This isn’t a future scenario. It’s happening now, and the traffic data is worse than most brands are prepared for.

What Google AI overviews are actually doing to clicks

When a generative summary appears at the top of a search results page, users stop there. They’ve got what they came for. Why click?

This “zero-click” pattern has the data behind it now. A multi-quarter analysis tracking 64 websites found that AI Overview appearances caused an immediate 16% drop in organic search clicks. As Google extended the feature, the decline deepened, eventually settling at a sustained 42% drop compared to pre-AIO traffic levels, per TSEG reporting.

Evergreen content takes the worst of it. Breaking news doesn’t.

A separate study with a pre-AIO baseline of 1.7 billion clicks per quarter found that the damage wasn’t spread evenly. Guides, tutorials, and reference articles lost 35% of their organic clicks. Breaking news grew 103% across Google’s surfaces.

Why the split? LLMs are slow with real-time information and prone to errors during active news cycles. Google knows this, so it doesn’t run generative summaries for fresh news queries and falls back to the traditional news carousel instead. Your evergreen how-to guide doesn’t get that protection.

That same study also found a major redistribution in how traffic arrives. Web search traffic for publishers fell 43%. Traffic from Google Discover grew 30%. For the first time, those two sources are roughly equal in volume. A big driver was a traffic spike following Google’s December 2025 Core Update, which was partially dialed back by a Discover-specific core update in February.

The CTR numbers from Seer Interactive

The most detailed data on what AI Overviews do to click-through rates comes from a study published by Search Engine Land, covering 3,119 informational queries across 42 client organizations, with 25.1 million organic impressions analyzed.

Organic CTR for queries featuring AI Overviews fell 61%, from 1.76% to 0.61%. Paid CTR on those same queries dropped 68%, from 19.7% to 6.34%.

The less obvious finding: queries without AI Overviews are also getting worse. Non-AIO organic CTR started at 2.72% in June 2024, briefly peaked at 3.14% in February, and fell to 1.62% by September 2025. That’s a 41% decline on queries that have no generative summary at all. Some users are going to ChatGPT or Perplexity before they ever hit Google. Others just aren’t clicking anything.

One exception stands out. Brands that get cited directly inside an AI Overview earn 35% more organic clicks and 91% more paid clicks than competitors who don’t appear in the summary. Getting named in the generated answer is now worth more than ranking second in standard results.

Query performance metricPre-AIO baseline (June 2024)Post-AIO standard (September 2025)Net changeOrganic CTR, AI Overview, uncited1.76%0.52%-70.45%Organic CTR, AI Overview, cited1.76%0.70%-60.23%Organic CTR, no AI Overview2.72%1.62%-40.44%Paid CTR, AI Overview, uncited19.70%4.14%-78.98%Paid CTR, AI Overview, cited19.70%7.89%-59.95%Paid CTR, no AI Overview19.10%13.04%-31.73%

SEO, AEO, and GEO: three different jobs

Traditional SEO still matters. It builds the technical foundation everything else sits on. But it won’t get your brand cited in an AI Overview or recommended by Perplexity. For that, two other disciplines apply.

Answer Engine Optimization (AEO) is about making content easy to extract. Systems that deliver direct answers through featured snippets, voice interfaces, and AI Overviews look for dense, self-contained paragraphs that hold their meaning without surrounding context. The core signal isn’t keyword density. It’s how much factual content you can pack into a short, citable block.

Generative Engine Optimization (GEO) operates at a broader level. It’s not about real-time crawls. It’s about how models like ChatGPT, Claude, Gemini, and Perplexity understand your brand inside their training data and retrieval systems. Entity associations, topical authority, and consistent brand signals across the web all feed into this. The catch: GEO changes take 6 to 12 months to show up because LLM training runs on its own schedule.

SEOAEOGEOPrimary objectiveRank URLs for organic clicksServe content as the direct answerGet cited and recommended by AI enginesSystem targetsGoogle and Bing web indexesAI Overviews, featured snippets, voice assistantsChatGPT, Claude, Gemini, Perplexity, GrokContent architectureLong-form topical pagesShort Q&A blocks, FAQs, structured listsWhitepapers, first-party data, authority guidesCore metricsKeyword match, click volume, backlinksFact density, intent matching, schema validationEntity disambiguation, citation volume, multi-source validationPropagation speedDays to 30 days30 to 60 days6 to 12 months

The entity debt problem

There’s a technical liability most brands haven’t addressed yet.

For years, SEO rewarded sites with strong backlink profiles and keyword-optimized pages. LLMs don’t work that way. They’re not reading your site like a crawler does. They’re trying to figure out who you are, what you sell, and how you relate to other companies and concepts in your market. If your digital footprint doesn’t give them clean answers, they’ll either guess wrong or leave you out.

Entity debt accumulates when a site lacks machine-readable metadata: no proper schema markup, business information that contradicts itself across platforms, duplicate Google Business profiles, missing Knowledge Graph entries. The LLM hits the site and can’t figure out what it’s looking at.

Fixing it requires two separate layers of work.

The first is entity identity. Schema markups, entity linkages, and Knowledge Graph alignment tell AI systems who the brand is without requiring them to infer it from context. Practically, that means claiming Place IDs through the Google Places API, auditing for duplicate profiles, and correcting fragmented maps listings that split brand authority across multiple entries.

The second is brand authority. Once identity is established, AI engines assess whether the brand is worth citing. This isn’t about backlinks. Media mentions, expert citations, verified reviews, and consistent content across channels all feed into the credibility signal that determines whether a model names you or a competitor.

What the UK regulators are requiring

The economic pressure generative search has put on publishers has pushed regulators to move. The UK’s Competition and Markets Authority has classified Google as having “strategic market status” and is now imposing transparency obligations around its generative features.

Under these rules, Google must provide clear attributions and direct links to publisher content inside AI-generated answers. Publishers get page-level and directory-level controls to block their content from being used to train or ground Google’s generative features. Google cannot penalize sites in organic rankings for opting out. The compliance deadline is nine months, with periodic reporting required.

In response, Google is testing a new toggle inside Search Console, starting with UK publishers before a wider rollout. The toggle controls whether a site’s content appears in AI Overviews, AI Mode, and generative Discover results. Opting out removes the generative impressions without affecting standard organic rankings. Google is also providing more granular impression and citation data broken down by country alongside the toggle.

Reddit and preferred sources

Two other changes are worth watching.

Google is now surfacing Reddit threads and social discussions inside AI search results under labels like “Community Perspectives” and “Expert Advice,” with usernames and creator handles attached. A “Further Exploration” feature suggests related follow-up topics to keep users in the search interface longer.

Users can also now choose preferred publications and websites that influence what appears in their AI Overviews and AI Mode results. Selected sources are labeled in results and act as a personalization filter layered on top of standard ranking signals.

The Skylumine framework

Skylumine, an Indian agency that describes itself as AI-first, has built a framework specifically around the entity debt and GEO problems described above. Its stack is called ZozoStack, a set of six tools aimed at making brands more legible to AI engines.

The six components, per documentation on kickassdigitalmarketing.com:

ZozoStack Context generates lightweight entity nodes designed for fast ingestion by LLM crawlers, bypassing the bloat common in standard CMS builds.

ZozoStack Schema automates Knowledge Graph alignment by converting site content into structured semantic profiles, so platforms like ChatGPT and Gemini can map a brand’s relationships back to authoritative external databases.

ZozoStack LLMS manages llms.txt files and bot access policies, blocking unauthorized scraping while ensuring that primary answer engines retain access.

ZozoStack Sight tracks AI recommendations, entity health, and brand sentiment across LLM platforms, providing visibility metrics comparable to traditional keyword rankings.

ZozoStack Flow is a transactional API layer that lets AI agents check inventory, retrieve pricing, and complete purchases without requiring human interface navigation.

ZozoStack Vision is a dashboard monitoring share of voice and generative visibility across major AI engines.

Pricing starts at Rs. 50,000 for website development, Rs. 15,000/month for SEO and entity alignment, and Rs. 20,000/month for a combined AI and performance marketing package.

The metrics that matter now

Page views and raw click volume aren’t the right scorecards anymore. Users are getting answers on the search page. The visit may never happen, but the brand impression did.

Three metrics are replacing the old KPIs, as Aja Frost outlined in a Search Engine Land interview.

Visibility is the lead metric now because it’s what starts the conversion path. A brand cited in a synthesized AI answer builds recall even without a click. Conversions often come days later, when the user opens a new tab and searches the brand name directly.

Share of voice compares your visibility to competitors across generative engines. If total platform traffic falls but your share of voice holds or grows, the GEO strategy is working. The ratio tells you what raw traffic numbers can’t.

Citations track how often your site gets named as a source in answer engine responses. More citations improve placement in synthesized answers and tend to generate more favorable descriptions. But don’t focus on citations alone. Broader brand mentions in LLM training data matter too, even when there’s no link attached.

The window for getting ahead of this isn’t unlimited. GEO takes 6 to 12 months to show results because LLM training cycles run on their own timeline. Brands that started the entity cleanup and authority-building work last year are already ahead. Waiting to see where this goes means starting from further behind every quarter.