A new study from Semrush and analyst Kevin Indig has found that nearly two-thirds of AI citations across ChatGPT, Gemini, Google AI Overviews, and Google AI Mode fail to include brand mentions — a finding that forces marketers to rethink how they pursue visibility in an AI-driven search landscape. The research, which analysed 3,981 domain appearances, lands alongside fresh guidance on optimising for the so-called agentic web, where AI agents increasingly decide which brands get surfaced and which get ignored.

Key Takeaways

• 62% of AI citations are “ghost citations” — they link to a source but never name the brand behind it.
• The agentic web demands a new five-layer optimisation stack so AI agents can find, trust, and select brands.
• SEO writing in 2026 must serve both traditional search engines and AI-powered tools simultaneously.
• Google Analytics 4 remains the baseline measurement layer, but marketers need to connect it to broader AI visibility tracking.
• Improving E-E-A-T signals and targeting low-competition keywords are now dual-purpose tactics for organic search and AI Overviews.

Most AI Citations Strip Out Brand Names Entirely

Semrush partnered with Kevin Indig to study how four major AI engines handle citations. They tracked 3,981 domain appearances across ChatGPT, Gemini, Google AI Overviews, and Google AI Mode. The headline number: 62% of citations linked to a domain without ever mentioning the brand by name in the AI-generated response.

These so-called “ghost citations” mean a brand’s content feeds the AI answer, but the user never sees the brand. The study found that each AI engine treats citations differently. Some favour domain authority. Others lean on topical relevance. None consistently credit the source brand in the visible text. For marketers, the takeaway is blunt: ranking well enough to be cited is no longer enough. You need explicit brand mentions baked into your content strategy.

The Agentic Web Requires a Five-Layer Optimisation Stack

Semrush has published a framework for what it calls agentic web optimisation — the practice of making brands discoverable not just by humans, but by autonomous AI agents that browse, evaluate, and recommend on behalf of users. The guide outlines five layers:

• Structured data and machine-readable content
• Authority signals that AI agents can verify
• Brand entity clarity across the open web
• API and feed accessibility for agent consumption
• Trust and safety markers that pass automated checks

Each layer builds on the last. Miss one, and an AI agent may skip your brand entirely. The framework signals a shift: optimisation is no longer just about Google’s crawler. It’s about every automated system that influences purchase decisions.

SEO Writing Now Serves Two Masters: Search Engines and AI Tools

Updated SEO writing guidance for 2026 makes one thing clear — content must perform in both Google’s traditional results and in AI-generated answers. The 12-tip framework from Semrush stresses clarity, entity-rich language, and direct answers to user questions. Writers who bury key information behind vague introductions risk being ignored by AI summarisation models that scan for concise, authoritative statements early in the text.

Short sentences. Clear structure. Named entities. These are no longer just best practices — they’re table stakes for dual-channel visibility.

GA4 Stays Central, But Needs Broader Context

Google Analytics 4 remains the default measurement tool for most marketing teams. Semrush’s beginner guide walks through setup, linking Search Console and Google Ads, and turning reports into decisions. But in an AI-first environment, GA4 alone can’t tell you whether your brand was cited — or ghost-cited — by an AI engine. Marketers now need to layer AI visibility monitoring on top of standard web analytics.

E-E-A-T and Low-Competition Keywords Pull Double Duty

Improving E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) and finding low-competition keywords have long been core SEO tactics. Now they serve a second purpose. AI Overviews and AI Mode tend to surface content from domains with strong trust signals. Targeting less competitive queries gives smaller brands a realistic path into AI-generated answers, not just page-one rankings. The strategy is straightforward: build genuine authority, then aim where the competition is thinnest.

The ghost citations study puts a hard number on a problem many marketers have only suspected. Being cited by AI is not the same as being seen by users. As AI agents take on more of the discovery and recommendation process, brands that fail to optimise for machine readability, entity clarity, and explicit brand mentions will find their content feeding answers they never get credit for. The playbook is changing fast, and the data says most brands haven’t caught up yet.

Frequently Asked Questions

What are ghost citations in AI search results?

Ghost citations occur when an AI engine links to a source domain but never mentions the brand name in the visible answer. A Semrush study found 62% of AI citations across ChatGPT, Gemini, and Google’s AI features fall into this category.

How do marketers optimise for the agentic web?

Marketers need to follow a five-layer stack covering structured data, authority signals, brand entity clarity, API accessibility, and trust markers. This ensures AI agents can find, verify, and recommend a brand autonomously.

Why does SEO writing need to change for AI tools in 2026?

AI summarisation models scan for clear, entity-rich, and concise statements — especially early in the text. Content that buries key information or lacks direct answers risks being overlooked by both traditional search and AI-generated responses.

What is E-E-A-T and how does it affect AI Overviews?

E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trustworthiness — signals Google uses to evaluate content quality. AI Overviews and AI Mode tend to favour content from domains with strong E-E-A-T, making it a key factor for AI visibility.