Decorative feature graphic to accompany the post

Google’s generative AI shift: understanding the impact on search and your business

Google search is going to significantly change in the next 12-24 months, and we urgently need to understand how and why.

In the next 12-24 months, I predict that Google will start generating AI-powered featured snippets for most search queries, fundamentally changing how we find and consume information online.

These AI snippets, generated from the top ranking search results, will attempt to directly answer user queries with a concise 2-3 sentence response. For users, it will be a faster way to get information. But for businesses relying on search traffic, it could be a tectonic shift in SEO strategy.

In this post, I’ll lay out why I believe this change is coming, what it might look like, and most importantly, what it means for the future of SEO. If you’re a business that depends on organic search traffic, you need to be preparing for this AI-powered future today. Let’s dive in.

AI featured snippets are coming

My working assumption is that Google is going to roll out generative AI results for most search queries in the next 12-24 months.

These will be 2-3 sentence answers against the search query, with the response generated based on the top-ranking results.

Imagine this as similar to featured snippets, but instead of text extracted from 1 source, the snippet is generated based on multiple sources (and the text is personalized based on the specific searcher query).

This solution makes sense as a next step:

  • It’s a pretty safe way of implementing generative AI, as you’re answering based on the top-ranking results. This lets you root your answer in a source of truth you can cite (and blame if the answer isn’t what you want).
  • The generated answer will be useful, but not too useful. This will work for basic questions but anything more nuanced and the searcher will still need to engage with the search results.
  • This does not disrupt Google’s advertising business. We already see featured snippets exist happily alongside ads and AI featured snippets would be fine with ads too.
  • You can generate these answers cheaply. Google mooted charging for search results recently. I don’t see such a dramatic change happening; surely the solution instead is to work out how to use AI cheaply. Because the generated answers are based on strong sources of truth, a relatively basic LLM would be able to generate something usable. In 12-24 months, the cost will have at least halved, and that’s what makes this plausible.

In many ways, the emergence of AI featured snippets is the natural next step in Google’s long evolution from a simple list of links to a sophisticated information delivery system. As Google continues to leverage AI to provide direct answers, the very nature of search is changing before our eyes.

Will this ultimately lead to a better search experience for users, or will it concentrate even more power in Google’s hands? And what does it mean for the businesses caught in the middle?

Any AI rollout in search needs to be cheap and basic – and that limits its usefulness

Since 2020, Google Search has been led by Prabhakar Raghavan, a former Google Ads executive known more for his focus on revenue growth than passion for organizing information. I see this as critical context to understand how Google is going to roll out generative AI in search.

In 2019, before Raghavan took over Search, he was involved in a “code yellow” incident at Google. This refers to a situation where the company’s finance team, concerned about a slowdown in search revenue growth, put pressure on the search team to make changes to juice short-term revenues. Raghavan, then leading Google Ads, was a driving force behind this push, even as search veterans raised concerns about the long-term impact on user experience.

Following the “code yellow”, Raghavan took over Search and many search veterans left. You should see this as a victory for those in favor of short-term revenue growth and a loss for those concerned about long-term product quality. We must see the rollout of generative AI in search through this lens.

Generating text costs money. I find it implausible that Raghavan is going to agree to anything that seriously impacts Search’s profitability, and that means any rollout is going to be basic. Just last week, Raghavan cited the need to control costs in an all hands meeting last week: “[we’re] spending a ton more on machines” because of genAI.

You keep costs down by using smaller, less capable LLMs. At current pace, quality is doubling whilst cost is halving for LLMs YoY. This is where I get my 12-24 month timeframe from: it’s plausible that in this period, the cost of generating simple text will be sufficiently low that something like AI featured snippets is doable.

But, this is going to be basic. The model won’t be capable of more than simple summarization from a couple of sources. This severely limits the usefulness of these answers.

In this context, AI featured snippets become a lot less to worry about because they’re just not very useful. The most basic questions will be answered, but any searcher looking for any level of depth on a topic will not be satisfied. This is pretty trivial to win against – as long as you’re paying attention.

Competitors are already doing this

You see paid-for search engine Kagi offering this already, with its Quick Answer functionality.

The Quick Answer does exactly what I’ve described above: reads the top results and returns a featured snippet-like experience with specific citations.

Here’s what this looks like if I search for reviews of Cal Newport’s new book, Slow Productivity:

The book “Slow Productivity” by Cal Newport offers a new approach to productivity that focuses on producing high-quality work while working less. The book argues against the “performative busyness” that has become common in knowledge work, where seeming busy has become a proxy for productivity. [1][2]

The citations are the New York Times’ review and Financial Times’ review. The Quick Answer keeps it simple and rooted in its sources of truth (which are the top-ranking results). Useful, but not too useful.

Interestingly, Kagi does not automatically generate a Quick Answer. You have to request it manually. I assume this is for cost: letting the user self-select where they want an AI answer means you’re not generating unnecessarily.

With enough searchers, you might be able to predict which searches bring results where people want an AI answer. One could certainly imagine Google, a company which loves Machine Learning, embracing this.

Winning in the next AI era

All of this presents an opportunity. You just need to know how to take advantage of it. If you’re not already thinking about how to position content today to win tomorrow, now is the time.

  • If the SERP has a featured snippet today, it will have an AI featured snippet tomorrow. You may want to avoid creating content against these terms, as your content will be vulnerable to genAI in the long term.
  • Zero-click searches need caution. This is where the AI rollout can be seen as much less radical: there are already large swathes of searches which do not result in any clicks. You can solve for this by making your understand of search volume wildly more sophisticated. High-quality data sources are critical here. How many people are searching for a query with an intention to click? What share of those clicks is going to ads vs organic? If you’re uncritically looking at a keyword’s search volume and accepting more volume is better, it’s time to urgently add more nuance.
  • SEO will be MORE winner-takes-all. The AI featured snippet will be generated from the very top ranking results. If your content is included in those, great! If not, your visibility decreases. You’re going to see more gains aggregate to the very highest-performing results.

Much of which is to say: keyword and topic selection is more critical than ever. Your success with SEO depends to a large extent on keyword selection currently. That is going to be more and more the case going forwards. Not all rankings are going to be equal going forwards.

Commercial value and the number of organic clicks against each keyword (rather than just volume) are a more important part of the picture today and this will be even more the case going forwards.

At Ellipsis, we’ve long used our FALCON AI to predict ranking before we create content. We’re rapidly evolving this to add more nuance: is that rank going to result in clicks? If there was an AI featured snippet, what might that be? Is it any good?

These are the questions one needs to answer, and you can! A model like pplx-online offers exactly this, or a response out of GPT-3-turbo would work too. Per the above, I suspect in 12-24 months, the quality of GPT-3-turbo will be possible with a much cheaper model and this makes it a plausible candidate for wider rollout in search.

SEO strategy becomes more important than ever

The rise of AI-powered search results is not a distant sci-fi future – it’s an imminent reality that every business needs to prepare for today. As Google rolls out AI-generated featured snippets in the next 12-24 months, the rules of SEO will be rewritten before our eyes.

Some will see this shift as a threat, a move by Google to hoard even more power and traffic for itself. But I see it differently. I see it as an opportunity for the smartest, most adaptable businesses to thrive in a new era of search.

The key is to start adjusting your strategy now. Double down on creating the kind of in-depth, authoritative content that Google’s AI can’t easily replicate. Rethink your keyword targets to prioritize queries where organic clicks will still matter.

Most of all, recognize that in the AI age, SEO will be more-than-ever a winner-takes-all game. The top spots on page one will be more valuable than ever, while anything below will be more invisible. The businesses that win will be those that see this shift coming and adapt their approach to stay ahead of the curve.

At Ellipsis, we’re seeing and preparing for this more than any company in the industry I’ve seen. Our FALCON AI was in the New York Times in March 2022, 8 months before the release of ChatGPT. We leverage technology where it’s best, but complement this with human expertise. Since the release of ChatGPT, we’ve hired a full-time team of excellent in-house writers to double-down on content quality.

The future of search is coming faster than most realize. Will your business be ready, or will it still be searching for answers?

Picture of Alex Denning

Alex Denning

Alex Denning is the Founder and Managing Director of Ellipsis®, a world-class SEO Content agency. Alex is the inventor of FALCON AI®, the next-generation AI SEO platform that helps you predict how your content will perform – before you create it. Alex is an international speaker and publishes regularly on the future of SEO. @AlexDenning on Twitter