ArticleWho’s Afraid of Little Old SGE?
If you haven’t heard – and most of you have – Google has officially unlocked the newly-rebranded “AI Overviews” (RIP, Search Generative Experience – it feels like we barely knew you).
Since the announcement of the “AI Overview” rollout (first at Google I/O, then at Google Marketing Live this past week), there has been Very Serious Concern™ from large swathes of the SEO community – along with more than few screenshots of AI Overviews Gone Wild.
Many people – some quite reputable – have declared this the beginning of the end for Google. Candidly, they’re wrong. This is merely the end of the beginning.
The search experience – not the aesthetic, or the functionality, or the ranking algorithm – has been relatively consistent for 20+ years. There have been step changes, to be sure: featured snippets, knowledge panels, FAQs, endless scroll, the list goes on – but the core of the SERP was largely the same: some ads, 10 blue links, some other stuff, more ads, more blue links. The reason for this was largely pragmatic: it kept the point of leverage consistent. Google served up some links, and a few billion humans served as (effective) mechanical Turks, helping the algorithm refine what was actually useful + relevant vs. what was trash.
And while the scale + functionalities of search changed, the point of leverage did not: there was still click-based interaction to use to evaluate quality – even if some (or more, or most) of those clicks were on the SERP features themselves, vs. on those 10 blue links.
But now, for the first time in 20+ years, we’re actually at an inflection point in search.
That has many people profoundly upset – because a sea change entails a fundamental reshuffling of the existing order. There will be many sites that are no longer economically viable as a result of this, and for the people who run those sites, that’s unfortunate. But for the internet as a whole, that’s a wonderful thing, because destruction of incumbents is what enables the growth of innovators. Both things can be (and are) true.
Destruction is necessary for creation. And in that vein, AI Overviews are the City Destroyers from Independence Day. And just like those City Destroyers, AI Overviews are coming for the big targets first, but eventually, they’re going to make their way to many other sites, too.
Unlike previous SERP Feature changes, where detecting if the feature was present on a specific SERP was simple + straightforward (use a tool to scape SERPs), AI Overviews aren’t included on Incognito SERPs.
That means most existing SEO tools will not be able to accurately determine if and when those SERPs have been impacted. This leaves marketers (not just SEOs) with two options:
- Find alternative ways to infer the presence of AI Overviews
- Prepare as if every SERP will be impacted
I am of the view that this is not an either/or situation; it is a both/and situation.
Fortunately, there are several ways to (relatively) accurately infer the presence of AI Overviews:
- Manually check core SERPs – this is the answer no-one wants to hear, but it does work at a limited scale. If you’re running any kind of paid search, you should already be checking your SERPs regularly, so this should not be anything more than an additional column on your spreadsheet. If you’re not manually checking your SERPs, I’d suggest you stop reading this, and start doing that.
- Correlate with Existing SERP Features: In our testing thus far, the SERPs most likely to have an AI Overview are ones that previously had a featured snippet. This makes sense, as those queries tend to be the ones that Google believes can be answered on-SERP, and AI Overviews are (gasp) answers on a SERP.
Note: we’ve observed many SERPs with a Featured Snippet / AI Overview double-dip; in many of those cases, the site referenced in the Featured Snippet was ALSO linked in the AI Overview. This may or may not persist, but it’s worth mentioning that it seems – for the time being – that the highest-probability method to being linked in the AI Overview on a double FS + AIO SERP is to get the Featured Snippet. - Measure Pixel Depth on SERPs: Multiple SEO tools (including Brightedge & SEOClarity) provide a metric called result pixel depth (or some variant thereof). Since AI Overviews are larger than Featured Snippets, if you observe an increase in pixel depth, that’s a good indicator AI Overviews are now serving on that particular SERP.
- Benchmark CTR + SERP Position: Search Console provides both Avg. Position + CTR; we know that AI Overviews (overwhelmingly) do two things: reduce clicks off the SERP + don’t change position. So, if your CTR is falling while your Avg. Position is constant, that’s a strong indicator that AI Overviews are present.
- Monitor The Number of Words in Your Query Strings: an interesting trend for people using GenAI tools is a tendency to write longer, more specific queries – whether those are more detailed prompts, or more specific searches. In Google Search Console and/or Google Ads Search Terms Report, you can download all (visible) queries going back years – and track the average number of words per query (if you want to make it easier…use Gemini or ChatGPT). If you then group those queries by topical and/or intent cluster, you can see which topics/types of searches are getting longer, faster – and (by extension) what kinds of topics + queries are more likely to be impacted by AI Overviews.
The above steps solve the first part of the AI Overviews question: identifying where AI Overviews are impacting your marketing efforts today. But none of this solves the big, thorny question: what does it take to win in the next chapter of search?
There are (at least) three things:
1. Get Back To Being Brilliant At The Basics:
The thing I love about automation – whether it’s ad account or SERP or process – is that it creates an imperative for people to think bigger, broader & more strategically.
At some point over the last 20+ years, marketers collectively lost sight of what was intrinsically important (the user) in favor of what was instrumentally important (the search engine). The results of that are littered about the internet today: recipes for Carbonara include a history of the Roman republic. Articles on removing stains from clothing include dissertations on the chemical composition of various garment afflictions. Even B2B landing pages are chocked-full of keywords (regardless of whether that’s a good life choice) to the point of being ridiculous. We’ve all observed examples of this nonsense. SEOs have peddled it for decades.
SEO has violated Goodhart’s Law: the measure (rankings) has become the target.
It’s time to change that. Digital marketers need to remember that “Digital” is modifying “Marketers” – we’re marketers first, and it just so happens that we do our marketing on the internet. But marketing, fundamentally, is about delivering the right message to the right person at the right time. That’s it. It hasn’t changed.
Getting back to the basics is all about doing that:
1. Actually understand your audience:
For 90%+ of marketers (agency + in-house), audience research is – at best – an afterthought. What’s presented as “Audience Insights” is, in reality, a half-baked cake thrown into the oven when the strategy party is already underway. It’s more likely to give you salmonella than it is to be delicious.
As this relates to AI Overviews, I start with a radical premise that (most) everyone in SEO probably finds horrifying: Google is doing this because people are using this.
Maybe not marketers. Maybe not the most educated 5% to 10% of society. But that still leaves 90% to 95% of the population – ordinary people, in their ordinary lives. Those people don’t think about SERPs, or what kind of link they’re clicking on (paid vs. organic) or most of the content that was jammed onto a page in order to improve the probability of it ranking highly. They don’t care. They just want answers with as little effort as possible.
Once you accept this, then the value of audience research becomes crystal clear: if people are using this, it’s because they want answers. We can only provide those answers if we know the questions, and we can only know the questions if we know the people asking them.
Audience research is all about understanding what drives, informs and motivates various segments of your audience. Invest the time necessary to understand your audience(s) at a deep, human level. Do surveys, interviews and/or focus groups. Subscribe to platforms like SparkToro (seriously, the best $39 we spend each month). Go to the YouTube channels they watch, and consume the content. Follow the influencers they follow. Read the publications they read.
What emerges when you do this is an understanding of both the motivations, challenges, fears and desires of the audience, along with an ability to communicate with the audience in their natural content language. That is the critical piece – because that’s what allows you to then go forth and create remarkable content that connects with that audience in the way that they are most likely to search AND accept.
Without a clear understanding of your audience, nothing else matters. You’re going to lose – it’s just a question of how long before it happens.
2. Craft Unique, High-Value, Authoritative Content on Specific Topics Relevant To Your Audience:
If your content strategy consists of repackaging stuff from other sites/people in slightly different formats, you’re likely to have a bad time in an AI Search world.
At a fundamental level, we know how LLM powered search works:
Basically, what this tells us is that the search engine filters sites based on the initial query, then the LLM runs only on the filtered site(s) to create an AI Overview. So, if your content isn’t getting past the initial filters, it’s NEVER going to show up in the AI Overview.
Craft content that has a unique PoV and adds value your audience can’t find anywhere else. Yes, Google (or Microsoft) may rip some of it off. Imitation is the highest form of flattery. If an AI Overview is straight-up ripping your content, that tells you you’re on the right track.
If you’re wondering how to get started, I’d suggest the following:
- Design For Answers – clearly indicate questions + answers in some or all of your content. This is similar to what many sites currently do for (among other things) FAQ snippets, so it should be familiar.
- Add Citations – add relevant citations to credible sources in articles
- Add Quotes – add direct quotes from individuals with relevant domain expertise; link to those individuals.
- Include Statistics – Quantitative data is always a good inclusion – especially if you have unique statistics or research.
- Unique Content + Perspectives – if you say the same things as every other page included in the LLM inputs, you’re unlikely to stand out or dislodge an incumbent; instead, find the areas where the current high-ranking pages are falling short, then create content that addresses those gaps.
- Demonstrate Relevant Topical Expertise – at the end of the day, Google maintains one of the most (if not the most) robust entity graphs in the world. The greater the degree to which your content mentions relevant, related entities, the more likely it is that you’ll rank highly, and the more likely it is that you’ll win.
- Provides Outsized Value – this is the most difficult one to quantify, but my rule of thumb is simple: if the quality of content you can produce on a given topic does not dramatically exceed the quality of content that Gemini, ChatGPT or a similar system can produce, then you’re not ready to publish content on that topic. The content that wins now – and the content that will win tomorrow – will be remarkable.
2. Think Bigger & Broader Than Search:
All marketing is a distribution game. For many marketers, for the past quite many years, Google has played a disproportionate role in that distribution. That’s likely to look different in an AI-Overview driven search environment. AI Overviews are going to increase zero-click searches for a significant tranche of informational queries – and likely quite a few transactional ones, too.
For brands that are relying on Google (or other search engines) to distribute their content, this represents a tectonic shift in their world order. There’s a reason they’re unhappy: because this evolution in search has the potential to materially change their economic fortunes.
Today, most marketing teams place 90% of their effort into content creation, and 10% of the effort into content distribution. Many of those brands have seen their traffic (and businesses) crater in the wake of AI Overviews + the most recent round of Google Core Updates.
The solution is simple: leverage that audience research to find the channels where your audience is already consuming information, format your content in a way that is likely to resonate, and find ways to distribute it in those locations. That might be influencer partnerships. It may be thought leadership positioning & digital PR. It might be native, programmatic or negotiated placements. It might be something else entirely, or all of the above plus some.
Less than 5% of time online is spent on search. It’s a critical 5% to be sure, but it still leaves 95% of time not on Google or Microsoft or Perplexity. If you can’t win the SERP, find ways to win before the search ever happens.
3. Defend The Clicks You Need:
That 5% is quite critical – and there will be certain searches that your (or your client’s) business absolutely needs to win.
The good news is that Google is in the advertising business – and they have already announced their intention to roll out ads within AI Overviews. Anyone who follows Google (or search more generally) should have seen this coming a few years ago, with the roll-out of RSAs, the blending of paid/non-paid results, auto-generated assets, PMAX, the addition of new “asset” types and the changes to match types. The end-result of all of those changes is a search product that is tailor-made to serve ads inside of AI Overviews.
Google is nothing if not ruthlessly capitalistic. They will give you the opportunity to get the clicks you need – but you’re going to have to pay for them. Candidly, that makes sense from Google’s perspective: if an ad can be made to be every bit as valuable + relevant to a user as an organic result, then their ideal outcomes are: ad click or zero-click. Pretty simple.
This means that more brands will (likely) choose to advertise via search (either search, PMAX, or whatever new campaign type they introduce for AI Overview Ads). In turn, that is likely to result in more competition for those particularly valuable queries – and an even-larger chunk of marketing budgets spent on not-so-valuable queries.
This is no different than poker – of the 169 possible Texas Hold’em starting hands, only 40 (~23.67%) are net profitable. And of those 40, 50%+ of the total profits come from just 5 hands (AA, KK, QQ, JJ, AK, if you’re curious). In search, every brand is likely to have a different set of winning hands – the game is (increasingly) to know what those hands are, and refuse to play the others.
This is what defending the clicks you need is all about: identifying the clicks you ABSOLUTELY need to have, bidding aggressively on them, and excluding everything else. Once again, this goes back to audience research, topical expertise and keyword management – what you exclude is more important than what you include.
4. Take Truly Valuable Content Off The Search Grid:
The final thing I expect to happen – at least in the short-term – is for more brands to hide their most valuable content from search engines – either by publishing it in a newsletter or downloadable resource, sending it directly to potential customers/clients, or sharing it via a private group/private community (though that might not be viable for long).
Candidly, I don’t know if that’s a stable long-term solution: whenever excess value is concentrated inside a small group, there’s an overwhelming tendency for some members of that group to take + leverage it for their own gain. I’m not sure how this plays out at scale, but I do think brands need readily-accessible discovery options in order to build their customer/client pipelines, and relying on indirect distribution to fill those pipelines is a bad life choice.
Finally: What If AI Overviews Are A Fad?
The single-most-common objection I’ve heard to an AI Overview-centric search experience is that AI Overviews are flawed and inferior to standard search. They frequently (relatively speaking) hallucinate or memorize. They don’t provide a ton of value relative to what already exists via a Featured Snippet. They stumble with basic logic.
These are all true. But here’s the thing: we’re comparing the pinnacle of search to the infancy of AI Overviews.
LLMs – along with our ability to control them – will only improve. Multi-Agent / Multi-Model approaches will catch errors/lies/hallucinations at higher rates, as one model generates a response, and a second/third distinct model validates it (just as people create drafts, and another person reviews/edits the draft). We’ll develop better integrations between deterministic systems + probabilistic systems, such that queries requiring a deterministic answer pull that, and queries requiring a probabilistic answer will have one of those generated. This is the technological reality of this space: things will get better. This gets talked about a fair amount (usually with a heaping helping of AI Doomerism).
But – I think – the more interesting part of the argument is that people will also get better at using AI tools. When the internet first became a thing (and I’m old enough to remember that, sad as it is to write), there was a considerable amount of effort + time that had to be spent learning what to ask what – Ask Jeeves was quite good at questions, and Netscape was largely OK for specific queries, and AOL existed if you had the perverse desire to turn your monitor into a virtual times square. Over time, people (collectively) learned what queries were best suited to which sites – and then Google won search, and had to create a whole bunch of SERP features in order to accommodate a whole bunch of queries that 10 blue links really weren’t appropriate for – but since there was no-where else for them to go, they went to Google.
My prediction is that we’ll see something similar emerge with AI Overviews + AI tools in the next few years: people will evolve. We’ll learn how to better phrase queries, we’ll develop an understanding of what kinds of responses/what kinds of queries each LLM is best suited to address, and, eventually, a few models will win out. One of those will be Google’s (because no-one else has the world’s information in the same format and level of detail), and the others…we’ll see. But, just as the machines will get better, people will get better, too.
That’s all for this time – thanks for reading!
Cheers,
Sam
