It usually starts with a message like this: “Traffic on our blog is down 40% and nobody changed anything.” So I open Search Console, filter to the informational pages — the guides, the “what is,” the “how to choose” articles — and the pattern is almost always the same. Impressions flat, or even up. Average position holding. Clicks sliding down a slope that started somewhere in the last year.
Nothing broke. No penalty, no botched migration, no lost rankings. The page still ranks. The user just never needed to arrive. They asked a question, an AI Overview answered it in full at the top of the results, and the click that used to be yours never happened.
The instinct at this point is to treat it as a catastrophe — to go looking for what's “wrong” with the content. But before you rewrite anything, it's worth asking a more uncomfortable question: what was that content actually for in the first place?
That content was never really about the clicks
Here's the thing most dashboards quietly made us forget. On an e-commerce site, the blog was never there to generate clicks. Nobody's business plan depended on 8,000 monthly sessions to a “how to choose X” guide. That content existed to do slower, less countable work — to build topical authority so the category and product pages rank, to answer the questions people have before they're ready to buy, to signal to Google (and now to language models) that this brand actually knows its subject.
The clicks were a proxy for that work, not the work itself. We optimized for them and reported on them because they were easy to count — not because a session on a blog post was ever the point.
Which is why the right reaction to AI Overviews eating that traffic isn't to panic, and it definitely isn't to stop producing the content. The content is still doing its real job. What's changed is that the measurable byproduct — the click — is being intercepted. And if you were quietly treating that byproduct as the goal, this era is going to feel like a crisis when it's really a correction.
But there is a genuine loss hiding in here. It's just not the one on your clicks graph.
What I actually saw in the data
This isn't a hypothesis for me. I spent a good chunk of last year analyzing a set of domains that were all showing the same signature — brands built heavily on informational and educational content, all watching organic clicks decline while Google impressions held steady or climbed.
One domain made the pattern unmistakable. Comparing the ~170 days before and after AI Overviews started appearing on its queries: clicks roughly halved while impressions stayed flat. Of the ~7,600 keywords it ranked for, more than 5,300 now triggered an AI Overview. And here's the part that matters most — of those 5,300-plus AI answers its content was helping to generate, the brand was cited as a source only a few hundred times.
The content was good enough to help the model answer the question. It just wasn't structured well enough to get named for it. That's the whole story of AI Search in one line.
The real loss isn't the click — it's the citation
Zero-click is only half of what's happening, and it's the less important half. Losing a click is a traffic problem. Feeding an answer you don't get credited for is a visibility problem, and it's the one that compounds.
In classic SEO the equation was simple: rank = visible. If you were in the top three, people saw you. In AI Search the equation is different: cited = visible. Your page can be read, parsed, and used by the model to construct its answer — and the user still never sees your name. You did the research, wrote the guide, earned the expertise, and the model quietly synthesized it into a clean paragraph attributed to no one. Or, worse, attributed to a competitor who wrote a thinner article but structured it so the model could lift and credit it.
Being read by AI is not the same as being recommended by AI. The gap between those two is where brands are losing right now.
What decides whether you're the cited source or just raw material
So why does one page get named and a better-researched one get silently absorbed? A few things, roughly in order of how often they're the problem:
Factor 01
Extractability
Answer-first structure. One question per section, answered completely and early — before the context and the storytelling. Models lift clean, self-contained answers, not conclusions buried under 400 words of throat-clearing.
Factor 02
Structured data
Article, FAQ, and Product schema that tells the machine what it's looking at instead of making it guess.
Factor 03
E-E-A-T signals
Named author, real credentials, sources, recency — the things that make a model trust the page enough to cite it.
Factor 04
Entity & brand strength
Models recommend brands they recognize. A brand the model already knows as an authority in a category gets cited far more readily than an unknown one saying the same thing.
That last point deserves care, because it sits right at the edge of what optimization actually is. Brand strength is one of the biggest levers in GEO — but it gets built somewhere other than the optimization itself: through digital PR, brand activation, the work of making a name recognized. My work is optimization — structuring what's already there so AI can read, use, and attribute it. That happens without digital PR, without brand activation, without building your recognition from scratch. Those are real and powerful GEO levers, but they're a different discipline. The principle underneath is simple: optimization uses the brand potential that already exists — it doesn't manufacture it. A brand the model already recognizes gets cited more readily; what I make sure of is that such a brand is structured so AI can actually use it. What no schema can do is conjure a recognition that was never built — but that's a bigger topic, one I'll write about on its own.
When this isn't your problem
Not every page is exposed, and the honest move is to say so before anyone rewrites content that's working fine.
- Transactional and commercial-intent queries. AI Overviews are thinner on “buy,” “price,” “near me,” and comparison-shopping intent. The click still tends to happen.
- Low-AIO niches. Some industries barely trigger AI Overviews yet. “Yet” is doing work in that sentence — but if your queries don't trigger them today, this isn't your fire to fight this quarter.
- Content whose value was always the click-through. To a calculator, a configurator, a product. If the page's job was to move someone one step, not to answer a question, AI has less to intercept.
- Steady informational clicks. If your guides aren't bleeding, don't fix what isn't broken. Check first, act second.
How to check your own exposure
You don't need a tool to see whether this is happening to you. You need about an hour in Search Console and a bit of honesty.
- Segment. Split your informational URLs (guides, glossary, “what is,” “how to”) from your commercial ones. You're only investigating the informational set.
- Look for the diverging lines. Pull clicks and impressions for that set over the last 12–18 months. The signature you're looking for is impressions flat or rising while clicks fall. That divergence is the AI Overview effect.
- Check the queries manually. Take your top informational queries and search them. Do they trigger an AI Overview or AI Mode? When they do — are you named as a source?
- Ask the assistants. Put the same questions to ChatGPT and Perplexity and see who gets cited. If a competitor keeps coming up and you don't, that's your answer. (LLM-visibility tools like Chatbeat and others now entering the market do this at scale, but manual checks tell you plenty.)
The output is a short list: pages that are exposed and not cited. That's your priority. Everything else can wait.
So what do you actually do
Not what the panic suggests. Don't delete the content, and don't stop producing it — that content is still doing the real job it was always doing, building the authority your category and product pages depend on.
What changes is narrower than it feels. Take the exposed-but-uncited pages and restructure them to be answer-first and genuinely citable: the direct answer up top, clean sections a model can lift, schema that labels what's there, real author and sources. Then change what you measure. Clicks on an educational page were always a proxy; in this era they're a broken one. Track whether you're the cited source, whether the assistants recommend you, whether your share of AI answers in your category is going up.
Make peace with the click. It may not come back, and that's fine — because it was never the point. The point was to be the brand that gets recommended when the answer gets written. Do the content, do it well, structure it so you get named. Don't cry after the clicks.
I'll be honest about the state of this work: the methodology is still forming, the tooling is being built as I write this, and anyone claiming a settled playbook is selling you certainty that doesn't exist yet. But the direction is not ambiguous. The brands that restructure now will be the ones the models cite in two years. The ones waiting for the clicks to come back will be waiting a long time.
FAQ
Frequently asked questions
Because ranking and clicking have come apart. Your page can still hold position one while an AI Overview answers the question above it, so the user never needs to click through. Check whether impressions are flat or rising while clicks fall — that divergence is the tell.
No. That content builds the topical authority your commercial pages rely on and feeds the models that now answer for you. The blog on an e-commerce site was never there to generate sessions for their own sake. Keep producing it — just structure it to be cited, and stop judging it purely by clicks.
A model can read your page, use its facts to build an answer, and never name you. Being read is invisible and gives you nothing. Being cited puts your brand in front of the user as the source. In AI Search, only the second one counts as visibility.
Take your top informational queries and search them. If an AI Overview or AI Mode answer appears, your content is likely feeding it. Then check whether you're named as a source. Pages that trigger an overview but don't earn a citation are your priority list.
Volume alone does neither. Breadth of genuinely useful, well-structured answers signals authority and gives models more reasons to cite you. A pile of thin, keyword-stuffed pages does the opposite. It's not about more — it's about more that's citable.
No, and the penetration is uneven. Informational and educational niches — health, finance, how-to, comparison — are hit hardest. Transactional and narrow local queries much less so, for now. Measure your own exposure before you act on someone else's headline.
