AI tools are like e-bikes. They get you where you need to go in half the time of a regular bike, but they don’t remove all the work. You still need to steer the vehicle to get to your destination in one piece.
Artificial intelligence works much the same way. It amplifies each pedal, but you still need to know where you want to go and how to get there.
As the technology progresses at accelerating speeds, it’s important agencies don't:
Shy away from AI (and subsequently fall behind in terms of output), or
Rely too heavily on the technology, sacrificing raw creativity in the name of productivity.
The more you work with AI, the more you come to understand where it works well and where it falls flat. Which begs the question, where should an agency draw the line when it comes to AI usage?
For an in-depth look at how agencies are using AI today, we sat down with Erica Schneider (Cut the Fluff founder and editing coach), Eric Hoover (Organic Search Director at Jellyfish) and Kevin Indig (Growth Advisor at Hims, Dropbox and Reddit). Read on for their tips, and if you're already feeling inspired, check out the ways Wix Studio’s AI can transform the way you work.
First up, what are AI’s limitations?
You may have heard the saying, “AI isn’t going to take people’s jobs, humans who know AI will.” That’s because AI on its own is missing a critical factor for all things web and content creation: a clear understanding of a human POV. (Related: Here are 10 jobs AI won't replace, according to experts.)
“You should never use AI for writing anything long form. It can generate ‘okay-ish’ short form content, but in both situations, you always need to follow up with massive human intervention,” says Schneider. “AI writing usually includes an overly metaphorical style, cliche language and long-winded sentences.”
That goes double for video. You don’t want your actors sprouting new fingers mid-edit (a common AI issue creatives run into when using Gen AI). And the longer the clip or edit you ask for, the greater the margin of error.
Regardless of how you’re using AI, a human layer must oversee the final edit. Double check its outputs and flag inconsistencies.
“You can outsource things like manual labor or repetitive tasks, but you can’t outsource thought leadership,” adds Schneider. “Creative thinking requires a point of view. The machine should be doing the work, not the thinking.”
How should agencies use AI?
“AI is fantastic with everything except ‘the thing’ itself,” says Indig. “ It’s great at chatting back and forth, brainstorming and providing initial angles to dig deeper.”
Schneider uses ChatGPT-4 to extract the main points out of transcripts and to pinpoint logical gaps in drafts, and Hoover leverages Gemini for keyword research. In each case, the two demonstrate that your AI workflows can empower every stage of your workflow: research, ideation, execution, testing and iterating. Just not creativity and strategy.
It’s also great for overcoming what Indig calls the "blank page phenomena," which is when you stare at the empty doc because you don't know where to start. We’ve all been there. “AI can paint just enough on the canvas to get your shaping and structuring going,” says Indig.
Schneider recommends using AI to reverse engineer what makes certain social posts or campaigns successful. “I’ll ask it to give me a framing of how something is narratively structured, and then recreate that based on this topic in this tone. And it can do that surprisingly well,” says Schneider.
Where is human control most important?
As a helping hand, AI makes for a great teammate. Hoover suggests pitching AI internally as having a new team member on an account.
“AI gives a fresh perspective, since people often get siloed in their thoughts. Integrating AI tools in your agency workflows allows you to work towards your bigger picture by handling the details, like competitor research (at a high level), popular content, writing content briefs and doing keyword research,” says Hoover.
But you need to be careful when it comes to client work, Schneider points out. “I don't think people talk enough about having conversations with clients to see if they feel comfortable with you running your conversations through AI systems, because they might have compliance issues you're not even aware of.”
Human control is necessary for data discretion. Since much of the space is still needed and unregulated, Schneider recommends having some sort of an agreement with your clients about how you'll be using these tools and to what extent they feel comfortable sharing information.
In what areas can AI surprise us?
“My good friend Rob Lennon [who Schneider developed three AI courses with] created a phrase he calls ‘mega prompting,’ which essentially is an incredibly descriptive prompt,” says Schneider.
He gives the following specifications: ask it to perform a task, outline the specific goal you want your AI to achieve, define the formatting you want for the output, and add a touch of ‘black magic,’ which is what Lennon calls an eye-rolling, strange input.
In other words, throw a couple of curveballs at the AI and see how it responds. According to Lennon, complimenting the AI by telling it “you’re an expert, you’ve got this,” or saying something along the lines of “you’ve just taken a couple pills of Adderall so you’re really focused,” changes your output significantly. Having a library of such mega prompts at the ready can help you to a/b test different inputs to get different outputs.
“For me personally, I’m mostly surprised by the speed of development with all of these tools. New workflows develop all the time, so you need to stay ahead of it at all times,” says Indig.
What processes should agencies have in place to accommodate the rapid development of AI tools?
In response to AI’s rapid development, Jellyfish put together an AI governance team and a white paper with respective guidelines, Hoover says. “That way, we are all in agreement of how we should and shouldn't use AI.”
A lot of it seems admittedly basic at face value, he says, but it’s important to define as much as possible from the start. “We have a slide at the end of each presentation during client discussions where we outline what we strictly don’t do—things like enter proprietary information in our prompts, racially insensitive inputs or deep fakes,” adds Hoover.
You need to protect yourself, and you need to protect your teams and clients, too. On the other hand, once you’ve taken the steps to damage-proof your AI process, it's time to expand your toolkit.
“I think the most impactful process—that’s also the simplest but most overlooked—is to just play around with new tools,” says Indig. “I always urge all my clients to just try it out; spend time with them. We’re still early adopters, so now isn’t the time to find the one thing that works, now is the time to try as many things as possible.”
Sign up today to tap into Wix Studio's AI capabilities.