AI Citation Optimization: How LLMs Select Brands to Mention

Classic SEO optimizes for a ranking algorithm that sorts ten links. AI citation optimization optimizes for a language model that writes an answer and names only a few sources in it. That is a fundamentally different game with its own rules. This article explains how LLMs such as ChatGPT, Perplexity, and Google AI Overviews select brands and sources, why unlinked brand mentions are becoming the most important currency in the process, which on-page and off-page levers demonstrably work, and how a B2B company in the DACH region builds and measures its own AI visibility.

Why AI citation optimization matters for B2B in the DACH region

Search is measurably shifting away from the classic list of links toward the generated answer. According to a representative Bitkom survey, half of Germans (50 percent) already use a chat with an AI instead of classic internet search at least sometimes, 7 percent mostly, and 5 percent even exclusively. The survey covers Germany only, but the trend is considered indicative for the DACH region. If you don't appear in these AI answers, you lose exactly the visibility that used to come from organic rankings.

What matters is that AI systems choose different sources than Google. An international analysis of 15,000 long-tail prompts shows that only 12 percent of the links cited by ChatGPT, Gemini, and Copilot appear in Google's top 10 for the same prompt and that around 80 percent of these citations don't rank in Google at all for the original query. A good Google position is therefore no guarantee of an AI citation. The playing field is new, and it partly rewards different signals.

For the Austrian B2B market, this means: if you bet on AI citability early, you secure visibility in a channel that is not yet as fiercely contested as the classic SERP. That is why the mechanics behind it are worth understanding in detail.

How LLMs select sources: retrieval, grounding, and query fan-out

On a technical level, an AI citation comes about differently from a Google ranking. Modern AI answers combine two sources of knowledge. First, parametric knowledge, which is stored in the model itself and comes from the training data. Second, live retrieval, in which the system fetches web pages in real time and bases its answer on them. This second step is called grounding: the answer is anchored in the retrieved sources and backed by citations.

The key difference from Google ranking lies in query fan-out. Instead of answering a single search query, the system breaks the original query down into several sub-questions and variants and retrieves pages that match this fan. The Ahrefs overlap study describes this mechanism as the reason why the AI retrieves pages based on several variants of the query and not just the exact original question. A source that delivers solid content across many related sub-questions is therefore more likely to be cited than one that only ranks strongly for a single keyword.

This leads to the central strategic insight. Being number one for a keyword is not enough. A brand has to be present on the web as an entity, broadly, consistently, and in connection with its core topics, so that it is recognized as a relevant source across the entire query fan-out. This is exactly where entity SEO comes in.

Brand signals and entity SEO as the foundation

For visibility in Google AI Overviews, brand mentions are clearly ahead of backlinks. An international Ahrefs analysis shows that branded web mentions, at 0.664, have the strongest correlation of all branding signals examined, while the number of backlinks reaches only 0.218. That is a ratio of roughly 3 to 1 in favor of mentions. The effect is even clearer at the top end: in the underlying study of 75,000 brands, brands in the top quartile for web mentions average 169 mentions in Google AI Overviews, more than ten times as many as brands in the second-highest quartile with 14.

The reason lies in how the models work. LLMs learn entities and their relationships from the frequency and context in which brands, topics, and terms appear together. This co-occurrence is the language an LLM understands. You build a consistent entity profile from several building blocks:

• Consistent brand naming: Company name, spelling, address, and core service must be identical across all sources. Contradictory information dilutes the entity in the Knowledge Graph.
• Wikidata and Wikipedia: Structured knowledge bases are a main source of parametric knowledge. A clean Wikidata entity and, where the notability criteria are met, a German-language Wikipedia entry anchor the brand in the Knowledge Graph.
• Co-occurrence with topics: The brand should regularly appear in the same context as its core topics. If you are an Austrian B2B provider and want to be cited for a field, you need to be visibly connected with that topic there.
• Structured data: Organization and sameAs markup clearly connects the website with the brand's profiles and helps systems resolve the entity correctly.

The importance of structured knowledge sources shows in both training and citation. In GPT-3's training mix, English-language Wikipedia had a weight of 3 percent and was processed about 3.4 times, more often than any other data source. At the same time, Wikipedia was the most cited single source in ChatGPT's answers, as a Profound analysis of 680 million citations from August 2024 to June 2025 shows. If you exist there as a clean entity, you sit closer to the source of parametric knowledge.

Unlinked brand mentions as AI currency

In classic SEO, the linked reference is what counts most. For LLMs, that holds only to a limited extent. A language model processes text. It recognizes a brand and its context even when the mention is not linked. This makes the unlinked brand mention a currency of AI visibility in its own right. That is exactly why branded web mentions clearly beat backlinks in the correlation data.

In practice, this shifts the focus from pure link building toward digital PR and earned media. The link is no longer the only goal. What counts is a context-rich mention in credible sources. There are several ways to get there:

• Digital PR: Media relations that lead to editorial coverage generate brand mentions in trusted media, often with topical context.
• Data-driven content: Your own studies, surveys, or industry figures get cited by third parties and repeatedly connect the brand with its subject area.
• Trade media and industry portals: Articles, interviews, and mentions in relevant DACH trade media place the brand in the right topical environment.
• Unlinked mention monitoring: Existing but unlinked mentions can be tracked down and maintained, for example by standardizing the spelling and sharpening the context.

The strategic consequence: a PR placement that used to count only because of the link now has additional value as an AI signal, even without a link. That fundamentally changes how earned media is valued.

On-page GEO tactics: building citable content

Off-page determines whether a brand is recognized as a relevant entity. On-page determines whether a specific page is citable as a source. The scientific basis for this comes from the KDD 2024 study on generative engine optimization. It shows that GEO can increase visibility in generative answers by up to 40 percent. The effective tactics aim to prepare content so that a language model can easily extract and cite it:

• Statistics with sources: Concrete, sourced figures are highly citable because they make an answer more precise. Every figure needs a clear source attribution.
• Expert quotes: Attributed statements from named experts increase credibility and give the model citable building blocks.
• Source citations and evidence: Transparent references to primary sources signal trustworthiness, a core criterion in source selection.
• FAQ and structured answers: Clearly worded question-and-answer blocks match exactly the format in which AI systems assemble answers.
• Schema.org markup: Structured data such as FAQPage, Article, or Organization makes the meaning of content machine-readable.
• Citable content blocks: A direct answer to a specific question, concise and complete under a fitting subheading, increases the chance of being quoted verbatim.

The common denominator of these tactics is extractability. Content that answers a question clearly and with evidence is the preferred source for featured snippets and AI answers alike.

Platform specifics: ChatGPT, Perplexity, and Google AI Overviews

AI citation optimization is not a uniform discipline. The major platforms select sources differently, and that has direct consequences for your tactics.

ChatGPT relies heavily on established, encyclopedic knowledge. According to a Profound analysis of 680 million citations (August 2024 to June 2025), Wikipedia was ChatGPT's most cited source, with 7.8 percent of all citations and 47.9 percent of citations among the ten most cited domains. The Digital Bloom's international 2025 AI Citation & LLM Visibility Report also shows that 87 percent of SearchGPT citations match the top 10 Bing results. If you want to be visible here, you need a clean entity in Wikipedia and Wikidata plus a solid Bing presence.

Perplexity relies more on real-time retrieval and user-generated content. In the same Profound analysis, Reddit led on Perplexity with 6.6 percent of all citations and 46.7 percent among the ten most cited domains. Perplexity also incorporates classic rankings more strongly: according to the overlap study, nearly one in three Perplexity citations points to pages that rank in the top 10. Here, fresh content and community presence work together with good Google rankings.

Google AI Overviews increasingly pull their citations from a wider field than just the top results. According to Ahrefs data analyzed by Search Engine Journal, the share of AI Overview citations from top-ranking pages fell from 76 percent in July 2025 to 38 percent, based on 863,000 keywords and 4 million AI Overview URLs. The remaining citations are spread across positions 11 to 100 with 31.2 percent and beyond position 100 with 31.0 percent. So Google, too, rewards topical breadth beyond a narrow ranking focus.

A single number shows how little the platforms overlap: according to the AI Citation Report, only 11 percent of domains are cited by both ChatGPT AND Perplexity. A source that is strong on one platform often isn't on another. Platform-specific strategies are therefore mandatory, not optional.

DACH and Austria focus: local authority sources as a citation base

For the Austrian market, the key question is which German-language sources count as authorities. Since Wikipedia was ChatGPT's most cited source in the Profound analysis, German-language Wikipedia is a central anchor point. Beyond that, the DACH region has specific authority sources that serve as a citation base:

• WKO profile and Firmen A-Z: A complete, consistent profile in the company directory of the Austrian Economic Chambers (WKO) provides a reliable, structured entity record for the Austrian market.
• German-language trade media and industry portals: Mentions in relevant trade publications connect the brand in the right topical and linguistic context.
• Local directories and chambers: Industry-specific registers and chamber profiles support the consistency of brand data.
• German-language Wikipedia and Wikidata: If the notability criteria are met, an entry in German-language Wikipedia with a linked Wikidata entity is the most stable anchor in the Knowledge Graph.

The advantage of a DACH focus is less competition. While international English-language sources are heavily contested, consistent German-language entity signals can be built in a more targeted way.

GDPR-compliant reputation and review management

Brand and AI signals also come from reviews, PR, and external data sources. In Austria and Germany, these activities are subject to the GDPR, in Switzerland to the revised Federal Act on Data Protection (revFADP). Both require legally sound handling. Reviews and reputation data are valuable co-occurrence signals, but they must be collected and used in compliance with data protection law:

• Collect reviews in a legally sound way: Customer reviews should rest on a clear legal basis, be collected transparently, and not be manipulated. Fake reviews are legally risky and undermine the trust signal.
• Mind personal data: For PR and case studies, get consent from the people named and use personal data sparingly.
• Check data sources: External data sources for brand monitoring must be GDPR-compliant. The legal basis for processing should be documented.
• Maintain transparency: A complete legal notice (Impressum), a clear privacy policy, and consistent company information also strengthen trustworthiness as an entity.

Legal certainty and AI visibility don't conflict. On the contrary: transparent, consistent, and verifiable information is exactly the kind of trust marker that the GDPR requires and that AI systems prefer when selecting sources.

From classic backlink to AI signal: reallocating the budget

With AI search and the insights from the Google API leak, the weighting of off-page measures is shifting. The plain backlink doesn't lose its effect, but it is no longer the dominant signal for AI visibility. The correlation data is clear: branded web mentions correlate with AI Overview visibility at 0.664, backlinks at only 0.218. That suggests reallocating the budget.

In practice, this means a shift across three priorities. First, less isolated link building that targets only the link. Second, more digital PR and earned media that generate context-rich brand mentions, regardless of the link. Third, dedicated GEO measures on your own site that make content citable. The backlink stays part of the mix, but it now shares its budget with disciplines that feed the AI signal more directly.

A sober assessment matters here: classic SEO and AI optimization are not mutually exclusive. A strong entity, sourced content, and trust signals work in both worlds. The budget question is a matter of weighting in favor of signals that scale in AI search, not an either-or decision.

Practical workflow for B2B: from entity to measurement

A realistic build-up follows a clear sequence and takes time. Initial results typically show after three to six months, because entity signals and mentions first have to flow into the models and into retrieval.

• Entity setup (month 1): Standardize brand data, check and build the Wikidata entity, make the WKO profile and directories consistent, implement Organization and sameAs schema.
• On-page GEO (months 1 to 2): Make core pages citable with sourced statistics, attributed quotes, FAQ blocks, and clean schema markup.
• Digital PR and earned media (months 2 to 4): Place data-driven content and PR deliberately to generate brand mentions in DACH trade media, with or without a link.
• Set up monitoring (from month 1, ongoing): Define prompt sets, track AI visibility and citation share, watch the sentiment of mentions.
• Iterate (from month 3): Adjust per platform based on the measurement data, for example by strengthening Wikipedia presence for ChatGPT or fresh content for Perplexity.

This workflow is not a one-off project. Because the platforms keep changing, as the drop in Google top 10 citations from 76 to 38 percent shows, adjustment belongs in a regular routine.

Common mistakes in AI citation optimization

• Relying only on backlinks: If you keep putting your budget primarily into links, you ignore the stronger signal. Brand mentions correlate at 0.664 versus 0.218 for backlinks.
• One strategy for all platforms: ChatGPT, Perplexity, and Google select differently. Since only 11 percent of domains are cited by both ChatGPT and Perplexity, a one-size-fits-all approach fails.
• Trusting a good Google ranking: A top position is not enough. Around 80 percent of AI citations don't rank in Google at all for the original query.
• Not making content citable: Without sourced statistics, clear answers, and structure, extractability is missing. GEO can increase visibility by up to 40 percent if the tactics are applied consistently.
• Keeping the entity inconsistent: Contradictory brand data dilutes the profile in the Knowledge Graph and weakens the co-occurrence signals.
• Ignoring GDPR risks: Manipulated reviews or opaque data sources are legally risky and undermine the trust signal that AI systems evaluate.

Metrics and measuring AI visibility

AI visibility is measurable, but it requires different metrics than classic SEO. Four metrics are central. First, citation share: how often the brand is mentioned or linked in the answers to a defined prompt set. Second, AI share of voice, which measures your share of mentions compared with competitors. Third, sentiment: the tone in which the brand is mentioned. Fourth, the development of branded web mentions as a leading indicator.

The methodological basis is a well-defined prompt set: a fixed list of the questions that potential customers in a B2B context would actually ask. This set is queried regularly across the relevant platforms, and the citation share is evaluated. Tools such as Ahrefs Brand Radar or Semrush's AI features support tracking brand mentions and AI visibility over time.

Platform context is crucial for interpretation. Because the systems' source selection shifts dynamically, as the drop in Google top 10 citations from 76 to 38 percent shows, a single measurement says less than the trend over several months. The leading lever remains the number of brand mentions, because brands in the top mention quartile are mentioned in Google AI Overviews more than ten times as often as brands in the second-highest quartile.

Next steps

AI citation optimization is the logical extension of off-page SEO into a world where language models make the selection. The foundation remains a strong, consistent entity that is anchored on the web through brand mentions, structured knowledge sources, and topical co-occurrence. Citable content and platform-specific tactics build on that. Sensible next steps for a B2B company in the DACH region are an audit of your own entity signals, defining a prompt set for monitoring, and gradually reallocating the off-page budget from isolated link building toward digital PR, earned media, and GEO. If you interlock these disciplines, you build visibility in a channel that increasingly complements classic search.