We’re making contextual advertising smarter with content clusters

Today, we’re going to be a little bit hipster and say, “We were doing contextual advertising before it was cool.”

With third-party cookies sending behavioral advertising (advertising based on individual users on a website) into a tailspin of uncertainty, everyone is looking back to contextual (advertising based on the content of the page). 

Over the last few years, huge publishers like Conde Nast and Vox launched contextual targeting systems.

And we’re over here like, “Yes. This is what we’ve been saying all along!”

For years, we’ve been investing in the contextual advertising game, matching up CafeMedia publishers’ content with high-paying advertising campaigns thanks to our artificial intelligence platform Marmalade that we built over four years ago and have been refining ever since.

Because contextual advertising does not rely on third-party cookies at all, it won’t be impacted when they go away. And we’re launching new tools like content clusters to make contextual advertising even better! 

4 key takeaways:

• While many major publishers and ad providers are just now ramping up their contextual targeting systems, we built our own artificial intelligence platform, Marmalade, years ago.
• Marmalade helps advertisers reach the right audience on CafeMedia publishers’ content by understanding the topic of posts AND the mindset of readers.
• We’re doing exciting new things with content clusters that can actually deliver better results than third-party cookies!
• And we have lots more contextual advertising improvements in the works this year.

You’ll walk away from this post with a better understanding of how contextual advertising works, how we’re building smarter targeting capabilities for advertisers, and why CafeMedia publishers have a head start on the rest of the industry!

How contextual targeting works today

Based on the powerful IBM Watson AI, Marmalade provides us with incredibly robust contextual data on every word published across thousands of CafeMedia publisher sites. (The English language has 470,000+ words, and we can target any of them.) 

Our ability to finely target through specific keywords is called “precision targeting”

We can write code to identify every single article that contains a specific keyword. For example, we can find every single article that mentions “motorcycles” or a “tape measure”. 

But it doesn’t end there. 

Keywords can be a blunt tool for targeting, as evidenced by the industry’s typical approach to brand safety — where the word “killer” can be flagged as problematic, whether it’s talking about “Gruesome killer strikes again!” or “These killer brownies will make your day!”

With the help of natural language processing, our data science team has been able to standardize language across many articles, making it more searchable and more attuned to meaning, rather than words taken out of context. 

For example, they’ve coded over 4 million recipes’ ingredients, allowing advertisers to target “bell pepper” even when a recipe calls for “sweet pepper” and to know that “1 teaspoon of salt” and “a pinch of salt” are both referring to salt as an ingredient. 

We also use language to group content together in meaningful ways. 

For example, we group content into “concepts”, allowing advertisers to target broader ideas such as “kids’ crafts activities”.

And we can write code that combines multiple keywords to build custom channels for advertisers, like content labeled as “New Home”, “Ready for Adventure”, or “Financial Independence”. 

Content clusters are the next evolution for contextual targeting

As advertisers plan for a future without third-party cookies, Marmalade is a powerful tool that’s allowed us to build our reputation as experts in contextual targeting.

We’ve been able to demonstrate to advertisers that they can reach the right audience across the high-quality sites in our network, and they’re turning to us for guidance. 

For example, a health food brand turned to us when looking for people in a self-improvement mindset — and we created custom targeting channels that span verticals like health and fitness, personal finance, and beauty. Retail brands looking to tap into Mother’s Day celebrations turned to us to create custom targeting that includes gift lists, kids’ activities and crafts, relevant recipes and entertainment content.

Now, we’re in the process of taking Marmalade a big step forward. 

Evolving our “content grouping” methodology, we’re creating innovative content clusters that allow advertisers to target based on both underlying meaning and intent.

Leaping beyond meaning, to behavior

Currently, our contextual targeting is based on semantics (i.e. language and language similarity). 

Today’s contextual targeting is based purely on language

While this is powerful in its own right, there’s often a missing piece of “related content”. 

For example, furniture advertisers are lucky to intercept people as they’re reading home decor content, but there’s so much opportunity to reach people through content that might not apply to furniture buyers directly but is part of furniture buying behavior — like browsing art content with an eye for new pieces to display. 

Our new content clusters go beyond semantics — they’re based on behavior, too. Where does a reader go after they’ve browsed home decor content? When a reader is bored on a Friday night and is looking for the perfect movie recommendation, where do they turn when they broaden their search to other activities as well? 

Behavioral-based clustering uncovers meaning that’s not captured by words on the page alone.

Content clusters are evolving to capture more customer actions and intentions

It helps them reach people who are reading content with intent — who are in the right mindset to be open to an advertiser’s message. Going beyond semantics also expands the breadth of relevant content, increasing the possibility that a reader will encounter the right message at the right time of their journey.

How content clusters work

Behavioral content clustering works by identifying commonalities among the articles that a reader is visiting. Today, this is done using third-party cookies, but in the future, identity, FLoC, and other developing technologies will allow this to continue.

With the help of machine learning, these commonalities are transformed into unique dimensions, on which every single article gets a score. Each article’s score, relative to another article, determines how closely they are related. 

We can create a map of an article’s relative “location” based on its score, and articles that score similarly will appear closer together, indicating that they reach common reader interests.

An article about the latest fitness craze might score similarly to another article about a publisher’s daily exercise routine. 

An article about a Keto recipe won’t be right next to these but could be fairly close because there’s a lot of overlap between readers who are interested in fitness and readers who are interested in Keto. 

But an article about names for pets will probably be much further away because jumping from a Keto recipe to an article about a new pet is not a typical course of events when browsing.

This content “map” (fancy name = “cartesian plane”) allows us to target a reader’s mindset and behavior, with high accuracy, without knowing any personally-identifiable information about them. 

We are able to target broadly (for example, “people who are health-conscious”) and more granularly (“people who are on the Keto diet”) by zooming in or zooming out of the map.

Content clusters are the future for advertisers and publishers!

Content clusters can mirror the third-party cookie — and even improve targeting

Contextual clusters based on behavior can replicate (or improve!) the experience that advertisers are looking for when they use third-party cookies. 

And they come without the downside of completely irrelevant context, which can detract from an advertiser’s message. 

Brand safety is such a huge concern for advertisers these days, but even beyond that, a readers’ mindset is critical. 

Say a bicycle brand wants to reach men ages 18-34 who are outdoorsy. Third-party cookies may allow them to identify that target person currently visiting a page with a cheesecake recipe. But when you’re in the mindset of “I’m going to bake a cheesecake!”, you may not be thinking about purchasing a bicycle — so even though this person is the right audience, the message won’t be effective. Content clusters help avoid a right-audience-wrong-mindset mismatch!

Coupled with our long-standing reputation in the contextual space, content clusters are a key step to making sure that advertisers keep spending with CafeMedia publishers on the open web, even when third-party cookies go away. 

Content clusters bring developing trends to light 

Beyond the obvious benefits, behavioral clusters act as great signals of emerging cultural shifts that can help inform content. 

For example, is our content cluster map seeing the distance between business and travel content shrinking? This could indicate that the business/leisure travel combo is coming back post-Covid. Based on these “map signals”, we can provide advertisers (and CafeMedia publishers!) valuable early insights!

In fact, we’ve already been using contextual trends to strengthen relationships with advertisers throughout the COVID-19 pandemic by providing brands with insights on ever-shifting consumer behavior and sentiment — and in a totally privacy-conscious way! 

Our contextual systems allowed us to leverage better, more actionable data than many other publishers, without relying on personal information. These insights into upcoming trends help shape advertiser messaging and help position us as a strategic advisor to brands. 

In many ways, our ability to pivot and prove value to advertisers throughout the last year has served as a proving ground for a future without third-party cookies!

What’s next with contextual advertising?

Our work to bring value to advertisers in the post-third-party cookie world doesn’t end with behavior-based contextual data. 

Not to give too much away, but these are just a few things on the horizon…  

Marrying contextual advertising with SEO keywords

A lot of this content cluster talk may sound familiar to publishers because they’re already thinking along these lines in terms of SEO. 

We’re exploring ways to help advertisers target articles based on SEO keywords, so they’re paying for the most relevant and most impactful traffic hitting a publisher’s site. 

The result is a more valuable impression for them and more revenue for publishers! 

Creating data bunkers and clean rooms 

We’re working hard to make sure that advertisers who want to use their first-party data to supplement our contextual advertising will be able to do so safely and easily. 

“Data bunkers” or “clean rooms” are special mechanisms that create a secure “warehouse” where a user’s data is first disguised and then matched with the advertiser’s data, helping the advertiser target the perfect audience for their ad, without knowing specifics about users.

Supplementing Marmalade data with primary research 

And we consistently use our expert research team’s survey data to help advertisers (and CafeMedia publishers) better understand readers. 

Because these insights are unique and proprietary to CafeMedia, they reinforce our value as a partner to advertisers who are always looking for a competitive edge!

—–––––

As we work across the ad industry to reimagine what digital advertising can look like from a privacy-first perspective, contextual advertising is one powerful tool in our box.

And we’re excited to continue making it smarter and more valuable for CafeMedia publishers and our advertising partners!