I have been following the generative AI search with active interest for a while now. Largely because from a marketing perspective it is still a bit of a puzzle. The more I read about generative AI search, its new capabilities, I'm amazed at the speed at which this technology is maturing. Yet, I am left with many questions and insufficient answers.
So, let's start from what we know so far, and see what we can do as marketers to become discoverable through Gen AI search.
Looking at online search from the lens of gen AI search language models
Based on my research so far, I think the best way to approach this topic is to look at it from the perspective of the large language models or LLMs that power Gen AI search; the user who is searching using a prompt; and us, the marketers.
Starting with LLMs. First the language models have to understand the context of the user's prompt. Basically, what is the intent behind what the user is searching for. It would use machine learning to identify websites that can provide information which is relevant to the intent behind the prompt. Then, it uses its language processing capabilities to understand the content of the website. This allows the gen AI search to identify the information that is most relevant to the intent behind the prompt.
Finally, Gen AI search uses deep learning frameworks to generate a response to the user. It goes beyond delivering a response, though. It also suggests questions related to the prompt.
Wait, all this is not as easy it's made out to be. There's so much more that these LLMs are doing that needs to be understood better. Once the LLMs gather understanding the context and the intent behind the prompt, they use machine learning to identify the websites. But how? Well, it is a multi-variate equation. The algorithm looks at the websites that have matching keywords, synonyms and related terms. Then, it goes beyond keyword matching and considers the context of the words on the website.
How does it do that? Well, gen AI search language models have been trained on massive text data that enable them to grasp the meaning and relationships between words, phrases, and sentences. This makes it possible for LLMs to gain understanding of context and theme of the website. These models have also been trained to identify named entities, i.e. people, organizations, locations, and products or services.
Now that the gen AI search algo has understood the user's context and intent and identified relevant websites relating to that prompt, it also considers certain technical aspects of the websites. It analyzes whether the website is adequately structured, has proper metadata, the quality of backlinks, page loading speed, mobile-friendliness, and more. It also looks at certain qualitative factors like if the website content gives signals of expertise, depth of knowledge, factual accuracy, technical language used, and references to credible sources. It determines the overall coherence and structure of information the website is providing.
LLMs evaluate website content based on other parameters like - whether the content has unique and up-to-date information; and it is mapped according to the user journey to give the best possible experience. They also factor in the user engagement metrics like click-through rates, dwell time and bounce rates on search result pages. Let me explain what these metrics mean. LLMs look at the data of which links got most clicks, how much time did the user spend on a website before returning back to the search results, and how much time has passed before they closed the search window also plays a role in generating the best possible response to a prompt.
If we take a few steps back, the content quality, trustworthiness, authority, performance, mobile-friendliness, and search metrics are some of the key factors the Gen AI search models look at before generating a response for the users.
Looking at online search from the lens of users
Users, on the other hand, sit at the centre of this entire play of online search and information discovery. Generative AI search engines like Google Gemini and Bing are designed specifically for one purpose - shorten the online search journeys for an average user and provide the information they are looking for. But then what is the difference between generative AI search and plan-vanilla search that we have been used to? For one, its chatbot-esque interface makes online search conversational. It is closer to how humans interact with each other. Since it is conversational, it can be personalized for each user in tonality. That is somewhat work-in-progress but the current output indicates that we are headed in that direction.
So, for an average user, it is about speed, relevance, accuracy, and experience. They want the information that they are seeking to be delivered quickly; be almost exact to the intent behind their prompt; be factually correct; and be accessible in simple, cohesive, and seamless way on a device of their choice. Their demographics, preferences and online behaviour become a major determinant of whether a particular online search journey was meaningful to them or not.
Looking at online search from the lens of marketers
Now, whether you are an experience marketer or a likely marketer, you are looking to tell story of your business to expand its brand visibility, keep your customers engaged, lead them to conversion, and nurture a warm relationship with them. To do this, you need to get your house in order. By your house I mean the understanding of your customer personas, the website, content, tone and voice of that content, a typical journey of your customer throughout the sales cycle, and make it all seamless and entertaining. Sounds easy, right? Nope. Far from it.
If you are marketing for an established business, you have a lot of information and data sitting within the organization that can get you started with having a good understanding of your customer personas and preferences. If you are a startup or an established business new to the digital marketing space, you can carve out an ideal persona and refine it by running multiple campaigns over 12-18 months to gain a better understanding.
With this understanding, you can start aligning with the user's perspective I talked about earlier. Review and/or develop website content according to the user journey through the funnel. Technically structure your website according to the user journey; create a proper website schema; inter-link relevant pages to inform, educate, engage and convert customers; use different content formats - blogs, videos, graphics, infographics; keep the content fresh and optimize everything with proper metadata; and make sure your website loads fast and is mobile-ready.
Of course, all that website content also gets promoted through e-mail marketing and social media channels. Get online brand mentions and social media engagement by posting website content combined with attractive visuals. Collectively, all these efforts will increase chances of your website content ranking higer and start appearing in generative search responses.
I have just listed a long list of activities that can be very overwhelming for any marketer to digest and implement. Thankfully, there are free analytics tools that can support and make this journey easier and meaningful. But what to measure and why?
Preparing to rank in Gen AI search responses
The best way to look at all this is from a user's journey perspective. Because, it is fundamental to how the users will look at your products or services no matter which stage of the funnel they are at. Also, that will give confidence to generative AI search models to consider your website content in generating response to your prompt.
You can start by building a list of search keywords that are relevant to your business. Make it exhaustive, make sure that you have a healthy mix of short tail and long tail keywords, and analyze the search volume data to narrow down to a targeted list of money keywords that your target audience is likely to search. It is very good indicator of the online demand for your products or services. It will also give insights into the user intent.
Compare your website content and structure to see if there any gaps in terms of website content and the shortlisted keywords. If there are keywords that your website is not talking about, then you need to develop content around those keywords. It is preferable to have a dedicated page for each keyword.
Extend this exercise to mapping your website content to a typical user journey that is relevant your industry and business domain. This might show gaps in your content strategy - you may have missed out developing content for a particular stage of the funnel. It will flow into your content development plan.
Then look at the depth of knowledge your content has by looking at the user sessions of each page versus the search volume of the keyword that page is catering to. You can also compare the volume of sessions with your competitors' pages to get a sense of where your page sits in the market. But you would need paid tools for such kind of information. If you don't have the budget, or have a small team, or just starting out, plan it for a later stage.
Bucket your list of keywords into broader categories to develop overarching themes. These themes would be specific to the intent behind a particular search term. Analyze the total sessions that the pages under each theme are getting vis-à-vis the total search volume. This will give you insights into how relevant is your website content to the user intent.
Now that you have understanding of the content depth, relevancy, and overarching themes, it is also necessary to look at how the users engaging with each content. Scroll depth will give you understanding of how far the user is scrolling. If they are not scrolling beyond a point, what can be done? Look at sessions per user to get understanding how many of them are returning to the website each month.
Bounce rate is also a good indicator of how relevant a particular page is for the users. Combine it with path to acquisition metric to understand how the user reached that page and which page did it move to? Or did the user closed the window altogether?
Look at average session duration to get the sense of how much time an average user is spending on a page. Slice it with new and returning users to understand which type of user needs to be catered to in terms of refreshing or updating the content. Is it even needed?
Please note that these metrics are mostly relevant for non-converting pages. Conversion pages have a single purpose - converting the user. These pages can have a high bounce rate, low scroll depth, and less session duration as we want the user to quickly submit the form.
Using this approach to look at your website structure and content can help with better search engine crawlability, improving your search rankings, and remaining relevant to users as well as gen AI search algorithms. This approach is likely to impact your current content strategy. But, leveraging the data that is currently available to improve your content depth and relevance, you can pivot and align your website content with the journey of your ideal customer and the Gen AI search LLMs.
It is important to bear in mind that Gen AI search models are rapidly getting better in terms of understanding the user intent and generating high-quality responses. This will soon allow them to personalize and hyper-personalize content to each user.
Marketers need to take advantage of the website and social media data available to them, to make necessary tweaks to their website, content and overall marketing strategies. This can help make their website, content and overall brand identity more discoverable and rank in gen AI search responses. It will also help align their website and content with the user journeys, create opportunities for higher engagement, and lead them to conversion.
Comments