“Think of things that are really time consuming in their in your business. And think of ways that you can automate that using AI,” he says. “It might not even be that AI is the solution to your problem – but it might explain how to set up the solution.”
“For example, we used to have a daily report of our sponsored content. We had a junior person doing data entry every morning, typing out Google Analytics traffic for these individual articles. We’ve automated all of that now using APIs, and we used AI to teach us how to do it all.
“That saves one staff member 1.5 hours per day.”
CASE STUDY ESSENTIALS
Name: Scott Purcell
Business: Man of Many, award-winning men’s publisher
Position: Co-founder
AI Platform: OpenAI
Dumpling: Xiao Long Bao
Great to hear what is working for small publishers - this is definitely good practical advice.