Somewhere in the last two years, the click quietly stopped being the prize.
Most SEO teams haven’t updated their dashboards to reflect that. They’re still reporting on rankings and sessions, watching both decline, and calling it a crisis.
It isn’t a crisis. It’s a correction. The click was never actually the goal; it was just the only thing Google let you measure.
Now that AI overviews are answering the question before anyone reaches your page, the industry is finally being forced to optimise for what it should have been chasing all along: being the source someone trusts, not just the link someone clicked.
Decline Is Real. That’s Not Interesting Though.
The topline numbers are dramatic enough on their own. According to SparkToro’s 2026 research, the share of Google searches that result in a click has fallen to around 23% in two years. When an AI Overview appears, click-through collapses by roughly 60%, according to the same study.
Everyone covering this story stops there, treats the falling number as the headline, and moves to tactics. That’s the wrong place to stop.
The actually interesting number is buried further down: the overlap between traditional top-10 rankings and AI Overview citations has collapsed from roughly 76% in late 2024 to somewhere between 17% and 38% today (Omnibound’s 2026 AI SEO data). Ranking #1 and being cited used to be nearly the same achievement. Now they’re two different games, and most teams are still only playing one of them.
Not Every Zero-Click Is a Loss
Here’s where most of the panic falls apart under scrutiny: it treats every zero-click search SEO as equally bad, and that’s not true.
A German SEO practitioner’s case study offers the cleanest way to think about this, splitting zero-click into “good” and “bad” categories based on a single question: does the zero-click generate a measurable brand return? the original framework stated that a definitional query that gets fully answered with no brand mention is a real loss. A query where your brand is cited by name inside the AI Overview, even without a click, is closer to a win.
Most teams don’t make that distinction. They look at falling sessions and assume it’s all damage. Some of it is. A meaningful chunk of it is just Google finally being honest: you were never the destination; you were the answer, and the answer got extracted with your name still attached.
That distinction matters more in some categories than others. Based on the data, it was established that informational queries now sit at 74% zero-click, compared with just 31% for transactional ones. If your content strategy has been built around “what is” and “how to” articles chasing blog sessions, that traffic is always going to be the first casualty. It is commodity content competing for a commodity outcome.
There Isn’t Just One AI to Optimise For
Most zero-click conversations still treat “Google” as if it were the only system deciding what gets cited. It isn’t, and treating it that way will leave gaps.
According to Jasper’s 2026 GEO guide, the engines doing the citing don’t all work the same way. ChatGPT tends to lift structured formats, bullet points and FAQs, almost verbatim. Perplexity always shows citations and clearly favours original data over secondhand summaries. Google’s AI Overviews respond well to FAQ and HowTo schema paired with short, direct definitions. Bing Copilot leans toward step-by-step guides and comparisons. Claude, distinctly, rewards longer, coherent passages with real explanation rather than fragmented chunks.
That’s not a minor footnote. It means the content that gets you cited on one platform can actively work against you on another. A page chopped into bullet-heavy fragments might perform well in ChatGPT but read as thin, low-effort content to a human reader landing on it directly. Most teams are still writing one version of a page and hoping it works everywhere. It won’t, not consistently.
The practical fix isn’t five different versions of every page. It’s writing the core content for actual human comprehension first, the way Google’s own guidance recommends, and then making sure the structural elements, headers, schema, and a clear FAQ block exist on top of that so each engine can extract what it needs without you having to predict which one will find you first.
Start With an Audit, Not a Content Calendar
Before you write a single new page, find out what you already have and whether it’s working for you or against you.
Open Search Console and filter for pages with impressions over 500, click-through under 0.5%, and a position between 1 and 10. That’s your zero-click list, the pages getting seen but not clicked. Now sort each one into two buckets.
Good zero-click: your brand name is visible in the answer, even without a click, an AI Overview citation, or a featured snippet with your name attached.
Bad zero-click: the query gets fully answered with zero trace of you anywhere in it, a generic definition, a basic how-to, a comparison Google can already complete on its own.
Google SERP visibility used to mean one thing, a blue link in position one, it now means showing up anywhere on that page, AI Overview, featured snippet, knowledge panel, whether or not anyone clicks through.
Everything in the bad bucket needs a decision, not a rewrite. Either make it citation-worthy with original data or a sharper angle, or stop investing in it. Chasing volume on content Google will always summarize for free is the single most common waste of budget right now, and it’s exactly the content type where zero-click rates are highest: informational queries sit around 74% zero-click against just 31% for transactional ones, so this triage matters more for some content than others.
Write Answer First. Literally First.
This is the part most teams get backwards, because it’s the opposite of how blog writing has worked for a decade.
Old habit: build up to the answer with context, a story, three paragraphs of throat-clearing, then land the point.
New rule: state the direct answer in the first sentence of the section. Then back it with one piece of evidence, a stat, an example, a source. Then stop. One idea, fully resolved, before you move to the next.
Why this works mechanically: AI systems pull whichever passage answers the question most cleanly, and a section that’s still building toward its point three paragraphs in just doesn’t get picked. Google has confirmed that you don’t need to chop content into artificial chunks to make it extractable; you need to organise it for a human reader first, and the extractable structure follows naturally when each section actually resolves.
Practical test: read your H2, then read only the next sentence. If that sentence alone answers the heading’s implicit question, you’ve done it right. If it takes a full paragraph to get there, restructure.
Make at Least One Thing a Year That Can’t Be Summarised
Most content is a summary of something that already exists, which is exactly why an AI can summarise it again, one layer removed, with your name nowhere in it. The fix isn’t writing better summaries. It’s occasionally producing something that isn’t one.
That means running your own small survey instead of citing someone else’s. Publishing real numbers from your own client work, even directionally, instead of repeating industry averages. Building a free calculator, checklist, or tool that requires someone to actually land on your site to get value, not just read a paragraph about it. Interviewing someone with a genuinely uncommon point of view instead of summarizing the consensus one.
None of this needs to happen constantly. One solid original asset a quarter, something with a number nobody else has or a perspective nobody else is offering, does more for citation-worthiness than another twenty well-optimised summary posts. It’s also the one category of content the research consistently shows holding up best as AI absorbs the rest: original research and primary data aren’t as easy to flatten into someone else’s answer box.
This is brand authority SEO in its most literal form, the thing that makes Google and every AI system trust your name enough to say it out loud.
Wire Up Schema on Pages That Already Deserve It
Don’t start a schema project. Start with the ten pages you identified in your audit that are already close to citation-worthy, and give them the technical scaffolding that makes them easy to lift correctly.
FAQPage schema on a genuine Q&A section makes those questions eligible for drop-down answers directly in Google’s results. HowTo schema does the same for step-by-step content. Article schema helps establish authorship and recency, both of which feed into how much an AI system trusts the source enough to cite it by name.
Knowledge panel optimization runs on the same signals as everything else here, consistent entity data, clear authorship, and enough citation volume that Google stops guessing who you are and starts just knowing.
Pair the schema with the basics that still matter and that most teams skip: put the actual question in your title tag and meta description, not a vague keyword phrase, and link related question-content together so a crawler, human or AI, can see the page is part of a coherent body of expertise rather than a single orphaned post.
Featured snippets optimization isn’t a separate project from the one you’re already running, it’s just the audit and the schema work, pointed at the questions Google is most likely to answer directly.
One honest limit worth setting expectations on: schema makes good content easier to find. It does not make thin content good. If a page wasn’t earning citations before, marking it up correctly won’t change that on its own. What does change, when the content is actually good, is real: being cited inside an AI Overview is worth roughly 35% more organic clicks and 91% more paid clicks on that same search result, compared to not being cited at all. That’s the actual prize this whole exercise is chasing.
Replace Your Click-Based Reporting Before Your Boss Asks Why Traffic Is Down
This is the step most teams skip, and it’s the one causing the most internal damage. If the only number you’re reporting is sessions, you will look like you’re failing even when you’re not.
Build a simple two-column report: impressions and assisted outcomes on one side, clicks on the other. Pull pages with falling clicks but stable or rising impressions, that’s often a sign you’re being seen and cited, just not clicked, which is a very different problem than actually losing visibility. Worth knowing for that conversation: total web traffic barely moved last year, down just 0.43% across 50,000+ sites measured, even as organic search fell in most industries. Demand didn’t disappear. It moved channels, mostly into AI-driven and paid search, which is the case to make when someone asks why the click count looks worse than the business actually is.
Check your existing analytics for AI referral traffic before assuming you have none. AI chat tools attach UTM parameters to the links they share, so that traffic is often already showing up in your data under a channel nobody’s bothered to label. Tag it properly before spending on new tooling to track something you can already see.
Add one new line to whatever you report monthly: share of citations or mentions across AI answers for your three to five priority topics. It won’t be perfectly precise yet. Tracking it consistently, even roughly, shows whether your visibility is growing while your click count falls.
Build a Way for People to Find You That Doesn’t Run Through Google at All
Everything above makes you better at the game Google is still playing. This step takes you partly out of needing to play it.
An email list, a branded app, a social following someone checks directly, these are the assets a falling click-through rate can’t touch, because the relationship doesn’t depend on search discovery to begin with. If your traffic strategy is entirely “rank, get clicked, repeat,” you have zero insulation against whatever Google decides to summarise next.
Pick one direct channel and actually commit budget to it this quarter, not as a side project squeezed between content deadlines. A modest, engaged email list of a few thousand people who open consistently is worth more right now than another ten thousand words of content competing for an answer box.
And Therefore…
None of this is complicated. It’s just different work than the work most teams have spent a decade doing.
Audit what you have. Write the answer first. Build one page that serves five different readers. Make at least one thing this year nobody else can summarise. Wire up schema on what already deserves it. Citation report, not just clicks. And build at least one channel that doesn’t need Google’s permission to reach someone.
Do that, and the falling click-through rate stops being a crisis. It just becomes one part of a bigger picture you’re already winning on.
If you want help running this audit on your own content and figuring out which pages are actually worth saving, bring us to the table, and we’ll build the plan with you.
