Digital music service Spotify recently acquired machine learning startup Niland to improve its recommendation and personalization technologies. In other words, Spotify wants to better connect its users to music they will like.
The heart of this concept is nothing new. Netflix and Amazon, too, use machine learning – a type of artificial intelligence (AI) in which machines learn when exposed to new data without being programmed – to suggest content and products their respective users might like. And while this ability to tap into AI – machines that perform smart, human-like tasks – to analyze internal data is increasingly common, it’s a bit more complicated when it comes to using AI to capture consumer attention externally, like, say, in search.
A brief history of AI in SEO
It's complicated in part because we don’t know a lot about how search engines use AI themselves.
RankBrain is the AI that is reportedly Google’s third most important ranking signal and has been in use since 2015. Ranking signals include all the factors search engine algorithms use to determine how sites rank. But Google has not been particularly forthcoming about what RankBrain does or how brands should optimize for it.
“It's actual use in search proceeds cautiously,” said Pete Meyers, marketing scientist at SEO software and tools company Moz. “Right now, RankBrain is more of a layer on top of the algorithm than a core reworking of it…the trick with search, for now, is that I think AI/[machine learning] is fairly subtle. It's helping [Google] solve difficult problems, like interpreting long search queries, but it's not fundamentally reshaping search results yet. It's probably going to have more impact sooner on things like image results.”
But it’s also because rankings are based on many different factors for each individual query.
“Machine learning…can detect so much more amongst the big data we’ve been talking about for so long. Could that suggest what content ranks best? You then have to ask the question: ‘What do you mean by 'rank?’” said Mike Grehan, chief marketing officer at intent-based digital marketing firm Acronym Media. “When I’m logged into my Google account and do a search, I get different results if I do that search in San Francisco to the ones I get in New York. The content that Google presents me with depends on all kinds of things from geolocation, past search history, time of day, context, intent and proximity data generally. So when you talk about ranking, it depends on who you are, where you are, etc.”
And, according to Patrick Reinhart, senior director of digital strategies at SEO platform Conductor, no brand will really ever be able to optimize its search strategy for AI because AI will instead become more human and better understand what consumers are trying to say.
“It’s more like Google is a kid that is growing up and understanding the world in a deeper way, and finally understands what we are saying to it,” Reinhart added. “Machine learning is more like talking to a person rather than an algorithm because the algorithm has become human-like.”
This also holds true for voice queries. And the only way to optimize for either is to make really good content that speaks to your users, Reinhart added. No pun intended.
Andy LaFond, group media director at advertising agency R/GA, agreed voice search – and chatbots, which themselves use machine learning – provide the greatest potential short-term opportunity for marketers in AI. That’s because chatbots can streamline and/or improve a brand’s buying and service processes and the end result is happier customers and more sales, he said.
What’s more, Google webmaster trends analyst John Mueller recently reminded marketers not to focus on optimizing for RankBrain itself, but to rather “make great sites for your users, folks”.
And that’s not bad advice – even if it isn’t particularly groundbreaking.
So what is new in AI and search?
1. Content optimization
Reinhart said some larger companies like Uber and Airbnb and are using machine learning to determine how consumers are using their sites and generating content based on that activity. (Uber did not respond to a request for comment. An Airbnb rep was not available.)
But this may not necessarily result in the best content – and it doesn’t come cheap either, Reinhart said.
For his part, Meyers said AI-based content ranges from “low-value content spinning” to “highly structured content” and advised skepticism when it comes to AI and content strategy – or even SEO overall.
“We use [machine learning] for things like helping sort how to group keywords together in [Moz keyword research tool] Keyword Explorer, so you could say we ‘use AI for SEO’, but I think that would be a gross overstatement of a system that solves one particular problem,” Meyers added.
Indeed, Meyers said many people claim to use AI to do a lot of things, but “it often ends up being a tiny piece of what they do or one step from smoke and mirrors”.
He added, “The challenge is that if an enterprise is using it effectively for something like SEO, they're going to keep that close to the vest as a competitive advantage.”
2. Content matching
AI could also help match consumers with the type of content they like, but that, too, comes with an asterisk.
Grehan said marketers could use pattern matching to discover whether certain query types prefer certain content experiences like video or text.
“But, ultimately, the end user decides what constitutes a quality experience for them personally,” Grehan said. “The answer to, ‘Who is the world’s greatest electric guitarist?’ is Jimi Hendrix, of course. Unless you’re an Eric Clapton fan. Or fan of another guitarist. Quality is subjective.”
In other words, Grehan said even if brands could use AI to get content to rank better in a search engine, it would still be poor if the end user didn’t get the right experience.
“And the less content is engaged with – the less it’s likely to rank,” Grehan added.
Grehan also pointed to Google’s recent update about its so-called Project Owl initiative, which noted human search quality raters will continue to assess the quality of search results – and Google itself released more information about how those raters should identify low quality pages.
“[A recent Search Engine Land story used] a decent analogy about quality raters being like [reviewers] in a restaurant,” Grehan said. “But more importantly, as quality is a subjective matter more easily evaluated by humans, the feedback from the quality raters helps make more rapid changes to results, where a machine learning algorithm would only see a pattern and have no regard for the meaning.”
In other words, Google is using something of a hybrid AI/human approach in ranking content – and so perhaps it’s not a bad idea for brands to do the same in creating content.
3. Predictive analysis
Ella Chinitz, group vice president of data science at digital agency SapientRazorfish, said she’d like to see a larger role for AI in predicting consumers’ future interests – potentially before they know themselves. That’s because AI can help brands better understand who individuals are and what they’re interested in. From there, brands could use AI to generate content using a model that anticipates the next steps a given customer might take. However, she said the aforementioned hybrid approach – in which the predictiveness of AI is combined with a “human layer of creative inspiration” – would still be necessary.
4. Paid search
Per LaFond, AI can also help with multiple nuts-and-bolts aspects of search campaigns, such as landing page optimization by testing many different pages to determine what drives quality score and conversion.
AI could also be used to optimize search ad copy, but, in a familiar refrain, LaFond said, “AI still can’t match human copywriters in framing a brand or product in different ways, but search will probably be the first type of ad copy that AI might write.”
Finally, AI could be used to optimize bids and budget allocations across many keywords, ad groups and campaigns.
“Google already offers tools to optimize campaigns, but not across other search platforms, such as Bing,” LaFond said. “In other words, AI can be great at optimizing to known objectives – like winning a chess game – but strategic and creative thinking still requires a human mind – for now.”
Kristoffer Belau, director of digital marketing strategy at marketing agency Firewood Marketing, agreed AI “really begins to shine” on the paid side.
“Paid search has always been a game of optimization and AI tools allow brands and agencies to maintain much larger and more efficient keyword portfolios without massive paid marketing teams,” Belau said.
Indeed, Chinitz said if brands use data across multiple sources to create a better picture of who a consumer is and to become more predictive, paid search results become increasingly relevant as well.
Belau also agreed AI must always be tempered by human intuition.
“It will quickly go off the rails and start to make decisions that are not in the best interest of a brand as it blindly chases a local maximum without insight into any macro strategy,” Belau added. “In both paid and organic search, it's better for AI to be used to inform your decisions rather than to make them itself.”
The Drum will host the first year of its US Search Awards in September with entries now being accepted.