SEO in 2026: What Changed When Google Added AI Mode

In short: AI Mode has worked in Croatian since 7 October 2025. AI Overviews have been in Croatian results since mid-May 2025, nearly five months longer, and they are the ones touching your traffic every day. Zero-click isn't the future, it's the present, and there is now causal proof rather than correlation: a randomised field experiment measured that an AI Overview cuts outbound clicks by 38%. But the biggest shift isn't the click loss, it's the quiet rule underneath it: the share of citations in Google's AI answers coming from the organic top 10 fell from around 76% to 37.9% in eight months. Ranking in the top 10 no longer guarantees a citation. Below are the numbers, the sources, and a clear line between what still works in SEO and what is dying. Wherever there is only one source, or the source is a vendor measuring itself, it says so.

What changed: Google switched on AI Mode

The timeline explains why the "something broke" feeling showed up exactly now.

AI Mode started in March 2025 as a Search Labs experiment, English only and invite only. Google announced it for all US users at I/O on 20 May 2025, the rollout ran through June and finished in early July 2025. Croatian wasn't on the list yet.

For the Croatian market, October is the turning point. AI Mode and the Croatian language arrived together in the global expansion of 7 October 2025, in a batch of more than 35 new languages and more than 40 new countries and territories. Croatian is named explicitly among the new languages.

Why is that a break? Because here it went from "not available" to "available" overnight. As late as August 2025 local commentators were writing that AI Mode wasn't available in Croatia. Two months later it was.

Keep two things apart. AI Mode is a separate, conversational tab. AI Overviews are a summary embedded in the classic results. AI Overviews showed up in Croatia far earlier, in mid-May 2025, the same week Google announced the expansion to more than 200 countries and more than 40 languages. Users see them without a single extra click, which is why they, not AI Mode, are what's eating your traffic right now.

What the zero-click reality looks like in numbers

Zero-click isn't new. Before AI Overviews went mainstream, 58.5% of US searches and 59.7% of EU searches ended without a click to the open web (SparkToro, Rand Fishkin, on Datos clickstream data, 2024). That's the baseline. AI pushes it higher: the same kind of measurement in early 2026 lands around 68%.

By how much exactly? Ahrefs measured 300,000 keywords and found a 34.5% drop in position-one CTR for informational queries when an AI Overview appears. That's the number everyone quotes here too. Worth knowing that a more recent Ahrefs measurement put the same drop at 58%, so the figure we all copy is already outdated in our favour, not against us.

The strongest evidence is neither of those two, because both are correlation. The strongest evidence is an experiment.

Saharsh Agarwal and Ananya Sen (Indian School of Business and Carnegie Mellon) ran a randomised field experiment on 1,065 Chrome users in the US, across two weeks in January and February 2026, pre-registered with the AEA RCT Registry. The result: the presence of an AI Overview cuts outbound clicks by 38%. Without it a user produces 0.61 clicks per search, with it 0.38. The zero-click share rises from 54% to 72%. The effect is strongest when the AI Overview sits at the top of the page, which it did 85% of the time.

In other words, the AI summary itself causes the click loss. That is no longer a matter of interpretation.

How often does an AI summary even appear? Pew Research Center tracked 900 American adults and 68,879 of their Google searches through March 2025. The result: 18% of searches produced an AI summary, and 58% of users saw at least one that month.

The trigger depends heavily on query shape. One or two words produce an AI summary 8% of the time. Ten or more words, 53%. Queries starting with a question word ("who", "what", "when", "why"), 60%. The longer and more informational the query, the better the odds that AI, not your page, answers it.

One more thing that matters. The trigger rate doesn't climb steadily. Semrush tracked it through 2025 from 6.49% in January to 24.61% in July, then down to 15.69% in November (single source, vendor's own data). Google actively tunes when it shows an AI answer. This is not a one-way ramp.

And Croatian data? There isn't any, and that's a finding in itself. For the Croatian market there is no published measurement of how often AI Overviews appear or how many clicks they take. The numbers circulating in local articles are either borrowed global figures or carry no source at all. Everything you've read above was measured on US results, mostly in English.

That is not a detail. Croatian has only been in AI answers since October 2025, the market is small, and the pool of sources a model builds an answer from is far thinner than in English. There is no reason to assume the US percentages transfer one to one, in either direction. Until someone measures it, the honest thing is to say so and go measure it yourself: that is why we run the Adria AI Visibility Index.

How AI Overviews pick sources (and why "92%" misleads)

This is where the biggest misconception in the local SEO conversation breaks.

You'll often hear that AI Overviews cite at least one page from the organic top in over 90% of cases. The best documented measurement of that kind is seoClarity's: across 362,000 US queries in October 2025, 94% of AI answers had at least one overlap with the organic top 20.

True, but it measures the wrong thing. It tells you how many AI answers touch at least one highly ranked page. It doesn't tell you what share of individual citations comes from there. An answer with ten citations of which one comes from the top 10 counts toward that 94% exactly like an answer with a single top-10 citation.

That second, more important number is far lower and falling. In March 2026, across 863,000 keywords and 4 million AI Overview URLs, Ahrefs measured 37.9% of citations coming from the organic top 10, against roughly 76% in July 2025. The rest splits almost evenly: 31.2% from positions 11 to 100 and 31.0% below position 100.

One caveat Ahrefs states itself, and which gets lost in the retelling: part of that gap isn't a change in Google's behaviour but a change in Ahrefs' own parsing method, plus wider use of "query fan-out", where Google breaks one query into several sub-queries and looks for sources for each. It lines up in time with Gemini 3 becoming the global default behind AI Overviews on 27 January 2026, but "almost overnight" overstates it. Part model, part measurement.

BrightEdge, over a 16-month sample and a different method, goes lower still: 16.7% of citations from the top 10.

And there's another side worth saying out loud. The same BrightEdge analysis shows citation overlap with organic results anywhere climbed from 32.3% to 54.5%. But that growth comes from positions 21 to 100, not from the top. Translation: ranking still helps, the benefit has just moved out of the top 10 and into the depth of the index.

This is where classic SEO turns into a new discipline, optimizing for AI answers. If you want to know what GEO is and why it exists, that's the bridge out of these numbers. And how visible Croatian and regional brands actually are inside AI answers is measured by the Adria AI Visibility Index.

What still works in classic SEO, and what is dying

Start with what still holds.

Ranking helps citations, but the link is weakening. The share of citations from the top 10 fell from around 76% to 37.9% in eight months (Ahrefs), with the methodology caveat above. Staying at the top is still the single most profitable thing you can do, it just isn't sufficient any more.

Links work indirectly. Across 75,000 brands Ahrefs measured that raw backlink counts correlate weakly with AI answer visibility, at 0.218. Brand mentions correlate far more strongly, at 0.664, and YouTube mentions strongest of all, at 0.737. The point: build a reason to be mentioned, not just to be linked. And don't ignore YouTube, which was the most cited domain in our own probes.

Freshness pays, but watch the viral claims. Across 17 million citations on seven platforms Ahrefs measured that content AI cites averages 25.7% fresher than the organic top 10 (1,064 versus 1,432 days). The same analysis warns that the popular "4.3x more citations for fresh content" claim has no source. A good reminder to check numbers, especially the ones that flatter you.

And traffic quality? There's a countertrend here. Semrush claims visitors from AI search convert 4.4 times better than the average organic visitor. That's one source, its own panel, with no independent replication. Other vendors publish even larger multiples, which is a reason for more scepticism, not less. Treat it as a direction, not a guarantee. If it holds, fewer clicks doesn't have to mean fewer customers.

Now what's dying.

Listicles and "best of" content are losing ground. BrightEdge measures a drop of around 60% in impressions for ranking-style content, and 14% for comparison queries. Over the same period impressions across all content rose more than 49%, while click-through fell about 30% against May 2024. Impressions up, clicks flat. That's the whole story in two numbers.

Thin, keyword-stuffed informational content is the most exposed. Ahrefs' 300,000-keyword study shows AI Overviews cut clicks hardest on informational queries. And the KDD 2024 GEO study measures keyword stuffing as the only tested tactic that delivers nothing at all in generative engines.

And briefly on Google's story. Google claims AI Overviews remove "bounce clicks" and that total clicks are stable, but it has never published numbers to support it. After the Agarwal and Sen experiment, that claim stands on weaker ground than before.

Implementation: where to invest in 2026

The data sorts the priorities cleanly.

First, keep ranking. Overlap with organic results is growing, it has just moved deeper into the index. SEO is the foundation, not a relic.

Second, build brand mentions and digital PR. They correlate with citations three times more strongly than raw links (0.664 against 0.218), and YouTube more strongly still.

Third, maintain freshness. Cited content averages a full year younger than the organic top 10. Updating an old page is cheaper than writing a new one.

Fourth, don't live on Google alone. The share of traffic that reaches you keeps shrinking no matter how well you rank.

Fifth, measure. If you don't know whether you appear in AI answers on your core topics, you're flying blind. It's the only item on this list you can start today and have a result by tomorrow.