In short: Getting cited by ChatGPT, Perplexity and Gemini in 2026 doesn't depend on keywords or bought backlinks, but on three levers: earned media mentions on authoritative news outlets and online newspapers that build the model's primary bias, an extractable structure (a 40-60 word TL;DR plus plain-text questions and answers) and a technically blazing-fast, clean site for AI crawlers. GEO replaces ranking with citability.
- 84% of AI citations come from earned media (journalism, third-party mentions), versus 0.3% from paid/advertorial content — analysis of 25+ million links cited by ChatGPT, Claude and Gemini (Muck Rack, May 2026)
- +40% visibility in generative answers by applying GEO tactics (Princeton/Georgia Tech/Allen AI, KDD 2024)
- -34.5% organic CTR on queries with AI Overview active, across 300,000 keywords analyzed (Ahrefs, 2025)
What is GEO (Generative Engine Optimization) and why does it matter in 2026?
GEO (Generative Engine Optimization) is the discipline that optimizes content to be cited as a source in answers generated by ChatGPT, Perplexity, Gemini and Claude. It doesn't replace SEO: it sits alongside it. SEO drives clicks from the blue link, GEO drives authority and mentions inside a conversational answer that often generates no click but builds brand preference.
The term comes from an academic paper authored by researchers at Princeton, Georgia Tech and the Allen Institute for AI (KDD 2024), which introduced the GEO-BENCH benchmark across 10,000 real queries. According to the study, applying GEO techniques increases visibility in generative answers by up to 40%, with effectiveness varying by domain: it's especially the sites starting from lower SERP positions that benefit most.
The market context is pushing in the same direction. Ahrefs (2025), comparing Search Console between March 2024 and March 2025 across 300,000 keywords, measured a 34.5% drop in organic CTR on queries with AI Overview active. Not optimizing for AI in 2026 means losing traffic twice: on the SERP and inside the chat.
How do ChatGPT, Perplexity and Gemini select their sources? The "primary bias"
Every LLM has two levels of source selection. Understanding them is the first step to getting cited. Level 1 is the primary bias: what the model "already knows" from its training data and the sources it considers authoritative. Level 2 is RAG (Retrieval-Augmented Generation): what the model retrieves from the web in real time to answer a specific query.
Primary bias dominates on generic, high-competition queries. When a user asks "what's the best SEO agency in Italy?", the model draws from the set of brands it finds cited most frequently in trustworthy sources: news articles, online newspapers, Wikipedia, directories and industry analyses. RAG dominates instead on long-tail, time-sensitive or product-specific queries, where the model opens the browser and reads live pages.
The operational consequence is clear-cut. To enter the primary bias you need earned media mentions and authentic PR on recognized outlets: it's journalistic coverage, not the bought backlink, that builds the model's trust over time. To enter RAG it takes weeks, provided the content is written in an extractable way and the site is technically clean for AI crawlers (GPTBot, PerplexityBot, Google-Extended).
One figure that changes the planning: the study by Yext (2025), based on an analysis of 6.8 million citations, found a very limited overlap between the sources cited by ChatGPT, Gemini and Perplexity for the same query. Optimizing for a single model therefore leaves out a significant share of generative visibility.
Why media mentions are worth more than backlinks
It's the most important shift of 2026, and it overturns ten years of SEO. The signal that leads an AI to cite a brand is no longer the backlink, but the mention on authoritative editorial sources. The analysis by Muck Rack (May 2026), conducted across more than 25 million links cited by ChatGPT, Claude and Gemini in 17 industries, is unequivocal: 84% of all AI citations come from earned media, while paid and advertorial content account for just 0.3%. Journalism alone makes up 27% of cited sources, and the figure has held steady between 82% and 89% across every edition of the study since July 2025.
Two practical consequences. First: the freshness of the news matters. An article's impact on AI answers peaks in the first seven days after publication, so press coverage well distributed over time is worth more than an isolated spike. Second: authoritative media that cite your brand — even without a clickable link — still build authority. AI crawlers recognize unlinked mentions — the brand name recurring alongside relevant editorial topics and contexts, the so-called co-occurrence — and count them as genuine citations.
The backlink, in this scenario, drops to a secondary signal. According to 2026 analyses, third-party mentions are roughly 3 times more correlated with AI visibility than traditional backlinks; the correlation between Domain Authority (the domain authority score used by SEO tools) and AI citations falls to around r=0.18 — a value close to zero, meaning an almost non-existent link — and nearly half of AI Overview citations come from pages that don't even appear among Google's top five positions. Gartner (2026) also recommends shifting budget from paid advertising toward answer engine optimization — optimization for the engines that provide answers, i.e. the AIs — pointing to earned media as the asset the AIs trust most.
In operational terms: the number-one lever for entering the primary bias is a press office that generates authentic coverage on online newspapers, magazines and vertical outlets. It's exactly the work we do with our press office and events service: direct relationships with the leading Italian newsrooms (Corriere, Repubblica, Il Sole 24 Ore, ANSA, RAI), press releases built on data and sources, and zero paid articles — only real journalism, the only kind the AIs reward.
Which content gets cited most often? What the Princeton and Ahrefs research says
The Princeton/KDD 2024 paper isolated nine GEO tactics, measuring their impact on the citation rate. The top three for effectiveness are not about keywords, but about evidence-based credibility.
- Cite Sources: include links to authoritative sources within the text, not just at the bottom (among the paper's top-performing tactics)
- Statistics Addition: quantitative data with an explicit source in the same sentence
- Quotation Addition: direct quotes attributed to experts or analysts
- Fluency Optimization: short sentences, short paragraphs, clean grammar
- Technical Terms: industry terminology used with precision
The paper documents that a combination of these tactics (quotes, statistics and citations in particular) produces the highest uplift, with the aggregate boost reaching up to 40% on the base source's citation rate. The study by Ahrefs (2025), on a dataset of 300,000 keywords (150k with AI Overview vs 150k without), then highlighted the strong impact AI Overviews have on organic traffic, cutting the average CTR by 34.5% on queries where they appear. The combined reading is that authority is built with earned media and third-party mentions (E-E-A-T), but it must be paired with a page structure designed for LLM extraction.
The operational summary is simple: the winner is whoever gets cited by the right media and writes with data, sources and structure — not whoever stuffs keywords or piles up backlinks. For the press coverage side you can start with our press office; for the technical side, with integrated SEO and GEO consulting.
The 7 operational rules for getting cited by AI
From academic research, earned media studies and the guidelines published by the leading publisher programs (OpenAI, Perplexity, Anthropic), a pragmatic checklist emerges. Seven rules, in order of expected impact.
- Build earned media mentions. Coverage on authoritative news outlets and online newspapers that cite your brand: it's the primary driver of primary bias (84% of AI citations, Muck Rack 2026). An active press office beats any link building campaign.
- Extractable 40-60 word TL;DR. A self-contained paragraph, up top, that answers the question in the title. It must stand on its own if copied out of context. It's the piece ChatGPT and Perplexity extract most readily.
- Quantitative data with the source in the same sentence. Never "studies show". Always "according to Ahrefs (2025), 34.5% of organic CTR...". The "Statistics Addition + Cite Sources" tactic from the Princeton paper is the most effective one measured.
- Plain-text questions and answers. A "Frequently Asked Questions" section with H3/P pairs, 40-80 words per answer, always visible in the code (never in a JS accordion). Not for Google's rich result — now deprecated — but because it mirrors the form of a conversational answer and is the block most easily extracted by LLMs.
- Headings as questions. H2s and H3s phrased as natural queries intercept both voice search and the conversational prompt.
- Blazing-fast site and clean markup. Semantic HTML, Schema.org/JSON-LD of type Organization and Article, sub-second load times. AIs favor fast, well-structured sites: it's the technical foundation that makes content extractable in RAG.
- Freshness signal. A populated
dateModifiedfield, content refreshed every 3-6 months on high-volatility topics.
The site as a technical foundation: speed and extractable markup
Earned media brings the brand into the primary bias, but when the model opens the browser (the RAG level) it needs to find a page it can read and cite in a few milliseconds. Three things matter here: load speed (sub-second), clean semantic HTML with no content hidden behind JavaScript, and structured data Schema.org/JSON-LD consistent on every page. A slow site or one cluttered with plugins gets crawled worse by AI crawlers and loses citations.
That's why we built Deep CMS: sites with PageSpeed 97-100, distributed on a global CDN, with Schema.org, JSON-LD, sitemap and semantic HTML generated automatically for every page and article — natively GEO-ready. The extractable structure this article talks about (TL;DR, plain-text Q&A, data with sources, dateModified) is built into the system, not manual work to redo every time.
How to monitor visibility on ChatGPT, Perplexity and Gemini
Tracking generative visibility is the most underestimated problem of 2026. Google Search Console won't tell you whether ChatGPT cites your brand in a private chat. Traditional rank trackers measure SERP positions, not mentions inside a conversational answer. Without dedicated metrics, every GEO investment runs on gut feeling.
There are two complementary approaches. Manual monitoring involves a set of prompts defined in advance (brand, competitors, the industry's informational queries) run weekly on each model, with answer logs. It doesn't scale well but is useful for validating hypotheses. Automated monitoring uses tools that send prompts via API to ChatGPT, Perplexity, Gemini and Claude, analyze the answers with NLP and extract three key metrics.
- Brand Mentions: how many times the brand name appears in the answer text
- AI Citations: how many times the brand's domain is listed as an explicit source
- Share of AI Voice: the percentage of brand mentions relative to competitors across a set of strategic queries
Tools like Profound, Conductor AI Visibility and Otterly.ai automate the process, tracking Brand Mentions, AI Citations and Share of AI Voice across multiple models in parallel. The choice of tool should be aligned with the mix of LLMs relevant to your industry and your budget: entry-level services start with manual prompt monitoring, while enterprise platforms include pre/post editorial-change attribution.
If you'd rather have a single platform instead of stacking multiple tools, DM Intelligence by Deep Marketing integrates GEO visibility (mentions and citations on ChatGPT, Gemini and Perplexity), Google Search Console data and site analytics into one dashboard. Integration is the point: cross-referencing SEO positions, real queries and AI citations in the same tool produces insights that GEO-only tools can't deliver, and it's today the most complete way to measure the return of a GEO strategy.
What does NOT work: common GEO mistakes in 2026
Cross-referencing AI crawler guidelines, earned media studies and the anti-patterns known in the literature, here are six mistakes that burn budget without moving citability.
A point that often gets missed: OpenAI's documentation distinguishes between GPTBot (the training crawler), OAI-SearchBot (the SearchGPT index) and ChatGPT-User (real-time fetch on user request). Blocking only GPTBot without understanding what it means cuts off access to future training, but not to SearchGPT. You need a granular policy, not a blanket refusal.
Schema and structured data: what changed with the deprecation of FAQs
A necessary correction, because a lot of GEO consulting has fallen behind. Google has deprecated FAQ rich results: according to the official announcement, as of May 7, 2026 the expandable FAQ panels no longer appear in search results, in June 2026 the report and support in the Rich Results Test are removed, and in August 2026 support in the Search Console API is dropped. It's the final step after the August 2023 restriction, which already limited FAQ rich results to government and health sites only.
Be careful not to confuse two things. FAQ rich results (the visual block in the SERP) are dead. The FAQPage schema, on the other hand, remains a valid Schema.org type: it causes no problems if present, it simply no longer produces a visible result on Google. The practical consequence for GEO is that the questions-and-answers section should be kept for LLM extraction, not for a rich snippet that no longer exists. The schemas that truly matter for AI in 2026 are two. Organization describes the company and, through the sameAs field, links it to its official profiles and external mentions, so the model understands that you're referring to exactly that brand. Article, with the author (who wrote it) and dateModified (when it was updated) fields, tells the AI who signs the content and how recent it is. Together they help models connect content, brand identity and authority.
In short: how you win at GEO in 2026
Lining up the evidence, GEO boils down to three moves that reinforce each other. First: get named by the media. AIs mostly cite whoever is mentioned by authoritative outlets and journalists: that's what builds the brand's reputation inside the model. It's the slowest lever but the most durable, and it's activated with a press office.
Second: write in an extractable way. A summary up top, plain-text questions and answers, every data point accompanied by its source. These are the blocks ChatGPT and Perplexity copy most readily when they read the page in real time. It's the fastest lever: it brings citations within a few weeks.
Third: give AIs a site that's easy to read. Fast, with clean code and structured data. If the site is slow or messy, AI crawlers scan it worse and cite it less, even when the content is excellent.
None of the three is a one-off revolution: they're activities that go into an editorial calendar and a communication plan. The brand that holds the three fronts together — media, content and tech — is the one the AIs end up recommending.
Frequently Asked Questions
How are sources cited by ChatGPT?
ChatGPT with browsing active (SearchGPT) lists sources as numbered links next to the sentences, preferring media and third-party editorial sources (~48.7% of citations according to Yext, 2025). Without browsing, the model draws on its training data, where earned media mentions dominate: 84% of AI citations come from journalism and third-party sources (Muck Rack, 2026). To increase the likelihood of being cited you need authentic press coverage, not just content on your own site.
Do backlinks still matter for getting cited by AI?
Very little. The 2026 analyses show a low correlation (r≈0.18) between Domain Authority and AI citations, and paid content accounts for just 0.3%. What matters far more is earned media: mentions on authoritative news outlets and online newspapers, even without a clickable link. A press office that generates real journalistic coverage is today the most effective lever for entering the models' primary bias.
Does GEO replace SEO?
No. GEO and SEO are complementary disciplines. SEO optimizes for the blue links of Google and Bing, measuring positions and clicks. GEO optimizes for answers generated by LLMs, measuring mentions and citations. In 2026 they should be managed together: earned media, correct schema, a fast site and authority remain the prerequisite for AIs to find and evaluate the content.
Is the FAQ schema still useful after Google's deprecation?
Google's FAQ rich results are deprecated (as of May 7, 2026 they no longer appear in the SERP), but the FAQPage schema remains a valid Schema.org type and causes no problems. For GEO it's the structure that counts, not the rich result: a plain-text questions-and-answers section (H3/P pairs, never in a JS accordion) remains among the formats most easily extracted by LLMs, because it mirrors the form of a conversational answer. Keep it for the AIs, just don't expect a visual block on Google anymore.
Which tool should I use for tracking AI visibility?
Among the specialized tools are Profound, Conductor AI Visibility and Otterly.ai, which track citations across multiple models. If instead you want a single platform that combines AI visibility, Google Search Console data and site analytics, DM Intelligence by Deep Marketing cross-references everything in one dashboard — useful for understanding which content actually drives citations. For small brands, manual monitoring with 20-30 weekly prompts on ChatGPT and Perplexity is enough for the early stages, before moving to a dedicated tool.
How long does it take to see GEO results?
It depends on the level. On RAG (live search) well-structured content with a TL;DR and plain-text Q&A can be cited within 2-6 weeks of publication. On the primary bias (training data and authoritative sources) it takes months of accumulating earned media mentions. The two levels work together: RAG generates immediate visibility, earned media builds long-term dominance.
Want your brand to be cited by ChatGPT, Perplexity and Gemini?
GEO today plays out on two fronts. The first is earned media: getting the authoritative outlets that feed the models' primary bias to talk about your brand. Our press office and events generates authentic journalistic coverage — Corriere, Repubblica, Il Sole 24 Ore, ANSA, RAI — with no paid articles. The second is the technical foundation: a blazing-fast, semantic, GEO-ready site like Deep CMS, which AIs find easy to read and cite. Request a free GEO audit and let's build the strategy together, calibrated to your industry and the models that really matter to cover.
Sources and References
- Muck Rack - What Is AI Reading? Earned media drives 84% of AI citations (May 2026, 25M+ links)
- Aggarwal et al. (Princeton, Georgia Tech, Allen AI) - GEO: Generative Engine Optimization (KDD 2024)
- Ahrefs - AI Overviews Reduce Clicks by 34.5%: Study on 300K Keywords (2025)
- Yext - AI Visibility: How Gemini, ChatGPT, Perplexity Cite Brands (2025, 6.8M citations)
- Search Engine Journal - Google Drops FAQ Rich Results From Search (2026)
- OpenAI - Overview of OpenAI Crawlers (GPTBot, OAI-SearchBot, ChatGPT-User)
- Google - FAQ (FAQPage) Structured Data Documentation
- Google - Generative AI in Search (AI Overviews)
- Deep Marketing - Press Office & Events service (earned media on authoritative outlets)
- Deep CMS - Blazing-fast, GEO-ready website
- DM Intelligence - Analytics, SEO and GEO platform
