Skip to content

How Generative AI Changes Information Seeking

By Matthew Edgar · Last Updated: April 08, 2024

One way of understanding how generative AI differs from websites and search engines is by evaluating frameworks that describe how we seek out new information.

There are many frameworks explaining the information-seeking process. One of the most popular is information foraging.  Nielsen Norman Group has a great article explaining how generative AI works within the context of information foraging. In that article, they explain information foraging as three simple steps:

  1. Find: The goal of this step is to find potential sources of information. In other words, what sources exist that might answer my question?
  2. Evaluate: After finding a list of potential sources of information, users need to evaluate each potential source to see if it will answer the desired question.
  3. Aggregate: After evaluating a list of potential information sources, the next step is aggregating the collected information into a cohesive answer.

Search engines and websites help people with each of these steps—in fact, search engines are specifically designed to support this process. However, generative AI can also help people seek out information and does so in different ways than existing tools.

To formulate a cohesive strategy about generative AI, we need to understand how generative AI works within each step of information seeking. Doing so highlights where generative AI poses a threat to current SEO practices and what we may be able to do about it.

Step 1: Finding Information Sources

The goal of this step is to find potential sources of information. In other words, what sources exist that might answer my question?

Search engines are the primary example because they focus almost exclusively on returning a list of potential information sources. At the most basic level, Google, Bing, and others return a list of links that match a specified query and, typically, that list of links is highly relevant to the specified query. Search engines simplify the finding process because they help users avoid the need to filter through irrelevant or useless information.

With search engines, users can quickly move from identifying a need for information and retrieving a list of documents. A 2023 study from Backlinko found that only 15% of searchers using Google modify their initial search query, indicating a higher degree of success with the initial search results returned. That same study found only 17% of users returned to the search result after clicking to a website. A low return rate suggests users were satisfied with the results provided by Google.

It is easy to take search engines for granted. However, the process of finding information prior to search engines took far more time. Along with being more time-consuming, it was also impossible to search through as much information without a search engine. Despite some problems, search engines deliver far more relevant results than other alternatives.

Generative AI serves multiple purposes related to finding information. The prompt may ask for a specific list of information sources, such as a prompt asking for the best articles or books to read before making a big decision. Alternatively, the AI service may be designed to return a list of information sources. Whether a result of how the AI tool is built or how the prompt was constructed, in these cases, the generative AI service operates more like a search engine.

Bing’s Copilot integration into search results and Google’s Search Generative Experience (SGE) are designed to support this type of use case. Perplexity is also designed as an alternative search engine to help users find information more quickly. These services respond to a query with some combination of search results along with a summary of information.

Perplexity returns a list of results, like a search engine but also provides a generated summary answering the query

More concerning, though, is that generative AI can also help users skip the finding step altogether. This a common use-case of tools like ChatGPT, Gemini, and Claude. The tools only return information related to the prompt and do not return a list of sources for users to consider. The AI tool operating in this way allows users to avoid the work involved in deciding which website listed in search results is worth clicking on. Instead, users can see the answer directly.

This helps explain why Gartner’s study found that search traffic could decrease by 50% by 2028. Why would people use a search engine to find information sources when conversational AI tools can eliminate (or significantly reduce) this step?

Putting the question that way reframes SEO: the current goal of SEO is to get websites ranking higher in search results and make the ranking compelling enough to encourage a click. If enough users would prefer to use a generative AI response to skip the step of seeing rankings, the decrease in search traffic could be even greater than Gartner’s projection.

Step 2: Evaluating Information Sources

After finding a list of potential sources of information, users need to evaluate each potential source to see if it will answer the desired question. After typing in a query on Google, the user must scroll through the search results and open some of the websites listed to see if they are relevant and helpful.

Search engines do aid in this process. Instead of returning a list of potential sources based solely on a simple keyword match, Google focuses on returning results that match the overall topic and the likely intention of that search. Google also matches the user’s desired action (shopping vs. researching, watching a video vs. reading an article). As well, search engines filter out spam and other invalid results (for the most part, at least).

While search engines are far from perfect, these efforts mean that at least one of the first three results listed will typically be helpful and relevant. This simplifies how much evaluation of each result returned by Google in a search result is required by the end user. Backlinko’s 2023 study mentioned earlier found that 59% of searchers only click on a single result; when using Google, people do not see a need to evaluate multiple websites returned in search results.

Evaluation is where generative AI starts to differentiate itself. Generative AI does not return a list of websites that users need to open and evaluate separately. Instead, generative AI provides a fully formed response containing information related to the prompt. The sources have already been evaluated by the AI tool itself during the training process, with the AI tool selecting which sources to draw upon to return an answer to the user’s prompt. Users still need to evaluate this information and may have follow-up prompts to refine the information, but the evaluation work is greatly simplified.

Google’s SGE, Bing’s AI, and Perplexity offer a hybrid between conversing with generative AI and using traditional search. A list of sources is available if users wish to open those websites and do their own evaluation. Or the user can simply rely on the generated response and trust it is sufficiently accurate. An open question is when users will want to do their own evaluation. Are there some circumstances when users would prefer to open the information sources, check citations, and verify information?

A study from Nielsen Norman Group found that only 22.43% of conversations with a conversational AI tool were followed up by a verification. This suggests very few users will be interested in doing additional work to evaluate the list of sources. Most were confident with the AI tool’s response. This study did find that users were willing to do the additional work of verifying information when the stakes were higher. As we think about the future of SEO for our websites, what content on our websites is important enough that people will want to verify the sources—when are the stakes higher for our customers?

Step 3: Aggregate Information

After evaluating a list of potential information sources, the final step is aggregating the collected information into a single answer. With search engines, that aggregation is done primarily by the end user. The user must open one or more websites presented in the search results and synthesize the information presented. That can take a lot of time and effort, plus there are only so many websites a user can (and will) open.

In contrast, generative AI does this aggregation work for users. The response to the prompt is an aggregation of all the information the AI system has been trained on. With the help of generative AI, the user can skip the process of having to read multiple sources of information and instead can get a fully formed, cohesive answer in response to a single prompt. A few refinements might be needed to clarify information, but this is considerably easier than reading more lengthy articles.

This is why Google and Bing are already incorporating conversational AI responses in search results. Users can see the aggregated answer immediately in the search result, creating a better experience for the user. These so-called answer engines would replace existing search engines. To survive, Google or Bing need to successfully transition from a search engine that supports finding information sources into an answer engine that provides an aggregated answer. Google or Bing may not transition successfully and may be replaced by a new service offered by Perplexity, ChatGPT, Anthropic, or some other company.

Whatever company provides the service, the shift toward answer engines will have an impact on websites. This can already be seen in the rise of zero-click search results. Over the last several years, Google has introduced a variety of search features, like knowledge panels, carousels, and featured snippets. These features are not the same as conversational AI but do use other forms of AI to extract relevant text from a webpage and show that text directly in the search result. As those features have expanded, many people have found little need to review and click on the provided search results. A 2022 study from Semrush found that 57% of mobile searchers and 53% of desktop searchers do not click on an organic or paid result. Many of these featured snippets allow users to aggregate the information more quickly because they minimize the work required to complete the search.

Generative AI’s response is far more sophisticated than a featured snippet. If existing search engines or new answer engines begin returning aggregated responses along with or instead of traditional search results, there will be even more searches resulting in zero clicks. Take the example search in the following image. The conversational response from the generative AI system has already found, evaluated, and aggregated the information the user was seeking. Why would a user click on any of the websites listed in the search results?

Comparing traditional search to generative AI search. Traditional search helps users find information while search powered by generative AI helps users aggregate information.

Understanding aggregation also explains why generative AI presents a greater threat to websites beyond just a loss of traffic from Google search results. Generative AI tools do more than aggregate information—it does so in a way that challenges the general premise of websites. Consider that Wikipedia, as just one example, also offers aggregated information. A Wikipedia page presents a cohesive set of information about a particular subject. Users do not have to visit multiple sources to piece together information that will answer their query because Wikipedia has already pieced the information together. The same is true of many informational and reference websites; these websites aggregate information to simplify the information seeking process for users.

Generative AI tools simplify the aggregation process more than a website can. Where Wikipedia, and other informational websites, provide an article about an entire subject, ChatGPT provides a response about only the specific aspect of that subject the user was interested in. More than that, the generative AI tools can personalize the response to the user’s specific needs related to the aspect of the subject of interest.

In other words, generative AI provides targeted aggregation whereas websites provide generalized aggregation. The following example shows how the Medical AI custom GPT was able to aggregate information about a medical question specifically for this user’s circumstances. Users may find this targeted and personalized response preferable compared to a general article on a trusted medical website about this same subject.

Response from MediSearch’s custom GPT providing a targeted answer to the user’s query

Search Helps Users Find, Generative AI Helps Users Aggregate

The threat to websites and SEO is that generative AI offers a better user experience because it simplifies information seeking. Generative AI tools are unique because of their ability to aggregate information specifically aligned with a user’s interests. Aggregated information is the goal of information seeking and generative AI helps people complete that goal much faster than search engines and websites can.

There are many use cases where users would prefer to skip finding and evaluating potential information sources. What person willingly wants to take on more work? Instead, users want an answer constructed from multiple sources of information that have already been found and evaluated. Generative AI aggregates information across millions of sources—way more than any person could ever consider. Plus, that aggregated answer is uniquely generated for each user, unlike an article that is written for thousands of users.

The open question is whether users will always want an aggregated response. In some cases, users may prefer to find and evaluate information themselves. This may be true for high-stakes situations. In those cases, generative AI systems may not deliver a helpful user experience. How many situations are high stakes related to your company?

It is important to understand how users prefer to seek out the information your website currently presents. The more you can focus your website on the information users will want to find and evaluate on their own, the more you can make sure people still have a reason to visit your website. If you only focus on the information users want aggregated as quickly as possible, then you end up competing directly with generative AI—and it seems unlikely websites will be able to compete.

Key Takeaways

  • There are many information seeking frameworks. The information seeking process can be simplified into three steps: finding, evaluating, and aggregating information.
  • Generative AI differs from search because is more effectively aggregates information. When AI tools aggregate information for users, users do not need to spend time (or as much time) finding and evaluating information sources.
  • People would likely prefer an aggregated answer without spending time finding or evaluating multiple websites. That is, people will likely prefer to use generative AI instead of search engines. Sometimes, though, users may not prefer an aggregated answer. Understanding the difference is important to determining how to adjust your SEO strategy.

Need Help?

If you would like help understanding the information seeking process related to your website, please contact me. We can work together to review your website and formulate an updated SEO strategy that will support your website as we transition away from traditional search toward generative AI.

You may also like

What Are Generative AI Hallucinations?

One of the biggest problems with generative AI responses is hallucinations. What are hallucinations? Why do they happen? More importantly, as generative AI starts to dominate search results, how do we factor hallucinations into our future SEO strategies?

Will People Click Links in SGE Results?

Will people click links in SGE results? How many clicks could we lose with SGE rolling out? We conducted an in-depth study to find out.

Understanding Prompt Types

How do we optimize for prompts and make sure our companies show up in generative AI search responses? How is this different than optimizing for search queries? It begins by understanding what types of prompts people use.