Generative AI is going through its more-is-better phase.
More ads! More content! More automation! More productivity!
But pumping out content in bulk isn’t what marketers should be using AI for, said Anda Gansca, CEO and founder of Knotch, a platform that helps brands analyze their content based on performance tied to specific business objectives, usually brand lift, engagement and awareness.
The more valuable proposition for generative AI, she said, is to quickly optimize existing content based on what’s working, including automatically tweaking low-performing content to make it perform better.
Knotch launched a revamped version of its platform called Knotch One that combines content measurement with generative-AI-powered content optimization. The platform uses the insights Knotch gathers about content performance to optimize the good stuff and produce more of what’s resonating and less of what isn’t.
“It might be surprising to hear,” Gansca said, “but a mature content strategy usually involves producing radically less content – way fewer of what we call ‘random acts of content.’”
The Knotch approach to measurement
To determine the effectiveness of content, Knotch collects as much quantitative data as possible, including word count, keywords, topic, tone, traffic volume, time spent, scroll depth, referral source and device type.
Knotch also measures sentiment tied to the customer journey across a brand’s entire website.
Did someone read two blog posts and then bounce? Did they hit a product page after clicking from a product review and then convert? Did they find the content they were served relevant, useful – or useless?
It’s important to track these factors and understand why people take the actions they do, because engagement metrics can be deceptive.
When someone spends a long time scrolling through a piece of content, for example, that could mean they’re getting value out of it – or it could mean they’re scrolling out of frustration because they got lured in by clickbait and can’t find the promised information.
That latter scenario is not what you want your generative AI model to optimize toward, said Andrew Bolton, chief customer officer at Knotch.
“So many brands mistake traffic for performance,” Bolton said. “If everything gets trained on the fallacy, it just amplifies the problem.”
Meet Knotch’s AI model
Knotch’s AI model is agnostic, according to Gansca, meaning that a client can apply any large or small language model they prefer.
But most use an out-of-the-box version built on ChatGPT Enterprise. The model has been “pre-trained” using a broad mix of sentiment analysis, customer journey tracking, content attributes and general performance benchmarks.
Knotch then further trains the model based on the brand’s own guidelines, content and performance data. These are private models that don’t communicate with the outside world.
From there, the platform can detect what content is popping in real time (and why) and diagnose content that isn’t doing as well. Marketers get recommendations about how to fix poor-performing content – whether that’s using a different tone or trying a different call to action – and have an opportunity to make edits and tweaks before sending new versions out into the world.
The AI model can also automatically “atomize” content into smaller pieces that brands can use across different platforms, channels and contexts, Bolton said.
“We think of it as a self-optimizing cycle,” he said. “Everything that gets published gets measured again, and so we keep learning more about what works and what doesn’t.”
‘Data-backed decisions’
Merchant services company Square, which has been working with Knotch since 2020, was one of the earliest clients to start testing Knotch’s new set of generative AI tools.
One of the first challenges it tackled was to better manage and optimize its existing content, said Mallory Russell, Square’s VP of global content and web.
Square produces and distributes a lot of content. It’s got its own online publication and content hub called The Bottom Line, which features research, case studies, financial tips and information about how to operate a business.
Using Knotch One, Square was quickly able to identify examples of high-traffic content where there was an opportunity to improve engagement, Russell said. Square took Knotch’s recommendations for what aspects to optimize – which “saved a ton of time,” Russell said – but also made edits of its own before deploying.
Next up, Square plans to use Knotch’s generative AI to create multiple versions of existing content, including breaking down long reports into shorter articles, social posts and emails that are appropriate for different distribution channels.
“All of this helps us make data-backed decisions about our content,” Russell said, “which improves performance and gives us more time back to think about what’s valuable for our sellers and what can drive impact for the business.”
