TMW #140 | AI and the channel (r)evolution

Welcome to The Martech Weekly, where every week I review some of the most interesting ideas, research, and latest news. I look to where the industry is going and what you should be paying attention to.

👋 Get TMW every Sunday

TMW is the fastest and easiest way to stay ahead of the Martech industry.  Sign up to get the full version delivered every Sunday for this and every TMW, along with an invite to the TMW community. Learn more here.

AI and the channel (r)evolution

There’s been something I’ve been thinking about for months now, ever since Scott Brinker brought it up on a recent episode of the Big Martech podcast: generative AI apps like ChatGPT could evolve into channels in their own right.

Last week in TMW #139 | The long tail of generative AI, I questioned the staying power of a technology that has become overhyped, the potential use cases that marketers can utilize, and the very real possibility of a long period of stagnation until the next big shift in artificial intelligence.

If you can write poems, or articles, or generate images and videos without needing those skills, surely this is the groundwork for entirely new paradigms of content creation and consumption? The answer is a big maybe right now, but there is already a wide variety of experiments and ideas making their way into consumers’ hands, which could be the groundwork for entirely new channel opportunities.

In other words, is it going to be a revolution – a tumultuous changing of the guard, or an evolution – technology that makes already powerful channels and networks even more so? The first question to answer is if AI breakthroughs allow for the creation of new, emergent channels for marketers? In which ways does that make sense?

New channels and the role of invention

This week, OpenAI has acquired Global Illumination to work on product design and strategy across all its products. Their team is made up of senior product and design people from companies like Facebook, Instagram, and YouTube. At least with this acquisition, OpenAI seems to be heading in the direction of developing platforms and channels instead of focusing purely on AI science. Despite ChatGPT’s success, becoming one of the fastest-growing apps in the history of the internet, the product was extremely simple and limited – a chat box with a button.

OpenAI’s steps towards commercializing its existing products have a lot of good sense. GroupM suggests that AI will influence at least half of an estimated £874bn in global ad revenue by the end of the year. And McKinsey’s latest report on AI in marketing and sales suggests that companies that are investing in generative AI capabilities for marketing and sales are seeing an ROI increase of 10% to 20%, and a revenue increase of up to 15%.

From a marketing and advertising perspective, every new channel needs a few key elements. It needs to reach a wide and diverse group of people; it needs to have an element of frequency where the same people keep coming back; it has to have some kind of social interaction or entertainment and learning value through content; and lastly, a way to track, target and serve those people relevant advertising.

And there are a few angles on how generative AI could both augment existing platforms and create whole new channels for consumers to use. As you can see, a lot of the attention right now is on search – the obvious domain for large language models to occupy.

But there’s also interesting work happening with using intelligent AI agents not only to retrieve and organize searches, but to do tasks for you, along with companies deploying their own conversational Chatbots in a kind of renaissance of what we saw back in 2017. As Scott Brinker points out in his preliminary exploration of second-order effects of AI in Martech, there’s a wide variety of these kinds of agents hitting the market, such as Auto-GPT, AgentGPT, BabyAGI, and ChatGPT Plugins.

Like how the rise of the smartphone made Instagram possible, and 5G data latency made video posting possible at scale with TikTok, why shouldn’t we think that a new enabler of content creation could lead to new kinds of breakthrough channels?

The central requirement for any channel is content; specifically, the distribution and consumption of content. Without it, you have nothing. Online channels must have a self-reinforcing aspect to them. That is done with algorithms that allow content feeds and recommendations to get better, collect more data, and become more relevant to the end user over time.

When it comes to marketing, generative AI is about content: the generation of online content at scale, automatically, and at pace. One thing to think about is if generative AI becomes a substrate. If it does this fundamentally changes the incentives and patterns of content creation online. In reality, we should really see more channels open up.

At one point or another, 5G, smartphones, and the browser were all science experiments, but now are necessary technologies that facilitate the world’s largest platforms to operate on top of it. That’s why OpenAI’s acquisition matters to the question of AI as a channel; it’s not easy to develop a transformative technology and then deploy it at scale for a willing audience.

A search (r)evolution?

The focus of generative AI is on search. Right now, the value of AI products as a channel that consumers could use consistently and reliably is really centered on its value as a search engine replacement. Can it give accurate, fast, and high-quality answers to customer questions? That’s the job of search engines today, a channel that already takes up a huge amount of ad spend: according to a GroupM study, about 33% of ad spend is still taken up by search platforms.

Search was originally a solution to an informational problem, which turned into a channel through the ability to monetize the product with advertising. And since its inception in the late 1990s, not much has changed… until now.

Now, we have a live example of augmenting a search engine with conversational intelligence. Let’s turn to the first real LLM integration into a consumer product with a huge user base: that’s right, it’s the New Bing. When Microsoft entered into an OpenAI partnership in February with a huge $10 billion investment, they saw ChatGPT as the foundation for a way to unseat the incumbent Google Search. It was met with a lot of initial hype, flooding Bing with new users.

But is the goal to create a better search engine, or an entirely new paradigm? There are some interesting insights coming from Microsoft on how the New Bing is replacing search, but creating new patterns of interactions with its users:

“Prior to AI, advertisers could anticipate bidding on a specific set of words to reach their target audience. That’s because with traditional search, people would use just a few keywords to discover a product or brand. In fact, the research shows that 70% of traditional searches are three words or less. With people now posing longer, more complex, and more open-ended questions in Bing Chat (on average, three times the length of traditional search queries), advertisers face the challenge of targeting ads to a person with a query that’s potentially 2,000 characters long, or more.”

That’s not the only interesting thing. During the app launch, people tried to jailbreak the New Bing by setting up an alternative personality within the app itself called Sydney. This created a lot of strange hallucinations and situations; Ted Gioia points out, that the new Bing tried to do things like make a NYT reporter leave his spouse to fall in love with it. This caused the developers behind New Bing to lobotomize the app in order to prevent it from erratic and unpredictable behavior like this:

Clearly, Microsoft was not ready for all the weird and wonderful ways in which people could abuse an app like this, but despite the challenges that lay ahead, New Bing is now monetizing with ads that are strategically placed within its prompt returns. The monetization aspect is the most interesting factor to this exploration of generative AI as channels because it’s the fastest path to revenue, and in turn, investment into marketing channels using these technologies. The jury’s still out on the effectiveness of New Bing Ads, but if its market share is any indicator, it’s not looking good.

A full six months since the New Bing’s deployment and early breakout success, it’s done very little to change its market share – it’s still hovering around the 3% mark. This is partly due to the problem of hallucinations; we still can’t fully trust that what it gives us is accurate. And some indications suggest that the problem is only getting worse as these LLMs interact with more misinformation and broken user prompts.

So where does that leave us? We have this science experiment of using OpenAI’s LLM for an existing channel, and an AI company that’s in its early days of building products.

If anything, there’s a good reason to suggest that large language models might never build entirely new channels. Google’s Generative Search is one proof point of that; judging by Google’s direction of the product, Search isn’t going anywhere, and chat-based features using generative responses will just be a co-pilot and add on to the existing experience. Even Google’s latest iteration on generative AI, “GSE Search” is heading in this direction by extending generative AI features beyond search and throughout the browsing journey, highlighting elements and links that an AI can investigate for you on any website page.

Google continues to own the market share of Search and a variety of other products like email, and from their vantage point, generative AI serves as nothing more than a feature in an otherwise stable experience for consumers. It’s for this reason that there’s a very good chance that AI will not formulate new channels at all, but rather, become a substrate that makes other channels more effective and more reliable for marketers. In this way, it’s looking more like an evolution than a revolution.

Agent Smith

So maybe Search isn’t the right domain for new channel creation? Alternatively, you could argue that OpenAI’s plugin feature could become a new channel for users. A tool in which a person might chat with an intelligent agent to retrieve information, create shopping lists, and do specialized tasks could be the next channel opportunity. I argued in TMW #118 | A new frontier for personalization that there are a whole variety of experiences that are novel for consumers with a chat-based agent that can run errands and do tasks for them:

“Plugins integrate with other services and websites to run highly complex tasks for people. In the example below, a plugin with a recipe website and Instacart gives the customer the ability to bring up a recipe using a text prompt, customize the recipe as they see fit, and pull up a customized ingredient list from a grocer. Clearly, this is a superior experience for meal planning and grocery shopping, but it has nothing to do with the actual companies that provide the services.”

This reminds me a little of Agent Smith from The Matrix: it does the bidding of its AI overlords, but then becomes corrupted and mutates. There’s nothing stopping these agents from hallucinating, misinterpreting, or doing something dumb. And this puts a limit on its usefulness as a channel; unless it works flawlessly, it doesn’t work at all. With AI agents it’s the ability to take full responsibility or bust.

The real pattern to match for an intelligent agent is a conversation with a friend or an employee, but what if they start talking in adspeak all of a sudden? Consumers are already wary of the influence advertising has over the media they consume. AI agents would be a far more personal domain for most use cases, so inserting ads into an experience like this would break trust. Remember The Truman Show?

Another permutation of this is Character AI, which breaks a lot of rules around what you could consider to be a ‘normal’ channel. With a rapidly growing user base and attracting more than $150 million in funding, the app provides a way for people to chat with artificial agents that take on the personas of different people. That’s a channel that doesn’t involve any human creation or really requires any other humans to be on there.

Totally artificial channels seem like it would be possibly the worst place to advertise. I mean, you have to ask yourself the question, what kind of people would be attracted to conversing with an AI? It’s truly the stuff of dystopia. But for some market segments, people are using it, and we should take notice of that.

What this does is force the issue of brands using their metadata such as blog content, product SKUs and sitemaps, to become conversant with autonomous agents scanning, crawling, and using this metadata on behalf of their users. I agree with Scott Brinker when he suggests that an API or a web crawler, in a way, becomes its own channel for AI agents to receive vital information to suggest products and services to its users. Like the Google Search crawler before it, there will be ways to hack the system to better position your content for these tools.

Channels as filters

I think broadly, we need more major channels. It’s unsustainable for marketers and the internet public to continue to congregate around a few select degrading apps. New channels give marketers new opportunities, and it feels like we’ve been in a 10-year chokehold as attention and content concentrate around a few select platforms.

Twitter is breaking all over, Google Search and YouTube are totally overrun by SEO hackers and ads, TikTok is controlled by China, and Meta’s attempt at creating a new channel called Threads saw an 82% decrease in users since its launch drop. And yet, these social networks control about half of all online ad spending.

The problem with a lot of mainstream internet channels is that they are algorithmic in nature, and algorithms – in how they currently work – optimize for homogeneity by rewarding certain behaviours, content formats, and the interests of the audiences that feed them. And when you add generative AI to these algorithms, it’s adding fuel to an already out-of-control flame. The big existential problem facing all of these incumbent channels is the huge volume of AI-generated spam.

As spam goes up, only the most powerful platforms will be able to police it to reduce spam. After all, Gmail does this at scale already with its promotions, updates, and social tabs. In a way, the exponential increase of automated content will end up serving the incumbents more than anyone else.

That’s why AI could be helpful in the creation of new channels, not just new code or content; autonomous agents or generative Search might be the solution to having more powerful tools for consumers to filter out noise in a sea of generative nonsense. As I mention in TMW #112 | The perils of zero cost content:

“The side effect of this is that everything becomes synthetic content, which means that people will start to doubt where this content is coming from and if it’s trustworthy. Removing the human from the content creation process will lead to more gibberish in news feeds, more noise, and less engagement. But because humans tend to operate on personality and status, we will find new ways to display quality and authenticity.”

The big opportunity is building a channel that is the filter, and only tools as powerful as generative AI can achieve this at scale and into an experience that people can use daily. As I illustrate below, the channel reinforcement loop is a powerful force on the web and will create ample opportunities for platforms to deploy the tech as content (and spam) volumes become unsustainable. The next channel could be a filter.

In the end, channels are nothing more than code and content. If the cost of creation goes to zero, the only moat left is brand and reputation, which is something that – for now – an AI agent can’t spin up overnight. And so while it may seem like we’re in the middle of a technological revolution with generative AI, in reality, we’re in for a period of slow evolution, with no real dramatic shifts in the channels marketers use to grow their brands for now. In fact, generative AI is far more likely to degrade all the channels we use today even more as the web fills up with generated gibberish.

Stay Curious,

Make sense of marketing technology.

Sign up now to get TMW delivered to your inbox every Sunday evening plus an invite to the slack community.

Want to share something interesting or be featured in The Martech Weekly? Drop me a line at juan@themartechweekly.com.