From Ranking to Reference: LLM SEO for AI Overviews and AI Answers

TL;DR

  • AI-powered search shifts SEO from winning clicks to visibility, citability and memorability.
  • What counts is whether your content appears as a reference in AI Overviews and in answers from large language models (LLMs).
  • LLM SEO works when that presence creates measurable demand: brand searches, direct visits, assists.
Privacy

Contents

The state of AI search in 2025: what has changed since spring

Rollout and reach. On March 25, 2025, Google rolled out AI Overviews to more EU countries, including Germany, Austria and Switzerland.

The official list names the language and country combinations of this first European expansion.

For websites in Europe, generative answer surfaces are now standard operation, no longer a test. blog.google Google Help

How often do AI Overviews appear? Reliable analyses of large query samples currently show an AI Overview share of more than 11% of Google searches, up from the launch year.

The share depends on the industry: it is especially high in health, education, B2B tech and insurance, much lower but growing in travel and entertainment.

Older estimates of 40%+ are not representative and systematically overstate visibility. BrightEdge Ken Yeung SE Ranking

Zero-click stays high. For 2024, a zero-click share of 58.5% (US) and 59.7% (EU) was measured. In 2025, the reach of generative modules grows: impressions rise, click-through rates fall.

This does not apply evenly to every industry, but it changes expectations for organic traffic. SparkToro Search Engine Land

New players, real usage. Perplexity handled around 780 million queries in May 2025, a visible block of demand beyond classic search engines, though at a lower overall level.

For marketers, this means: distinguish your audiences. Not everything is Google-only. TechCrunch Search Engine Land

Consequence: in 2025, success shows not only in positions, but in whether and how content appears in answers, and whether people later type your brand actively (brand demand, direct).

An SEO forecast puts numbers on this shift: it estimates per channel which traffic, brand demand and leads are realistic and what they are worth.

From ranking to reference

Top rankings still matter, but they deliver clicks less reliably when generative answers sit above the organic results.

The goal: being named as a reference source, with consistent messages, verifiable evidence and a clear link to the brand.

The three core goals of LLM SEO:

  • Increase AI mentions of the brand and domain,
  • Raise citability (original data, methodology, authorship),
  • Strengthen memorability through brand search and direct visits.

Structuring content so LLMs pick it up

This section turns the short checklist into clear instructions, so a head of marketing can judge effort and impact.

Key statement first: answer the main intent

Open with the answer to the dominant search intent in one or two sentences: what is the result or recommendation? Then add context (for whom, under which conditions) and evidence (figures, sources, limits).

This way, people and LLMs find the essence immediately and get material suited for AI Overviews.

Example (fictional, B2B SaaS):
Key statement: “For mid-market teams, tool X cuts lead time by 18% on average compared with Y.”
Context: “Applies to SDR teams of five or more with at least 10k inbound leads per month.”
Evidence: “Study Q2/2025, n = 84 accounts, 90-day measurement window.”

Compact formats instead of walls of text

LLMs prefer structured units that are easy to extract. Put core answers into FAQ blocks, step-by-step sequences and comparison tables.

The benefit: people scan more easily, and LLMs are more likely to adopt precise wording, the basis for brand mentions.

In practice:

  • FAQ: 6 to 10 typical questions, each with a one- or two-sentence answer plus a short rationale.
  • Steps: no more than 7, each with result, duration and risks.
  • Table: columns for criterion, option A/B/C, trade-off, when it fits.

Clean heading structure (H1/H2/H3)

This means a logical outline with exactly one H1 per page and H2/H3 headings in a clear order.
It improves segmentation and makes individual sub-questions referenceable, which matters for AI Overviews that assemble answers modularly.

Make transparency and authority visible

  • Author box: role, short bio, area of expertise.
  • Methodology box: sample, time frame, data sources, tools.
  • Sources next to the claim: LLMs value evidence close to the statement, and so do people. This setup supports E-E-A-T signals and raises the chance of being cited as a reference.

Using Schema.org correctly: a practical guide for 2025

Why: structured data is the most precise context carrier for machines. It increases the chance of correct citation and supports rich snippets. JSON-LD is the standard; use Microdata or RDFa only if your processes require it.

LLMs primarily “read” the visible content, but JSON-LD still supports entity mapping during crawling.

Priority schema types and why they matter

Core types (almost every site):

  • Organization / LocalBusiness (+ subtype): legal entity, contact, logo, profiles. The basis for consistent entity signals.
  • WebSite + SearchAction: sitelinks search box where relevant.
  • BreadcrumbList: clarifies navigation and strengthens context.
  • Article / BlogPosting: author, headline, date, image, a citable unit.
  • FAQPage, HowTo: only if the content is really structured that way; improves extractability.
  • ImageObject, VideoObject: make media available with clear labels.

Commerce and services:

  • Product / Offer / AggregateRating / Review: clean offer and rating data.
  • Service: service descriptions with clear evidence of performance.

Expertise and people:

  • Person: authors, expert roles, sameAs profiles to support trust.

Vertical examples (selection):

  • MedicalBusiness / Dentist / Physician / MedicalClinic (services, doctors, booking link),
  • Attorney / LegalService (practice areas, admissions),
  • AutoDealer / AutomotiveBusiness / Vehicle (inventory, services),
  • Restaurant / Menu / Reservation,
  • EducationalOrganization / Course,
  • SoftwareApplication / APIReference.

The result is industry-specific precision, not generic markup.

Compact JSON-LD pattern

For blog posts, a “BlogPosting + Organization” snippet is often enough (headline, description, datePublished, author, publisher/logo, image, mainEntityOfPage).

What matters is consistency with the visible content and complete required fields, a common source of errors.

Common structured data mistakes

Wrong types (for example Product without a product), markup that does not match the content, missing required fields, “decorative” schema without substance. Check regularly in the Rich Results Test and update during content refreshes.

Note: for AI answers, verifiability counts. JSON-LD never replaces well-structured content; it amplifies it.

Technical requirements for LLM visibility

Crawlability: AI needs free access.

  • No unnecessary blocks in robots.txt for GPTBot, ClaudeBot, DeepSeekBot and others.
  • Keep XML sitemaps up to date; maintain priorities and change frequencies.
  • Main content without JavaScript hurdles; HTML first where possible.
  • Performance and stability: timeouts and server errors cause crawling gaps.

Tools and feeds. Use Google Search Console and Bing Webmaster Tools actively; provide RSS feeds for content and data products where it makes sense.

A GEO audit checks these points systematically: blocked AI crawlers, outdated sitemaps, JavaScript hurdles and faulty schema markup, ranked by impact.

Video as a citation booster

LLMs draw on explainer videos remarkably often. Properly prepared, they deliver concise knowledge units, ideal for AI Overviews and answer engines.

Formats that work:

  • Explainer videos (3 to 8 minutes) on clearly defined problems,
  • chapter-based demos with timestamps,
  • short clips (30 to 90 seconds) with one strong statement,
  • long-form content and webinars with chapters and content spin-offs.

How videos get cited more easily:

  • A meaningful title and a precise description with the key statement,
  • a transcript on the landing page and on YouTube, plus subtitles (SRT),
  • chapters and timecodes in the description and on the page,
  • a companion article with the same statements and sources,
  • VideoObject with name, description, thumbnailUrl, uploadDate, duration, a reference to the transcript, contentUrl/embedUrl.

Audio and accessibility: clean audio, a clear speaker role, subtitles and, where useful, two audio mixes (full and compact) improve usability and the chance that snippets are quoted verbatim.

Beyond Google: places with a high chance of co-citation

Generative systems aggregate from many sources. If you are present in the right reference places, your chance of co-citation rises:

  • Wikipedia/Wikidata: a neutral entity base with consistent evidence.
  • YouTube: chapters, transcript, metadata (see above).
  • Trade media and industry portals: editorial mentions, study reports, guest articles.
  • Press section: stable URLs, cleanly versioned media kits.
  • Science and data: arXiv, SSRN, Zenodo, open data portals, Kaggle.
  • Developer contexts: GitHub, API docs, package READMEs.
  • Q&A and communities: Stack Exchange; Reddit selectively and fact-based.
  • Associations and standards: white papers, references to standards.

Publishing strategy: consistent core evidence across channels, formats tailored to the audience (video, short text, data package), and a focus on lasting assets over short PR spikes.

A competitive analysis shows which of these sources already cite competitors in AI answers and where the company itself is still missing.

Measuring what works: KPIs in a post-click world

Traditional SEO metrics alone are no longer enough. Add influence metrics that work without a direct click:

  • AI mentions (brand or domain in AI Overviews and answer engines),
  • brand demand (brand searches) and direct visits,
  • GA4 assists and first-touch attribution for early touchpoints,
  • survey attribution (“How did you hear about us?”),
  • earned mentions and social listening (are your figures and models being reused?).

Technical view: LLM referrals. Segment potential LLM traffic by referrer patterns and user agents (for example OpenAI/ChatGPT, Bing/Copilot, Google, Anthropic, DeepSeek, Perplexity, Kagi).

Create your own channel groups in GA4 and Looker Studio so trends become visible. Interpret trends, not daily values.

How does management recognize impact?

  • Correlations: do brand searches, direct visits or newsletter sign-ups rise when a topic gains AI Overview visibility?
  • Attribution shift: first-touch and assists highlight SEO contributions before the last conversion.
  • Qualitative indicators: add CRM fields on “first contact through …” and do not rely on click data alone.

Governance and quality assurance

Process, not activism.

  • Quarterly refresh of key pages (facts, examples, screenshots),
  • maintain E-E-A-T signals (author bio, source transparency, methodology),
  • build co-citations actively (partnerships, guest articles, joint studies),
  • manage volatility: models change, so look at trend lines, not daily values.

Added benefit: content that is “citable” for LLMs often improves user experience and traditional SEO signals at the same time.

Compact LLM SEO checklist (extended)

  • Key statement first: answer the main intent, then context and evidence.
  • Q&A, lists and tables instead of walls of text; no more than 7 steps per process.
  • H1/H2/H3 clear and logical; only one H1 per page.
  • Schema.org (JSON-LD): Organization/LocalBusiness, Article/BlogPosting, FAQPage/HowTo, VideoObject; add matching industry types.
  • Own data and studies with transparent methodology; make authors visible.
  • Video spin-offs with transcript, chapters and VideoObject markup.
  • Core messages also on YouTube, Wikipedia and in trade media.
  • Monitor AI visibility; track brand demand and assists; segment LLM referrals.
  • Quarterly refresh and E-E-A-T maintenance as a fixed routine.
  • Build co-citations systematically.

Figures and trends (August 2025)

  • AI Overviews: more than 11% of Google queries; EU rollout since March 2025; strongly industry-dependent (high in health, education, B2B tech, among others). Trend: growing and diversifying by query type. BrightEdge Ken Yeung blog.google
  • Zero-click: 59.7% (EU) and 58.5% (US) in 2024; in 2025 more impressions and lower CTRs in many clusters. Trend: consistently high, shifting toward brand and direct. SparkToro Search Engine Land
  • Perplexity: 780 million queries (May 2025); about 20% month-over-month growth; new ad formats still experimental. Trend: growing, but a selective channel. TechCrunch Search Engine Land Business Insider

Conclusion: LLM SEO means influence, not just clicks

SEO in 2025 means influence, not clicks alone. If you structure content to be citable, mark it up clearly with Schema.org, provide your own evidence, use video deliberately and measure impact beyond the click, you stay present in generative answers and build lasting demand.

Next step: a focused audit shows where content can quickly become more referenceable and which metrics reflect its real impact. Book a short orientation call to set priorities and the approach.

Get in touch now
Lorenz Wacker, Managing Partner and Digital Strategist at digitalagenten Berlin

Lorenz Wacker

Managing Partner

Lorenz Wacker is Managing Partner and Digital Strategist at digitalagenten in Berlin and specializes in holistic digital strategy.

As managing director, he brings deep expertise and a visionary approach to designing digital multichannel strategies.

Lorenz has extensive industry experience in automotive, financial services, insurance, online media, telecommunications, retail, travel and healthcare.