Google’s AI Overviews & Perspectives

Editor’s Note: The title of this blog post has been updated. AI Overviews was formerly known as Search Generative Experience (SGE).

What is it?

Google’s AI Overviews are live in most countries. The only way to opt out of having your content included in the AI answer is to block Googlebot (Google’s crawler) from accessing your website.

The design and sources shown in the Overviews change almost hourly. Most of the sites being pulled into the answers are ones shown in the top 10 results.

There is no official data for impressions or CTR in Google Search Console as the data is there, but there is not a filter for “AI Overviews”. When Bing rolled out their AI chat responses into Bing Webmaster Tools, they also included them with web visits, making it impossible to see chat-only engagement.

What will be the impact?

We don’t know how many people use the AI Overviews or if they click through.

For some queries, the AI Overviews response auto-loads; for others, you need to click “generate.”

This example auto-generated:

How are responses generated?

We have some insight based on Google’s patent behind AI Overviews. From that patent and Google’s announcement, we know the following:

• Google uses multiple Large Language Models (LLMs) to generate the response.
• The system received a query (either entered or automatically generated)
• It selects a set of search results documents (URLs) relevant to that query or related queries.
• It generates a content snippet from each search result document (this can be an image, video, or text)
• It uses an LLM to process the search results, document content snippets, and generate an LLM output.
• That is then rendered to the client’s device along with optional search results, confidence annotations, links to verifying search results documents, etc. 
• The engineers on the AI Overviews update previously worked on Featured Snippets at Google. 

SEOs have spotted that for local (location-based) queries, AI Overviews uses content from Google reviews in its answers, which means that a robust review management and acquisition process needs to be in place for local businesses.

What do you do to appear in those generative answers?

• For Google, the sites mentioned in the top two results are most often pulled through and referenced in the generative answer, along with having the most relevant passage of content.
• How do you get the most relevant passage of content? Understanding the query’s search intent and writing something succinct to that topic that also covers the relevant subtopics and correctly using HTML markup (H1, H2, tables, bullets, etc.) on your post/page, along with structured data markup, if relevant.
• Rand Fishkin looked at the data sources Bard uses for its responses (in the chart below) and highlights that if you’re a brand people are talking about related to their issue, you will do better in generative responses and the currently live perspectives feature.

• For Bing AI search, he recommends making it easy for Bing to index your site (IndexNow) and having your content based on a basic HTML template with structured data.

Other search changes during 2023 Q4 that will impact rankings:

Helpful Content Update

Google rolled out a helpful content update that focuses on serving answers from content that demonstrates EEAT (Expertise, Experience, Authority, and Trust) and is helpful to the end searcher. We know that Google uses “what is the purpose of this page” in its human quality rate guidelines and how well the content meets the user’s needs. To be “Highly Meets,” the content must be very helpful for many or most users. Some users may wish to see additional results. To be “Fully Meets,” Google indicates that this is a “special rating category,” which only applies to certain queries and results. All or almost all users would be immediately and fully satisfied by the results and would not need to view other results to satisfy their needs. 

How does Google/Bing measure this type of satisfaction? We know they use user click behavior (thank you, DOJ trial!). Users clicking on results to websites and staying (a positive signal) vs users quickly returning to the search results or revising their queries.

What do you do to ensure that your content is helpful?

For most, this is a chance in the writing process to ensure that one’s content is focused on answering the user’s burning questions and follow-up-related questions (to keep them on-site longer). 

Developing content based on search intent research and a strategic content brief are recommended. 
The other part that a website owner can work on is encouraging those positive user signals by influencing the search snippet click (optimizing the meta title, URL, and meta description). Additionally, we recommend that the target keyword is in the header and first paragraph to create strong signals related to information scent and that each page has a natural next step for the user (internal link) to keep the user on your site continuing their learning journey (vs. quickly returning to search).

Perspectives Feature

Google now shows a Perspectives feature highlighting first-person expert experiences for some queries, and most often on mobile devices. These are often from social media platforms. Here are some examples of what that looks like:

How do you ensure your content is found in these perspectives?

This is where SEO teams will need to start to branch out toward strategic social media efforts. The first step is to develop relationships with established influencers already seen as topical experts on platforms that appear for your target queries.

The second step is to work to highlight the experience and expertise of your staff and writers. This year, Google also updated its Knowledge Graph, focusing on authorship. You can build Google’s understanding of the expertise of your authors through profile and bio pages with structured data markup and reference to those experts from databases used in the knowledge graph. The largest ones for individuals are below; the rest can be found here.

Our Recommendations

Experts should be creating videos via YouTube, starting with appearances on well-established YouTube channels. We recommend organizational markup on your About page, references, and mentions in references used in the Knowledge Graph (which no longer includes Wikipedia). Ultimately, you should ensure that your organizations and experts trigger Knowledge panels.

Organizations should also monitor for which social media platforms appear for their target terms and start developing social media engagement strategies on those platforms.

Drop us a note if you need assistance in integrating social media into your search strategies.