Tag : digital marketing

Everyone knows what search engine optimization (SEO) is, and many companies take great efforts to ensure they show up near the top of organic search results and benefit from the resulting traffic which comes at no unit cost. Traditional organic search results are slowly being replaced, however, with a lot of the focus being shifted to what Google calls a search generative experience (SGE; note that this is synonymous with AI Overview on Google Search, and the SGE is titled AI Overview on the search results page). It is widely accepted that as SGE becomes more prevalent, traffic to websites from legacy organic search results will decrease. This is due to two factors:

• Fewer people will click on organic search links – or any links – when SGE is present.
• The webpage links referenced in an SGE answer have lower clickthrough than standard organic search links.

In other words, some searchers will see the answer provided by the AI overview, accept it as accurate and sufficient, and take no further action. These searchers who would have clicked through to something else in the past may simply not click on anything. The bounce rate of SERPs likely increases markedly when SGE is present. SGE also contains its own reference links, which will inevitably cannibalize some legacy organic search traffic. Data from FirstPageSage shows that the result is not dramatic (yet), but just the first link in the AI overview is already garnering 9.4% of clicks. While this compares to 39.8% for a top search position result or 42.9% for a rich snippet result when SGE results are not present, it still has to come from somewhere, and the FirstPageSage data shows SGE is now appearing on 31% of SERPs.

In this post, we’ll address what life science marketers can do, and should be doing, to address the new search paradigm of Generative Engine Optimization (GEO).

How Generative Engine Optimization and Search Engine Optimization Overlap

Luckily for search marketers, GEO and SEO have a lot of overlap. If you are doing well at optimizing for search, you are probably doing a fair job at optimizing for generative engines. A number of key SEO principles apply to GEO:

• Perform keyword research to ensure you are addressing popular user queries and develop content targeting those keywords.
• The content you create should be helpful, reliable content that demonstrates experience, expertise, authoritativeness, and trustworthiness (what Google calls E-E-A-T).
• Ensure you are signaling the relevance of your content through optimization of on-site and on-page factors (copy, metadata, schema, etc.) for targeted keywords.
• Further signal the relevance of your website and content through off-site link building.
• Ensure all your content is getting indexed.

Increasing the quantity of content, using clear language, and using technical language when appropriate also improve performance in both generative and organic search results. Other practices to improve the authority of a page or domain such as backlinking almost certainly play a role in GEO as well, as search AIs pick up on these signals (if not directly, then through their own understanding of organic search ranks).

There is further overlap if your goal in creating content is to get it seen by the maximum number of people instead of solely driving traffic to your website. In that case, disseminate your content as much as possible. While AI Overviews are not citing Reddit and other discussion forums as much as they once did, the more places your content lives, the more of a chance you’ll have that the AI will cite one of them, especially if your website itself is not well-optimized.

How GEO and SEO Differ in Practice

Optimizing for GEO is akin to specifically optimizing for rich snippets: there is additional emphasis on the content itself vs. ancillary factors. You need to pay more attention to how you provide information.

A seminal preprint paper by Pranjal Aggarwal et al uploaded to arXiv in late 2023 which coined the term generative engine optimization investigated a number of factors which they believe might help optimize for inclusion in SGE. Note that this paper has yet to pass peer review and was subject to a lot of scrutiny by SEO professionals, most intricately by Tylor Hermanson of Sandbox SEO who gave a number of compelling reasons to believe the data may be overstated, but having read the paper and a number of critiques I still think the paper contains meaningful and actionable lessons. There are two figures in this paper which I believe summarize the most interesting and useful information:

Table 1 shows how different tactics affected results. They used a metric called position-adjusted word count to measure the performance of websites in SGE before and after various GEO methods. I am more interested in this because it is an objective determination as opposed to the subjective impression metric, which basically involves feeding results into GPT-3.5 and seeing what it thinks. We can see from the results that specific types of content addition – adding quotations, statistics, or citations – have a notable impact on the position-adjusted word count for those websites. I point those out specifically not only because they have the greatest impact (along with fluency optimization), but they are not things which would necessarily be considered important if the only consideration for content creation was SEO. All the others which they tested and found to be useful – speaking clearly, fluently, technically, and authoritatively – are things which good SEO copy already needs to do. The inclusion of quotations, statistics, and citations are simply additional content.

The other interesting lesson from this paper is that the most impactful GEO methods differ based on the topic of the content.

While I would like to see this data presented the other way around – what methods are the highest performing for each category – it still makes the point. It also suggests that scientific content may receive disproportionate benefit from fluency optimization and authoritativeness. Again, those are already things which you should be factoring into your copy.

Practical Steps Life Science Marketers Should Take for GEO

If you are looking to optimize for generative engines, first ensure you are doing everything required for good SEO, as outlined above in the section of how GEO and SEO overlap. That is 80% of the job. To reiterate:

• Perform thorough keyword research to address popular and relevant queries
• Write in a way which demonstrates experience, expertise, authoritativeness, and trustworthiness (EEAT)
• Optimize of on-site and on-page factors (copy, metadata, schema, etc.) for targeted keywords to demonstrate relevance
• Further demonstrate relevance through off-site link building
• Stay on top of Google Search Console and ensure your content is getting indexed
• Write more / longer content
• Write clearly and use appropriate technical language considering the subject matter

To specifically optimize for generative search beyond normal SEO, make a note to cite your sources and include statistics and / or quotations when possible. That is the lowest-hanging fruit and where most life science marketers will be fine stopping. If you really want to deep dive into generative engine optimization, however, you can use a tool such as Market Brew’s AI Overviews Visualizer to investigate how search engines’ semantic analysis algorithms perform cluster analysis with your website content and see how content is grouped and related.

Since AI overviews decrease overall clickthrough rates, another consideration for some marketers may be getting their content into the AI overviews independent of whether the content is hosted on your website or not. In these situations, you should try to disseminate your content widely across high-reputation sources, particularly Reddit. While it is not cited in SGE as much as it used to be, having your content in multiple places still increases the probability your content will be used.

Product Companies: Don’t Forget Merchant Center Feeds

While our anecdotal data shows that shopping results aren’t yet being included much in the life sciences, they are occasionally included in other industries and it would not be surprising to see them included more frequently in the life sciences in the future. When shown, these shopping results are very prominent, so ensure your Merchant Center feeds are functioning, include as much of your product portfolio as possible, and are well optimized. (Product feed optimization is a topic for another day.)

Summary

If you want to improve the likelihood that your content will appear in AI overviews and those overviews will contain links to your website, start with SEO best practices. That will get you far in both legacy organic search, which still receives most clickthroughs, as well as in SGE. From there, ensure your content which is the target of optimization efforts cites sources and includes statistics and quotations. If you sell products, ensure you are making optimal use of product data feeds.

GEO is neither difficult nor rocket science. By taking a few relatively simple steps, you’ll improve the likelihood of being included in AI overviews.

As this is a complex and novel topic, we’ve included an FAQ below.

FAQ

Is employing current SEO best practices sufficient for good ranking in generative search?

Helpful? Yes. Sufficient? It depends.

If your products and services are relatively niche, and the questions you seek to answer with your content are likewise niche, then current SEO best practices may be sufficient. If there is a lot of competition in your field, then you may need to incorporate GEO-specific best practices into your content creation.

You can think of this similarly to how you think about SEO. If you are optimizing for niche or longer-tail terms, you might not need to do as much as you will if competing for more major, high-traffic terms. The more competition, the more you’ll likely need to do to achieve the best results. If your terms are sufficiently competitive that you are not ranking well in organic search, you should definitely not presume that whatever you are doing for SEO will reliably land you in AI overviews.

If my website has high organic search ranks, will it perform well in SGE?

I’m not sure anyone has a clear answer to this, especially since the answer still seems to be changing rapidly. Many of the studies which exist on the topic are almost a year old (an eternity in AI time).

Taking things chronologically:

• A January 2024 study by Authoritas using 1,000 terms found that “93.8% of generative links (in this dataset at least) came from sources outside the top-ranking organic domains. With only 4.5% of generative URLs directly matching a page 1 organic URL and only 1.6% showing a different URL from the same organic ranking domain.”
• A January 2024 study from seoClarity looked at the top 3 websites suggested by SGE and compared them to just the top 3 organic results on the basis of domain only. In contrast with the Authoritas study, they found that only 31% of SGE results had no domains in common with the top 3 organic results, 44% of SGE results had 1 domain in common, 24% had two domains in common, and 1% had all three domains in common. This suggests much more overlap between generative and legacy organic results, but it should be noted that it was a much smaller study of only 66 keywords.
• A January 2024 study from Varn Media, using a similar but less informative metric to Authoritas, they found 55% of SGE results had at least one link which was the same as a top-10 organic result on a given SERP. One result in the top 10 is a low bar. They did not publish the size of their study.
• A February 2024 study from SE Ranking which looked at 100,000 keywords found that SGE included at least one link from the top 10 organic search results 85.5% of the time. I don’t like this very low-bar metric, but it’s how they measured.
• A slightly more recent Authoritas study from March 2024 using 2,900 branded keywords showed that “62% of generative links […] came from sources outside the top 10 ranking organic domains. With only 20.1% of generative URLs directly matching a page 1 organic URL and only 17.9% showing a different URL from the same organic ranking domain.” Obviously branded terms are a very different beast, and it should be no surprise that SGE still references the brand / product in question when using branded terms.
• SE Ranking repeated their 100k keyword study in June 2024 and found similar results to their February study: 84.72% of AI overviews included at least one link from the top 10 organic search results. Again, I don’t love this metric, but the fact that it was virtually unchanged five months after the original study is informative.
• Another seoClarity study published in August 2024 found far more overlap between legacy organic results and SGE results. Their analysis of 36,000 keywords found that one or more of the top 10 organic web results appeared in the AI Overview 99.5% of the time and 77% of AI overviews referenced links exclusively from the top 10 organic web results. Furthermore, they found that “80% of the AI Overview results contain a link to one or more of the top 3 ranking results. And when looking at just the top 1 position, the AI Overview contained a link to it almost 50% of the time.”

The most recent seoClarity study, suggesting a much greater deal of overlap between organic web results and SGE links, tracks with my recent experiences. While I would ordinarily discount my personal experiences as anecdotal, in the face of wildly different and rapidly evolving data I find them to be a useful basis of reference.

How much could my organic search traffic be impacted by SGE?

No one has any reliable metrics for that yet. Right now, I would trust FirstPageSage when they say the impact of SGE is not yet substantial, although I view their classification of it being “minimal” with some skepticism.

A lot of people like to point to a “study” posted in Search Engine Land which found a decline in organic search traffic between 18% and 64%, but it should be noted that this is not a study at all. It is simply a model based almost entirely on assumptions, and therefore should be taken with a huge grain of salt. (Also, 18-64% is a not a narrow enough range to be particularly informative regardless.)

Is SEO still worth doing?

Absolutely, hands down, SEO is still worthwhile. Legacy organic search results still receive the majority of clickthroughs on SERPs. However, as AI continues to improve, you should expect diminishing returns, as more people get their answer from AI and take no further action. It is therefore important that whatever you need to get across is being fetched by AI and displayed in SGE – regardless of whether it leads to a click or not.

I heard there is a hack to get your products cited by generative AI more often. What’s up with that?

A paper by a pair of Harvard researchers originally posted to arXiv in April 2024 titled “Manipulating Large Language Models to Increase Product Visibility” generated a lot of interest by both AI researchers and marketers looking for a cheat code to easily generate demand without any unit cost for that demand. As the paper suggests, they did find that LLMs can be manipulated to inserting specific products when the LLM is providing product recommendations. It is unrealistic that this is going to be applicable by life science marketers, however. It is a trial-and-error method involving high-volume testing of random, nonsensical text sequences added to your product’s metadata. This means that it would be nearly impossible to test on anything other than an open-source LLM which you are running an instance of yourself (and therefore able to force the re-indexing of your own content with extremely high frequency).

Another paper submitted to arXiv in June 2024 by a team of researchers from ETH Zurich titled “Adversarial Search Engine Optimization for Large Language Models” found that LLMs are vulnerable to preference manipulation through:

• Prompt injection (literally telling the LLM what to do within the content)
• Discreditation (i.e. badmouthing the competition)
• Plugin optimization (similar to the above, but guiding the LLMs to connect to a desired API from which it will then obtain information)

While preference manipulation is simpler and feasible to implement, the problem with any overtly black-hat optimization technique remains: by the time the method is found and published, LLM developers are well on their way to fixing it, making it a game of whack-a-mole which could potentially end up in your website finding itself on a blacklist. Remember when Google took action against unnatural link building and had marketers disavow links to their sites? That was not fun for many black-hat search marketers out there. BioBM never recommends black-hat tactics for both their impermanence, likelihood of backfiring, and ethical reasons. There’s plenty of good things you can focus on to enhance your search optimization (and generative engine optimization) while providing a better experience for all internet users.