Optimizing and preparing for SGE: strategies for site owners

machine learning code on a screen
AI- Google

I wrote recently about Google’s limited rollout of SGE (AI-powered search answers) and how it’s going to shape the search arena in the years to come. tl;dr – 92% of all brand and product questions are answered by SGE. Researchers expect a resulting “erosion of traffic levels”.

So, let’s dive in to some strategy for how publishers can still get their content and brands inside SGE. If you can’t rely on click-throughs to your landing page, the least you can do is get Google to say it for you.

The first thing to understand, fundamentally, is that SGE doesn’t magic information from thin air. It’s derived from a Large Language Model (LLM). LLMs are trained on data (duh) and Google’s LLM is based on their proprietary Transformer model, first developed in 2017. At it’s core, an LLM is a predictive engine. Using training data it’s fed, it predicts the next most statistically likely word to form intelligible responses. Most of the time anyway.

As an aside, words scoring highly on predictive analyses is also how AI content detection tools like Originality.ai or Education Plagiarism tools work. If a certain sequence of words follow a statistically likely pattern as would be generated by an LLM, said tools will determine you’ve generated the content via AI means. Google themselves have publicly joined the effort to identify AI content at scale, and recent rollouts in the March Spam Update suggest they have succeeded to some degree.

The second element to consider with LLMs is relational learning. Think parent and child category relationships. Recognizing the patterns and connected-ness of different topics. In practice, that means the machines powering SGE need to be ‘taught’ that a piece of content, product or brand is connected with, recognized within, and a trustworthy component of the answer being generated.

So, in short, LLMs like SGE present content by extrapolating from training data and producing predictive responses. The likelihood of a response is determined by relational connections.

It’s easy to see an example of that in action by trying it out a query in SGE for best mail sequencers:

We can see an example in this screenshot, where the list of best email sequencers is derived from training data taken from the URLs/pages listed underneath. In fact, the exact copy (just slightly re-arranged) from Mixmax’s own site features as both the top response and the benefits answer that SGE displays:

To add even more sting in the tail, because Mixmax produced (what SGE/Google determines as) a comprehensive article covering the options, it also takes their description of YesWare – a competitor to Mixmax – and displays that in the SGE. Described as best for small businesses… ouch. Maybe the enterprise sales team over at YesWare would disagree with that one.

So, first thing we can determine is that in order for your product or service to be displayed in a SGE result, it needs to be part of the LLM’s training data.

If we revert back to non-SGE Google and input the same query, we also see something interesting:

The number one organic result is user-generated content (UGC) via Reddit. Mixmax’s article is actually the second result.

Authoritas’ analysis of SGE questions and answers found that the majority of the time, the number one organic result was used as the basis of the SGE answer. Given that Reddit has a high domain authority and currently dominates Google search results, why is it that Reddit was not cited in the SGE results?

The simplest answer is E-E-A-T, with an emphasis on the Expert. It’s reasonable to conclude that Mixmax has greater expertise on email sequencers than some randoms on Reddit. Maybe UGC holds relatively little weight overall in SGE.

Content strategies to deal with Google SGE

If you want to head off SGE as a product/brand and get yourself in the results, these examples give you a clue of how to do it. But I’ll also spell it out nice and clear:

  1. Build clear, simple and engaging content that resonates with core queries related to your product. Do that by taking a look at the keywords and keyphrases that bring you traffic right now. If you’re a named SaaS product that’s bringing traffic via another brand’s keyword, consider how you can widen that content to hit specifically questions related to your product.
  2. Make efforts to review the marketplace and alternatives to your product, and make articles about what people can find out there. We can see exactly that in Mixmax’s article. They’re selling and describing their own product, but by writing about competitors, other options, and how they work, they get to explain the marketplace to Google in SGE results. Sneaky actors could, of course, manipulate how competitors are described to better position their own software to potential customers.
    ^n.b. I realise on first glance this second point conflicts (write about alternatives) with the first (write about your own stuff), but it’s more about broadening your approach so your content is included in training data, instead of relying on coincidental keywords to funnel visitors to your site

Overall, this starts with a content audit. Go ahead and review your keywords and top landing pages and see if any can be improved or updated to be included in Google’s broader training data. Optimize it using schema to be included in citations. If you don’t have it, create it.

Wider brand strategies

More widely, going back to the idea of ‘connected-ness’ and parent to child conceptual links, brands and product owners need to make efforts to teach the LLMs that your name and content is part of the story that SGE creates.

The more your company is referred to in third-party areas, the more general brand awareness you have, and the wider penetration of your name and material, the more likely you’ll be included in SGE results.

This is where publishers might be tempted to fall to traditional link building strategies, but simple links aren’t likely to be enough, as they haven’t been enough for a while now (especially paid, in bulk links).

There are a few different strategies you can consider here:

  1. Collaborations and Partnerships that sync up with your brand. Things like getting people to speak about you, attending and featuring in events, appearing on podcasts, being interviewed and generally widening your digital footprint.
  2. Schema and Knowledge Graphs to mark up your content in a way that’s easily consumed by LLMs. If you’re not doing schema already, you’re missing out. Directory listings in relevant publications and entries on Google Maps would also likely help here to bulk out that profile.
  3. Digital PR, creating stories and content get that picked up by high authority sites, news sites and major blogs will be a huge boost to SGEs ability to integrate your content/brand/product
  4. Guest Blogging to get those links, brand mentions and penetration across other notable and reputable sites.

Getting your organic search result to the top of Google will hugely boost your ability to feature in SGE, but as we’ve seen in examples above, it’s not always the number one results that gets consumed and regurgitated by SGE.

Teach the LLMs that you’re important, connected to the queries you want to be shown for, and a good source of data, and you’ll get ahead of the SGE traffic drop-off we’re all expecting.

GET IN TOUCH

Speak with our team

Planning how to protect and boost your brand and product traffic with SGE incoming? Our team can help.

Stay ahead of the game
Receive regular insights, expert advice and analysis direct to your inbox.
icon

One response

Leave a Reply

Your email address will not be published. Required fields are marked *