On 19 May 2026, Google took to the stage at I/O and announced something it had been dropping in bits and pieces for months: AI Mode is switched to work with Gemini 3.5 Flash as the default model, worldwide, within the search engine. What sounds to a user like “the search bar answers better”, to those of us working in digital marketing means something different: the shop window where we've been scraping pennies to appear for years has changed shape. And, as I write this from Barcelona at the end of May, we are already seeing the first real scare in the reports of clients who depended too much on classic organic traffic.
At Vandelay we have spent the last two weeks looking at data, comparing reports and cross-referencing what Google says with what we see in real accounts. This article is the summary that we would like to have read ourselves: what exactly is Google AI Mode with Gemini 3.5 Flash, What's changing in SEO and paid campaigns, and where to start if you don't want the next quarter to catch you on the wrong foot.
What exactly was announced at Google I/O 2026?
Let's get down to the verifiable, unvarnished facts. At the I/O 2026 keynote, Google confirmed three things that directly affect advertising and positioning:
- AI Mode surpasses 1 billion monthly users and becomes the fastest growing search mode within the search engine, according to the data published by Google in its Search official announcement at I/O 2026.
- Gemini 3.5 Flash becomes default model of AI Mode in Search and of the Gemini app globally, according to the official summary of the 100 new features of Google I/O 2026.
- Gemini 3.5 Flash beats 3.1 Pro in coding, agent and multimodality benchmarks while maintaining the cost and speed of the Flash series, in line with the technical coverage of the MarkTechPost of 20 May.
Translated into agency-speak: the engine that decides what to answer and who to cite is now faster, cheaper for Google and better able to reason through several steps in a row. This explains why AI Mode no longer simply summarises three pages: it strings together queries, launches actions and returns a single response with citations, instead of the ten blue links of a lifetime.
The inconvenient fact: the age of the click is emptying out
While Google celebrates the launch, independent analysts put another fact on the table. According to analysis published in May 2026 by Pasquale Pillitteri, based on aggregate data from SISTRIX and Similarweb, 93% of the searches within AI Mode end without a single click to an external website.. The same report documents that the CTR of the first organic position in queries with AI features has dropped from 27% to 11% as measured by SISTRIX in March 2026 (source).
I'm not going to sell you that SEO is dead, because it's not true and no one with a modicum of honesty would claim that. What is dead, or dying in a hurry, is the idea that appearing first equals getting traffic. Today, appearing first can mean that your phrase appears literally within the generated response and that the user closes the tab without going through your website. Good for your brand, terrible for your Analytics.
What changes, in particular, for a brand or an e-commerce site?
1. Visibility is no longer measured only in positions
Until two years ago, the Search Console panel was the oracle. Today it coexists with new metrics that SEOs now call it, almost shamelessly, AI Share of VoiceHow many times your brand is cited in responses generated by AI Mode, AI Overviews, ChatGPT Search or Perplexity. Tools such as Profound, Otterly or Peec AI have been offering this tracking for months, and what was once a whim has become indispensable for any client relying on the organic channel.
2. Content has to be citable, not just positionable
Gemini 3.5 Flash, like all LLMs, tends to cite sources that meet recognisable patterns: descriptive headlines, clear definitions in the first 80 words, data with number and date, visible authorship and linked entities. This is very similar to what SEO has been asking for all along, but with a twist: generic paragraphs full of filler no longer compete. The model needs to extract a short, verifiable sentence. If your post doesn't contain it, it doesn't cite you.
3. Google Ads reorients around the conversational user
At the same keynote, Google confirmed that ads within AI Mode will be integrated as a natural part of response, not as a separate block. AI Max for Search Campaigns, which we were already testing in Spanish accounts, gains prominence because it allows Google to match fuzzy intent with creative without the advertiser having to define each query. It's more convenient, yes, but it also forces you to have solid fundamentals: clean feeds, well-crafted exclusions and conversion metrics that actually mean something, not phantom clicks.
4. Latency and cost-per-response matter more than ever before
This is where Google's business move comes in. Replacing the Pro model with a faster Flash allows the company to serve more complex answers without increasing the cost per query. For an agency this translates into something very practical: the answers will be more ambitious (more steps, more cross-sources, more reasoning), but also faster, and therefore the user will have less reason to leave the SERP and open your landing page. Every extra millisecond that your website takes to load is a direct competitive disadvantage compared to Google's own response.
What we have been doing in Vandelay since May 20
So that it doesn't remain in theory, I'll tell you what we already have in place in the accounts we manage. There are five fronts and each one has an owner within the team, because not all digital marketing profiles are prepared for the same thing.
Citability audit. We review the posts with the best traffic history and rewrite their first 80 words to contain a self-contained definition and a recent figure. We don't touch the URLs or H1s, just the block that a model would need to cite the page without hesitation. It's a surgeon's work, not a bulldozer's.
Implementation of llms.txt. We wrote it a few weeks ago in this blog: this file tells the crawler which pages you want to cite and which you don't want to cite. With Gemini 3.5 Flash doing more reasoning on the indexable corpus, controlling what goes in is as important as the classic robots.txt.
Mention tracking in AI Mode. We have set up dashboards that cross classic positioning with mentions in generative responses, so that the client can see what is happening in each channel without having to open three different tools. The difference between “losing positions” and “losing citations” determines the decision: one can be fixed with good link building, the other with rewriting key fragments.
Review of Google Ads campaigns. We start migrating traditional campaigns to AI Max for Search in accounts where attribution is well resolved. Where it is not, we first fix measurement. Activating AI Max without a good GA4 or Enhanced Conversions is like hitting the accelerator with the brakes loose.
Human messages where AI cannot reach. This is what matters most to us. AI Mode solves queries, but it doesn't build brand. People still buy from people, so we are reinforcing newsletters, LinkedIn founder content and short video formats where the customer's face and voice are recognisable. If Google is going to keep the click, at least stay clear on who you are quoting.
Mistakes we are seeing and should avoid
For every client that is adapting well, we see three others reacting badly. The most common mistakes in these weeks are three and I'll list them quickly in case they ring a bell: confusing traffic drop with demand drop (it's not the same thing, and sometimes the brand is growing even if the CTR drops), asking the AI to rewrite the whole blog “to please Gemini” without understanding what the model actually does, and cancelling paid campaigns because they “no longer work”, when in fact what has changed is attribution and not performance.
If you ask me which signal to watch out for between now and the end of June, I would say two: the evolution of the percentage of impressions of your brand within AI Overviews for your core queries, and the ratio between direct and organic traffic. If direct traffic rises while organic traffic falls, you are not losing your audience: you are converting it into a recurring audience. If both fall, that's where you need to act.
Where we go from here
Gemini 3.5 Flash is just the first link. Google has already advanced that during the summer of 2026 Gemini Omni and Gemini Spark capabilities will enter into the AI Mode itself, with access to external tools and persistent user memory. For an agency like Vandelay, that means the job in the coming months is not to decorate landings or move budgets in bulk, but to understand how each customer behaves when half of their funnel occurs within a generated response that doesn't even open a tab.
Digital marketing has not become harder, it has become less lazy. And that, frankly, doesn't strike us as bad news.
Do you have active Google Ads campaigns or a blog that relies on organic traffic and want to know how the change affects you? Write to us at vandelay.co.uk/contact and we'll set up a no-obligation citation audit for you.