In late April 2018, Google Research published a paper titled Ask the Right Questions: Active Question Reformulation with Reinforcement Learning that shares a vision of a new way to rank pages. This research paper is not a notification of a new algorithm being implemented or an announcement that it will be used. But, this new ranking system is actively being explored and tested by Google.

If it ultimately is implemented, traditional methods of page ranking will be significantly changed, and the impact will be far-reaching. As far as who will benefit, companies that genuinely understand what specific information their users want and need and then clearly provide it on their websites.

High-Level Overview

Google’s 15-page paper, released at the Sixth International Conference on Learning Representations, focuses on two processes:

• Query reformulation
• Reinforcement learning

The query reformulation part of the process has been dubbed “active question answering” (AQA). Under the current ranking system, when someone types in a query, Google’s ranking algorithm kicks in and returns results in response. With AQA in the mix, though, before ranked results are presented in response to a user query, a machine learning algorithm (also called the “agent”) will analyze the query and formulate follow-up questions based on the query.

Interestingly, this agent would have no knowledge of how pages would be ranked. Instead, as SearchEngineJournal.com summarizes this concept, “This new algorithm uses a learning system that reformulates the user query, asking the ranking engine many questions, then choosing the best answers from the multiple sets of answers.”

Here’s how the explanation reads in the research paper. “We propose an agent that sits between the user and a black box QA system and learns to reformulate questions to elicit the best possible answers. The agent probes the system with, potentially many, natural language reformulations of an initial question and aggregates the returned evidence to yield the best answer.”

The research paper notes how this actually mimics how human beings find answers to their questions, especially more complex ones. They start with an initial question, perhaps: “What is the best software for an optometrist’s office?” and then ask more in-depth questions, perhaps about features, cost, usability and so forth. In 2018, this involves the person searching on Google multiple times and then pulling together the answers in a coherent form.

That’s the overview of query reformulation. Reinforcement learning, meanwhile, is a type of machine learning that focuses on which actions software agents should take to maximize cumulative rewards. So, in short, Google is investigating how seamlessly and effectively this dual system (AQA and reinforcement learning) would take users from their initial question to the end of their researching process.

It isn’t possible, of course, to predict exactly how this system would work or how it would impact certain sites and pages. Even Google can’t yet know how this would unfold in the real world setting. Their plan, according to their paper, involves continued development of AQA and to further investigate the “sequential, iterative aspects” involved in seeking information. Ultimately, Google experts recognize that the loop between the reformulator process (when follow-up questions are asked) and the selection process must be closed.

So far, Google says the process is proving to be “highly effective.”

How This Would Affect SEO Strategies

Currently — and, actually, throughout Google’s existence to date — the ranking algorithm has been responsible for determining how pages rank in the search engine results pages (SERPs) for each keyword. If this new system is adopted, though, the agent will ultimately make ranking decisions, using information provided by the ranking algorithm.

So, what would really matter is how well content on a particular web page answers the question being asked — as well as the follow-up questions devised by the agent (but not seen by the person making the query). Search Engine Land suggests there could already be a similar, less complex process in play, which would explain how pages with fewer inbound links and/or with inbound links of lesser quality rank above pages with a better link profile: they provide better answers. Perhaps the use of Google’s Answer Box is a rudimentary step in this process, too, as it lifts what it sees as the best answers to questions, regardless of how they rank using today’s ranking factors.

Clearly, SEO strategies will need to focus, even more than they are today, on user intent and how to provide answers to the questions people have about products and services. How clear are you about the questions your prospects are asking? What surveys are you taking and what conversations are you having? How highly honed are your social listening skills?

It will also become increasingly important to effortlessly keep visitors on your site until they convert, however your business defines conversion, because subsequent wording of queries may take users to competitors’ sites. Plus, it will also be crucial to monitor site traffic as Google transitions into this new system (when they do) and adjust accordingly. Doing what you’ve always done will not be the answer.