Problem — what’s changing and why it matters
Search is shifting from sending people to pages toward giving them instant, sourced answers. AI-driven assistants now surface concise, grounded responses that often remove the need to click through — which is siphoning traffic from publishers and brands.
The data, at a glance
– Zero-click results have surged. In experiments, Google’s AI overviews generate zero-click outcomes for as many as 95% of queries where they appear. ChatGPT-style answer chains have shown zero-click rates from about 78% up to 99% in some verticals. – Organic CTR is slipping. Top-position click-through rates have dropped from roughly 28% to about 19% — a ~32% relative decline — with even sharper falls for lower spots. – Traffic losses are real and measurable. Large outlets like Forbes and Daily Mail reported traffic declines near 50% and 44% in affected areas. Smaller players see impacts too: a German price-aggregator (Idealo) reported roughly 2% of clicks diverted to ChatGPT-derived interfaces in testing.
Why this is happening now
Two things converged in 2023–25. Foundation models became more capable, and practical RAG (retrieval-augmented generation) pipelines scaled into production. Platforms began embedding AI overviews directly in their UIs around 2024–25, while major indexing efforts from OpenAI, Anthropic and Google accelerated. The combined effect: engines now prefer concise, citable snippets — answers that can be confidently grounded — rather than pages that simply rank well.
How “citability” actually works (technical overview)
The move toward answer-first search is pipeline-driven, not magical. The core distinction is:
- – Foundation models: generate fluent, general-purpose prose from learned weights, but they can fabricate details when unchecked. – RAG systems: couple generation with live retrieval. Which documents get retrieved, how they’re ranked, and how well the model can match its claims to those documents determine which sources get cited.
Key stages: retrieval → ranking → grounding. Any weak link here reduces a publisher’s chance of being referenced.
Signals that make a source more likely to be cited
RAG and retrieval systems typically prefer sources that are:
– Close in embedding space (dense-vector similarity). – Ranked highly by relevance, freshness and topical authority. – Easy to ground against the generated claims (high grounding confidence). – Machine-readable and standardized: explicit provenance, consistent formatting and structured metadata help a lot. Systems favor pages where answers are extractable, clearly attributed and corroborated across sources.
Concrete levers publishers can pull to increase citability
– Structure answers for retrieval: short, labeled answer sections and machine-readable metadata (JSON-LD). – Optimize for embeddings: use clear, extractable statements and concise anchor language. – Improve grounding signals: include explicit quotations, timestamps and direct links so the engine can match snippets to assertions. When grounding confidence is low, engines either omit citations or default to generic responses. Clear, extractable facts raise the odds a page is shown as evidence.
What this means for motorsport publishers (and similar verticals)
Racing coverage and technical analysis play to data-driven engines if formatted correctly. Priorities should be:
– Publish short, structured summaries and Q&A blocks for common queries. – Mark up key stats and results with schema. – Make primary sources — timing feeds, official releases, team statements — easy to access and clearly attributed. RAG systems will tend to prefer consistent, accessible sources such as official series sites, timing providers and recognized technical specialists.
How major answer engines differ (short comparison)
– ChatGPT-style deployments: typically private RAG stacks, producing condensed citations tuned to the prompt. – Perplexity: shows visible, linkable citations with short source snippets. – Google AI Mode: leans on knowledge-graph signals and often favors established domains. – Claude (Anthropic): may use different retrieval indexes and vary citation verbosity. Crawl intensity also varies wildly — reported crawl ratios span from Google (~18:1) to OpenAI (~1,500:1) to Anthropic (~60,000:1) — helping explain why some engines cite certain sources more frequently than others.
Three operational must-dos
1. Be crawlable: make pages accessible to major indexers and bots. 2. Add structured signals: implement schema, clear source lines and concise extractable text. 3. Don’t assume parity: visibility in one engine doesn’t mean you’ll be cited across the board. Publishers that adapt — by making facts easy to find, verify and machine-read — will be in a much better position to retain visibility and capture whatever click-through remains.