Google’ SGE (AI Snapshots) Patent and the Future of Search (?)

A highly anticipated #SGE patent from Google is somewhat a disappointment..

Last week Google published its first “Generative Summaries for Search Results” patent explaining how Google’s AI snapshots might work.

It describes methods for generating query-based summaries of search results using LLMs to process the content and generate natural language summaries.

Some important highlights from that patent include:

• Snapshots don’t just respond to the query itself: They provide “additional” information that is likely helpful to the searcher

  • Note: The algorithm uses “related queries” that are likely sourced from Google’s “also searched for” algorithm. “Related queries” are those that are searched in “close temporal proximity”.

• AI snapshots use Google’s index of documents that respond to the searched query or to those closely related queries in the best way (these pages are referred to as “search result documents (SRDs)”

  • Note 1: Source documents are summarized and optionally referenced

  • Note 2: ALL documents are taken from Google’s index but some documents may be ranking for related queries, not the one that is being searched.

Note that the patent doesn’t explore important aspects of search Google is struggling with, such as:

• Bias in the selected documents

• Identifying official or trusted sources

• Content helpfulness, etc.

This is why URLs referenced in AI boxes are not necessarily taken from the top ranking positions. SGE algorithm is lacking what Google has been trying to work on for years. The selection process appears focused solely on using relevancy to queries as the criteria. I personally find it absolutely hilarious, given Google’s history with all these signals.

Furthermore, there’s no explicit mention of advanced personalization of the snapshots. Yes, past searches are taken into account when creating AI answers but you’d think in 2023 these answers should also be adjusted based on the perceived experience of the searcher with the topic. 

Somehow I expected more from Google…

All in all, the patent is merely about selecting documents to summarize and optionally reference (sorry, I cannot emotionally let go this “optionally” part)

But this is what we are dealing with at this point…

What to do:

1. Prioritize your keywords based on how important they are to your bottom line.

Then take a close look at the related queries Google serves for it and evaluate how good of a job your site is doing responding to all this stack of keywords.

The patent discusses the process of selecting documents based on their responsiveness to a given query, related queries, and recent searches. These selections can include documents that responded to the original query, as well as those responding to related or recent queries.

2. Monitor AI snapshots for your most important queries

Things will change many times in the future, but keeping an eye on your queries is certainly going to help adjust your strategy.

3. For ecommerce brands, make sure your products are fed into Google’s shopping index. Google uses that index for product-related searches for recommendations.

Finally, this patent may have some really huge holes in it but there’s one important thing to note here: It makes organic rankings even less important and prioritizes understanding your target audience and their buying journeys.

Yes, this conclusion is neither helpful nor actionable but it is still a learning curve!

Must-reads!

• Here’s my article on using AI tools to understand patents (I tested them to analyze this patent)

• Catch the replay of our webinar discussing the future of search and AI. We had a blast!

Please let me know your thoughts!