Understanding the advancement of AI and its impact on search marketing

By Justin Sorrentino, Group Director, Integrated Search & Social and Linh Dang, SEO Supervisor, SOLVE(D)
 

AI stands for "Artificial Intelligence." It refers to the development of computer systems that can “look” at large amounts of data to perform tasks that would typically require human intelligence, such as recognizing speech, understanding natural language, making decisions, and solving problems. AI systems can learn from data, adapt to new inputs, and improve their performance over time without being explicitly programmed to do so. It is built to learn and advance on its own.  

The concept of Artificial Intelligence has been around for centuries, but the first modern AI system was not created until the 1950s. The term "Artificial Intelligence" was coined by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon at the Dartmouth Conference in 1956. 

Perhaps to no surprise, the first two paragraphs were written in part by OpenAI’s ChatGPT when prompted: “what is AI and when was it first developed?” It shows how far AI has come from originally completing math problems to now entering creative spaces, program coding, and having uninterrupted conversations with humans. While all of this is exciting, it does raise important questions. How far will we let these AI systems advance on their own and how will we let it impact our lives? The applications are endless, but we must draw a definitive line in the sand. Automation has its utility, but when do we need a human operator to take the wheel?  

Since the 1950s, AI has continued to evolve, as there have been countless advancements in the field, including the development of neural networks, deep learning, natural-language processing, robotics, and automotive autonomy. Within our own industry, content creation is the newest advancement as AI is currently used in a wide range of applications across speech, video, and image creation. AI technologies powering this class of use cases are often referred to as Generative AIs.  

Digging deeper, some form of AI has been used for years in Paid Search as it continuously learns from real-time data. Auction bids, positioning decisions, ad copy variation, automated reporting, and which sites receive our impressions have all been directly influenced by AI over the past decade. Since 2019, Google has also adopted the language processing model BERT in Organic Search to better understand human language, thereby returning responses that more accurately capture a user’s search intents. 

Because AI models are built by human hands, they are flawed in nature and subject to human bias. The technology also has limitations. This is why any current prediction made about AI’s impact cannot be taken as fact; we may have to change our opinions depending on future developments. As AI evolves, so will its influence within our society and our digital industry. 

The recent introduction of ChatGPT is the perfect example. Over the past two months alone it has made an immediate and astounding impact. It does, however, raise questions. Industries are including its capabilities in their own software as they try to be more automated and user friendly. Some organizations have welcomed content being created by AI while others have barred its use altogether. Wherever these advancements in AI take us, as exciting as they may seem, we must keep a watchful eye and be prepared to pivot in the future.  

As Nikola Tesla once said, “A new idea must not be judged by its immediate results.” 
 

What is ChatGPT? 

ChatGPT is an AI chatbot designed to assist with customer support, conversation, and information-gathering tasks. It utilizes natural-language processing to generate tailored responses to users’ complex questions. It was created by a company called OpenAI and was partially funded by Microsoft with a billion-dollar investment. Even though AI chatbots have been around, ChatGPT is different because it has been trained to continuously learn what the “user means” when asked a specific question only during that specific session. However, all open AI GTP models are only trained wit data until 2021. In fact, OpenAI GTP models are only trained on data for the vast majority of events up until September 2021 and does not learn from experience.¹ The responses are designed to be human-like, which is what makes it so innovative. OpenAI’s ChatGPT uses Large Language Models (LLM) such as Davinci003. LLMs are AI systems that have been trained on a massive amount of text data—such as books, articles, and websites—to understand and generate a human-like response. The more data or text that is fed into the model, the “smarter” it gets and the better the response. This technology has moved beyond predicting the next word and into predicting the entire answer of what the searcher is expecting. 

Microsoft released The New Bing 

Recently, given the success of its initial adoption, Microsoft has incorporated the GPT-4 technology into their Edge browser, and it augments search results, Bing. Microsoft was able to do this through its “Intelligent Answers” feature. When a search is done on Bing, the AI analyzes the query and responds with a conversational answer, creating a personalized and engaging Search experience. Bing also included the ChatGPT technology in their virtual assistant chatbot feature. Their chatbot functions the same way it does within Search, but in a different interface. Only time will tell if Bing is able to steal market share from Google through greater thought-provoking and humanistic Search results. 

All of this naturally has caught the eye of professionals across multiple industries. Product integration has taken center stage as companies (such as Snapchat) were fast to claim ChatGPT tech is being included in their platforms. Some online news channels have been churning out content calling ChatGPT the biggest thing since Google was invented, while others are calling it a flash in the pan. Regardless of which direction it eventually takes, its immediate impact has been impressive. 

Google answers with Bard 

Upon seeing the success of ChatGPT, Google rushed to the table with their own version called Google Bard, which uses LaMDA-2 LLM. It is a ChatGPT-like application that utilizes less computing power. It has yet to be released and is available only to an exceptionally small group of Beta testers. Google has looked to launch some new AI capabilities within Search called NORA—which stands for No One Right Answer—but the timing at this point is unknown. NORA is utilized for more difficult questions that include a preference or a choice. The goal is to provide the most useful recommendations, not necessarily the most human-like answer. There is no timeline for when Google will be incorporating any type of AI into their search engine. However, they have been investing a great deal of money and time into Google AI and have been working on conversational technology for a few years now. We do expect that whatever Google does release into the market, it will be impactful and transformative.  
 

Immediate impact and use cases of AI in Paid Search 

The eventual impact of AI on Paid Search from either Google or Microsoft is still up for discussion. As these tools generate advancements in their technology, we will need to continuously reevaluate how we are incorporating them in our everyday lives as marketers. Any Search Marketer looking to advance their clients’ businesses should be open to understanding and utilizing these new technologies.  

Here are some immediate use cases for ChatGPT within any Paid Search program:  

• New keyword research: Finding new and different terms your targeting audiences are utilizing to ask the same questions not normally found within current tools. Solve(d) has been utilizing AI and Language Modeling for years now with our Synaptic Search product. Long-tail keywords will have far greater importance in the future as searches will go from short snippets to long elaborate questions. Negative keywords will also need to be revisited as Broad/Phrase match keywords may connect to broader searches.  
   
• Help writing new ad copy and adjusting existing copy: Can provide a new twist, version, and way of speaking with your target audience.  
   
• Landing page refinement: Find LPs with richer content that may connect better with existing keywords. 
   
• Audience enhancement: Find new and existing audience behaviors that can be used to test audience overlays not only within SEM but programmatically as well.  
   

What to immediately look for in your SEM campaigns:

• Fluctuations of all front-end metrics within each engine’s campaigns, especially CPCs and impression volume. 
• Campaign-level budgets in Bing – Microsoft’s goal by releasing ChatGPT was to steal market share from Google. If successful, budgets will need to be shifted to capture this change in impression volume.  
• An influx of competitors on Bing – advertisers who originally have not been investing in Bing Search may now feel the need to do so. Bing has always been monetarily more efficient; this may change over time.  

Immediate impact and use cases for Organic Search 

As an AI language model, ChatGPT does not have an immediate impact on Organic Search results. As a generative model, however, ChatGPT excels at creating new content and summarizing complex concepts. Keep in mind that Google has stated it will flag AI-generated content as spam if it has not been fine-tuned by human hands. This does not mean it should be ignored. ChatGPT can help generate ideas, speed up drafting and brainstorming, and summarize complex concepts. Once the new content has been created, time and effort will be needed to measure its impact on rankings over time. As the Search Engines crawl new content that AI helped you write, in theory a higher quality score will result and an increase in visibility across the SERP may occur. Again, this is speculative and content on any given landing page is not the only aspect that goes into Google and Bing’s algorithm. There are a variety of factors that go into organic rankings alongside content quality, like keyword competition and overall quality and SEO strategy of the entire site.  
 

Quick and immediate applications for ChatGPT in SEO: 

• Metadata creation (similar to ad copy creation for SEM) 

• Metadata pulls for various landing pages 

• Speeding up repetitive research prep tasks by brainstorming variations of keywords and topics
   

What to immediately look for in your SEO monitoring: 

• Potential shift in rankings for certain core keywords 

• Organic CTR and traffic to pages that rely on rankings below third position 

• If Bard/Bing Chat only includes three websites as the "sources" below each response, and a lot of our traffic are from keywords in position 4 to 10 going into interior pages, we may see a decline in overall site visits 

Limitations of AI within search 
 

Low factuality and blatant falsehoods 

Large Language Models still struggle with factuality: even best-in-class models have presented falsehoods as facts while providing highly convincing responses, a phenomenon referred to in literature as “hallucination.” InstructGPT, the sibling model of ChatGPT, has a hallucination rate of 21%; in other words, the model has a 21% chance of fabricating information that is nonexistent or false.² This can cause issues across a plethora of situations, especially when dealing with pharma brands and federal regulations.  

Their reliability with facts can pose serious risks to brand and patient safety. In terms of brand safety, AI models can misrepresent disease states and conditions, especially in the case of rare diseases and new treatment pathways for which available data are limited. Users who start their exploration journey with a search can be discouraged or distracted from finding disease state education content created by a brand. It is even more dangerous when a consumer is looking for dosing information. Getting this information from AI as compared to a brand’s fact-checked and legally approved website could lead to a dangerous situation. This presents Search Marketers with an opportunity to reinforce content, especially with dosing. SEO experts will need to figure out how to ensure specific, validated information is utilized in the training models. 

To help address the lack of factual accuracy, we leverage domain-specific and fine-tuning models. We have partners that fine-tune and/or develop domain-specific datasets. This gives us the ability to utilize larger language models like OpenAI Davinci but give us results that are based on factual data.  

Lack of ethical guidance 

Data and methodologies behind AI models remain vulnerable to the influence of bias and noise, which influence the quality of search results returned by the model. As noted above, AI has been developed by humans and we are all inherently biased. Depending on their training/programming, multipurpose AI models can take on unexpected personalities and behaviors that can harm users and erode brand perception by association.  

Copyright infringement is also something users should be taking into consideration when utilizing the answers to their questions. As ChatGPT pulls from various online sources it could at times utilize images and copy from sources that are copyright protected. The US Copyright Office has begun a new initiative which will be looking at copyrighted materials being utilized by Generative AI systems and new guidelines on registering AI generated content.  

What could be the future impact? 

AI-powered Search services such as Bing Chat and Bard will continue to enhance the Search experience, permanently altering the way users communicate with Search engines. AI can also boost the viability of nontraditional search experiences on browsers, virtual assistants, and other applications, which will become increasingly complex to manage and coordinate. 

As Search Engines continue to update their framework for advertising placement and targeting, the Paid Search world will see major shifts in strategic and tactical focus. Organic Search (SEO) as a practice will also evolve to capture new content and technical guidelines from Search Engines as well as a change in user behavior. 

The recent integration of GPT-4 into Bing Search results has signaled to a select few the beginning of the end of SEM. However, we must all keep in mind that Google built their empire on the back of advertising within the SERP. Roughly 80% of Google’s revenue in 2021 came from advertising. On the Microsoft side, even with the success of Azure, Office, and Xbox, advertising is still an important component of their revenue stream. If there is any prediction we can make, it is that ChatGPT & Bard will only make Search results stronger, eventually increasing impression volume and the need to advertise within the SERP.  

References 

¹ OpenAI GTP Training Model  

² Cornell University Research