Between January 2024 and March 2026, Google shipped more changes to how search works than in the previous five years combined. AI Overviews replaced featured snippets. Zero-click searches climbed from a trend to the default behavior in AI-powered interfaces. An entire category of websites — built on mass-produced AI content — lost 30 to 60 percent of their organic traffic in a single quarter. The rules for what ranks, what gets cited, and what drives actual clicks shifted beneath everyone's feet.
We operate content hubs and landing pages across 9 LATAM markets. Organic traffic is one of our two primary acquisition channels, alongside paid search through Google Ads. What follows is not speculation about where search is headed. It is a reconstruction of what already happened, built on Semrush's published datasets, Google's own announcements, and changes we observed in our own Search Console data.
2024: The Helpful Content Update and the E-E-A-T Reckoning
The Helpful Content Update had been rolling since late 2023, but 2024 was when the consequences became impossible to ignore. Sites that had built their entire strategy on high-volume, surface-level articles — many of them using early AI generation tools — started seeing 20 to 60 percent traffic drops. Not gradual declines. Sudden, cliff-edge losses that showed up in analytics over a single week.
What made this different from previous Google updates was the mechanism. The Helpful Content signal operates at the site level, not the page level. One section of low-quality content could drag down rankings for everything else on the domain. A site with 200 solid articles and 50 thin AI-generated ones didn't just lose rankings on those 50 pages — the quality penalty spread across the entire domain. Google's own documentation confirmed this: the classifier evaluates a site's overall content quality, and low-value pages act as a tax on every other page.
E-E-A-T — Experience, Expertise, Authoritativeness, Trustworthiness — stopped being an abstract concept discussed in SEO blogs and became the primary filter Google applied to separate useful information from noise. The "Experience" component, added in December 2022, started mattering in practice during 2024. Google's systems got measurably better at detecting whether content came from someone who actually did the thing they were writing about, or someone who summarized other people's summaries.
The March 2024 core update was especially aggressive. Google explicitly targeted "scaled content abuse" — producing large quantities of content purely to rank, regardless of whether AI or humans wrote it. Sites that had been publishing 30, 50, 100 articles a day with no editorial process saw their indexed page counts drop by half overnight. The update also went after expired domain abuse and site reputation abuse (parasite SEO), where operators rented subdirectories on high-authority domains to borrow their ranking power.
SGE — Search Generative Experience — was still in beta for most of 2024, available to users who opted in through Search Labs. Most site owners ignored it. That turned out to be a mistake, because everything SGE previewed became the default experience less than a year later.
2025: AI Overviews Roll Out — and the Data Is Complicated
AI Overviews went from experiment to production feature in 2025, but their rollout was more volatile than most coverage suggested. Semrush tracked over 10 million keywords throughout the year. In January 2025, AI Overviews appeared on 6.49% of tracked queries. By July, that number peaked near 25%. Then it dropped — by November, coverage was back down to 15.69%. Google was testing, adjusting, pulling back, and re-expanding throughout the year.
The distribution was uneven. 82% of desktop AI Overviews appeared for keywords with less than 1,000 monthly searches — the long tail, not the high-volume head terms. About 35% were triggered by question-based queries (who, what, why, when, how). And the intent mix shifted dramatically: in January 2025, 91.3% of AI Overviews served informational queries. By October, that dropped to 57.1%, with commercial, transactional, and navigational queries making up the other 42.9%. Google was expanding AI Overviews beyond pure information into buying decisions.
The click-through rate impact was real but more nuanced than early panic suggested. Semrush's analysis of 200,000+ keywords found something counterintuitive: when examining the same search terms before and after AI Overviews appeared, users actually clicked slightly more with the overview present. The overall zero-click rate for keywords with AI Overviews slowly declined through 2025. But this masks a larger truth — AI Overviews tend to appear on queries that already had high zero-click rates. The feature didn't create zero-click behavior so much as it concentrated where zero-click queries live.
The broader zero-click picture was less ambiguous. 58.5% of US searches and 59.7% of EU searches ended without a click in 2024. By Q4 2025, zero-click in the US rose to 27.2% (up from 24.4% at the start of the year, measuring only the queries that had previously generated clicks). The overall direction is clear even if the AI Overview-specific impact is debatable.
AI Mode: 93% of Sessions Never Leave Google
In March 2025, Google launched AI Mode — a standalone conversational search interface powered by Gemini 2.0. This was fundamentally different from AI Overviews. AI Overviews are summaries above traditional search results. AI Mode is a separate experience entirely: a chat interface where users ask questions, get follow-up responses, and can refine their queries without ever seeing a list of blue links.
Semrush analyzed 69 million search sessions between May and July 2025 to measure AI Mode's impact. The zero-click rate: 93%. Only 6 to 8 percent of AI Mode sessions resulted in a click to an external website. For comparison, traditional search had a 34% zero-click rate, and searches with AI Overviews showed about 43%. Each step up the AI ladder roughly doubled the chance that the user never left Google.
AI Mode was initially available to Google One AI Premium subscribers and then expanded to broader audiences. As of early 2026, it remains a separate tab in Google Search, not the default. But its existence signals where Google believes search is heading: conversational, answer-complete, and self-contained within the Google ecosystem.
Google Citing Google: The Self-Referential Loop
A pattern emerged in AI Mode that deserves its own section because of what it implies for the future. SE Ranking analyzed 1.3 million citations within AI Mode responses. By February 2026, Google.com accounted for 17.42% of all citations — nearly triple its 5.7% share from June 2025. Google was increasingly citing itself as a source.
59% of those self-citations pointed to traditional Google Search results pages — not Google Business Profiles or YouTube, but links that sent users back into Google's own search results. Combined with YouTube and other Google properties, roughly one in five AI Mode sources was Google-controlled.
The implication for independent publishers is straightforward: AI Mode is designed to keep users inside Google's ecosystem. External citations exist, but they compete with Google's own content for a shrinking share of references. Being cited in AI Mode responses is valuable — but the share of external citations is decreasing, not increasing, as the feature matures.
2026 Q1: Three Updates, One Message
Google shipped three updates in the first quarter of 2026. Each one targeted a different symptom of the same disease: low-quality content built to game rankings rather than help users.
The January core update reweighted E-E-A-T signals. The specific target: AI-generated content published under the byline of a human "expert" who had no verifiable connection to the topic. Google began cross-referencing author names against LinkedIn profiles, professional registrations, and academic publications. Sites in health, finance, and legal niches — YMYL categories — saw immediate ranking shifts. Attribution without verification stopped working.
February brought a spam update with explicit enforcement against scaled content abuse. Two categories hit hardest: sites publishing dozens of articles daily with no editorial process (the AI content farm model), and expired domain abuse (buying dropped domains with existing authority and loading them with unrelated content).
The March core update was the largest. Sites relying on templated, mass-produced AI articles lost 30 to 60 percent of organic traffic. SimilarLabs tracked the decline across their index; SiteGrade confirmed similar patterns in their agency portfolio. Parasitic SEO — coupon sections on news sites, review directories on university domains, affiliate content riding hospital trust signals — was deindexed or massively downranked.
On the other side: sites with original research, proprietary data, and first-party analysis gained an average of 22% in visibility. The signal was unmistakable. Google is not anti-AI. Google is anti-content-without-value, and AI made it trivially easy to produce content without value at scale.
Original Data Becomes the Only Moat
If there is one clear pattern across every update since 2024, it is this: content containing information that exists nowhere else on the web gained ground. Everything else lost it or held steady at best.
The +22% visibility increase from the March 2026 update came specifically from pages with proprietary data. Not pages that cited someone else's data. Pages where the data originated. First-party surveys. Campaign performance numbers. Market analysis from original datasets. The kind of content that requires actually operating in a market, not just writing about it.
For us, this validated a decision we made in late 2024: publishing operational data from our own campaigns as content. Click costs by market. Conversion rate benchmarks by vertical. Infrastructure approaches we built and tested. None of it is secret. But all of it is original. Nobody else has our campaign data from 9 LATAM markets. That makes it, by definition, content that AI cannot generate and competitors cannot replicate without doing the same work.
This is not an argument against using AI as a writing tool. It is an argument that what you write about matters more than how you write it. AI can polish prose, fix grammar, suggest structure. But it cannot invent data points from campaigns it never ran, and Google's systems are getting better at distinguishing between content that reports original findings and content that rearranges existing information.
Passage Ranking and Entity Optimization
Google now indexes and ranks individual passages within a page, not just the page as a whole. This has been true since late 2021, but it became dramatically more important with AI Overviews, because the AI system pulls specific passages as citations. A well-structured article where each H2 section answers a specific question completely — with the answer in the first two sentences, then supporting detail — gets indexed passage-by-passage. Each section is a potential citation in an AI Overview.
Pages outside the top 10 in traditional rankings are now getting cited in AI Overviews. Originality.ai analyzed citation patterns and found that a significant share of AI Overview sources came from pages ranked well below position 10 — positions that would have been effectively invisible in pre-AI search. Being on page two is no longer irrelevant if your content gets pulled into an Overview.
Entity optimization — mentioning specific, named things instead of generic descriptions — correlates strongly with AI Overview citation. Not "a popular digital wallet" but "Nequi, Colombia's largest digital wallet with 17 million users as of 2025." Not "a major search engine update" but "the March 2026 core update that began rolling out on March 3." Google's Knowledge Graph maps entities across the web, and content dense with correctly identified entities gets matched more reliably to user queries.
Structured data (schema markup) plays into this. Article schema, FAQ schema, author schema, organization schema — these give Google's systems explicit signals about what a page contains, who wrote it, and what entities it references. Sites with complete schema markup are overrepresented in AI Overview citations relative to their traditional ranking positions.
What This Means If You Buy Traffic for a Living
If your business depends on web traffic — whether you buy it through Google Ads or earn it through content — four things have changed structurally since 2024.
Organic traffic from informational queries is declining in absolute terms. Not dying, but shrinking. The zero-click trend is real and accelerating, especially in AI Mode. A content hub that generated 50,000 monthly sessions in 2024 might generate 35,000 to 40,000 in 2026 with identical rankings. Planning content investments based on 2024 traffic baselines will overestimate returns.
The content that survives and grows is content that AI cannot generate. Original data from your own operations. Campaign performance metrics. Market-specific benchmarks from markets where you actually operate. Real testing with documented results. Google rewards the things that are hard to fake and impossible to generate from a prompt. For a media buying operation, the data you collect daily is more valuable as content than any article an AI could write about the same topic from publicly available sources.
SEO and paid traffic are converging. Organic visibility amplifies paid performance. A site that ranks organically and gets cited in AI Overviews creates a trust signal that improves ad click-through rates on the same queries. Running search ads without any organic presence is increasingly expensive. Running content without ads leaves money on the table. The two channels reinforce each other now more than at any point in the last decade.
The companies that handle this transition well are treating content as a data publication, not a keyword-targeting exercise. Publishing what they know from operating in their markets. Attaching verifiable author credentials. Building entity-rich, structured pages that AI systems can parse and cite. None of that requires a massive content team. It requires having something genuine to say — and saying it in a format that both humans and machines can understand.
