AI changes SEO by adding a new citation layer on top of the ranking layer, by shifting query traffic from blue-link results to AI-generated answers, and by changing the economics of content production so that depth, originality, and entity-clarity matter more than volume. The crawl-index-rank pipeline still exists; what’s different is that an additional pass now reads the indexed content, decides which pages are credible enough to cite in a generated answer, and surfaces those citations in surfaces like Google AI Overviews, ChatGPT, Perplexity, Bing Copilot, and Claude.<\/p>\n
For a working SEO, the change is not philosophical. It is mechanical. Indexing now optimises for entity recognition as much as keyword matching. Ranking now sits next to citation as a parallel objective, not a downstream consequence. Surfaces have multiplied beyond Google’s SERP. The economics of thin, aggregator-style content have collapsed because LLMs synthesise answers from the substantive sources and ignore the rest.<\/p>\n
This article walks through what’s actually different, layer by layer, with the operational implications. It is the current-state explainer rather than a prediction piece – what has changed in indexing, ranking, surfaces, content economics, and measurement, as of 2026.<\/p>\n<\/p>\n
The indexing layer has been the quietest change but the most foundational one. Search engines and the LLM crawlers that feed answer engines now build an index that is structured around entities and their attributes, not only around terms and the documents that contain them.<\/p>\n
Entity-first indexing.<\/strong> A page about a product, person, place, or concept is parsed not as a bag of words but as a structured entity record – the entity itself, its type, its attributes, its relationships to other entities, and its provenance. Schema markup (JSON-LD) feeds this directly. The entity record is what the answer-generation layer queries when it builds a response.<\/p>\n Structured data has gone from optional to expected.<\/strong> Articles, products, organisations, FAQ blocks, breadcrumbs, and reviews ship with schema as default. Pages without schema can still be indexed, but they enter the index as less structured records and lose the citation lottery to pages that gave the parser an explicit, machine-readable definition.<\/p>\n Multi-modal indexing.<\/strong> Images, video, and audio are increasingly parsed for content rather than treated as opaque assets with alt-text labels. Transcripts, diagrams, and screenshots all contribute to what the index knows about a page.<\/p>\n Operational implication.<\/strong> The on-page work is no longer just title-meta-H1 and keyword density. It includes schema completeness, entity disambiguation (clear definitions of what the page is about), and clean information architecture so the parser can resolve ambiguity.<\/p>\n<\/p>\n Traditional ranking – the position your page holds in the blue-link results – has not gone away. It is still the largest single source of organic traffic for most sites. What has changed is that ranking is no longer the only objective. Citation in AI-generated answers is a parallel objective with different mechanics.<\/p>\n The two objectives can diverge.<\/strong> A page ranked at position 3 may not appear in the AI Overview for the same query. A page ranked at position 12 may be the AI Overview’s primary citation. The two systems read different signals – ranking weights backlinks, on-page relevance, and engagement, while citation weights entity-clarity, source credibility for the specific claim, and synthesis-readiness of the prose.<\/p>\n Synthesis-readiness is the new on-page property.<\/strong> LLMs build answers by extracting and synthesising statements from sources. Pages that present claims clearly (one claim per sentence, attributable, with surrounding context) get synthesised more reliably than pages that bury claims in marketing prose.<\/p>\n Source credibility for the specific claim.<\/strong> The citing model evaluates whether the source is credible for the specific question being asked, not in general. A government website is credible for statistics; a practitioner blog is credible for methodology; a news outlet is credible for events. The same page may be cited for some queries and ignored for others.<\/p>\n Operational implication.<\/strong> SEO work splits into two tracks. The ranking track continues with on-page optimisation, technical health, and link building. The citation track adds entity-first content design, statement-clarity editing, and surface-specific monitoring.<\/p>\n<\/p>\n The single SERP has fragmented into multiple answer surfaces, each with its own citation behaviour and traffic dynamics. Working SEO now requires understanding what each surface does and how it sources its citations.<\/p>\n Google AI Overviews.<\/strong> The expandable AI-generated summary at the top of Google search results for many queries. Sources are typically pulled from the top organic results, but not always – the AI Overview model has its own selection logic that sometimes promotes lower-ranked but more synthesis-ready sources. Citation here drives both visibility and click-through to the cited page.<\/p>\n ChatGPT (with browsing).<\/strong> When ChatGPT searches the web for a query, it reads the top results and synthesises an answer with inline citations. The selection draws heavily from authoritative domains and well-structured content. Citation here drives traffic from a user base that increasingly uses ChatGPT as a research starting point instead of Google.<\/p>\n Perplexity.<\/strong> Citation-first by design. Every claim in a Perplexity answer is footnoted with the source. The selection is broader than Google’s AI Overview and rewards clearly-written practitioner content. Perplexity drives a smaller but more research-intent audience.<\/p>\n Claude (with web search).<\/strong> Web-augmented Claude searches and cites sources. Citation behaviour skews toward thorough, well-structured content with clear claims and source attribution.<\/p>\n Bing Copilot.<\/strong> Microsoft’s answer surface across Bing search and Copilot. Citation patterns favour authoritative sources but also include practitioner content for niche queries.<\/p>\n Operational implication.<\/strong> Citation tracking has to be multi-surface. A page can be cited on Perplexity but absent from Google AI Overviews, or vice versa. Single-surface tracking misses 60-80% of the picture.<\/p>\n<\/p>\n The economics of content production have changed in a measurable way. Thin, aggregator-style content has collapsed in citation share because the LLMs synthesising answers prefer original, substantive sources and skip the aggregators that historically captured rank with thin pages on high-volume keywords.<\/p>\n Aggregator collapse.<\/strong> Pages that summarised what other sources said, without adding original observation or data, were reasonably effective at ranking when ranking was the only game. They collapse in citation because the LLMs would rather cite the original source than the summariser.<\/p>\n Original observation gets cited.<\/strong> Pages that contain practitioner observation (what we saw when we did this), original data (numbers from work the author actually did), or distinct framing (a way of explaining the topic that is genuinely the author’s) get cited more reliably than pages that don’t.<\/p>\n Depth gets cited.<\/strong> Pages that cover the topic substantively – 1,500-3,000 words of genuine substance, not padded length – get cited more than thin pages that ranked on freshness or keyword density alone.<\/p>\n Entity-clarity gets cited.<\/strong> Pages that define their entities clearly (a clear opening definition of what the topic is, a clear taxonomy of subtopics, schema markup that resolves ambiguity) get cited more than pages where the parser has to guess what the page is about.<\/p>\n The penalty for thin content has shifted from ranking to citation.<\/strong> Thin content can still rank for low-competition keywords. It cannot get cited.<\/p>\n Operational implication.<\/strong> Content production budgets shift toward fewer, deeper articles with original observation and complete entity coverage, rather than higher volumes of thin pages.<\/p>\n<\/p>\n Measurement has expanded from rank tracking on a single SERP to citation tracking across multiple answer surfaces, plus AI Overview presence and zero-click impact analysis. Teams that haven’t expanded their measurement stack are missing the half of the picture that AI surfaces now control.<\/p>\n Citation tracking.<\/strong> For each priority query, check whether each major answer surface (Google AI Overview, ChatGPT, Perplexity, Claude, Bing Copilot) cites the page. Track citation presence over time, just as rank is tracked over time. This is its own discipline – the multi-LLM citation tester is now a category of tooling.<\/p>\n AI Overview presence.<\/strong> For each priority query on Google, check whether an AI Overview appears, whether the page is cited in it, and what fraction of the answer’s substance comes from the page. Track this alongside organic rank.<\/p>\n Zero-click impact.<\/strong> AI Overviews and other answer surfaces sometimes resolve the user’s question entirely, with no click to the source. The traffic impact varies by query – informational queries see large zero-click impacts, while commercial and transactional queries still drive clicks. Audit the query mix to understand exposure.<\/p>\n Branded query lift.<\/strong> AI Overview citation often produces branded query lift – users who saw the brand in an AI Overview later search for the brand directly. Track branded query volume as a downstream indicator of AI surface visibility.<\/p>\n Operational implication.<\/strong> The reporting cadence and dashboard need rebuilding. Rank is one column; AI Overview presence is another; multi-LLM citation share is another; branded query trend is another. Single-column dashboards undersell the work that is happening on the new surfaces.<\/p>\n<\/p>\n AI changes SEO mechanically, not philosophically. Indexing now favours entity-recognition and structured data. Ranking sits next to citation as a parallel objective with different signals. Surfaces have multiplied beyond Google’s SERP to include AI Overviews, ChatGPT, Perplexity, Claude, and Bing Copilot. Content economics have shifted toward depth, original observation, and entity-clarity. Measurement has expanded from single-column rank tracking to multi-surface citation tracking, AI Overview presence, and zero-click impact analysis. The crawl-index-rank pipeline still exists. What’s been added is a citation layer that reads the indexed content, evaluates it for synthesis-readiness and source credibility, and surfaces it in AI-generated answers. Working SEO in 2026 means doing both tracks – the ranking work that has always existed, and the citation work that is now its own discipline. Sites that hold both win. Sites that hold one and ignore the other lose share on whichever surface they neglected.<\/p>\n<\/p>\n If you want a structured view of where your site stands on both ranking and AI citation, we can scope an audit that covers both layers and produces a remediation plan.<\/p>\n<\/p>\nRanking: still real, now sitting next to citation<\/h2>\n
Surfaces: AI Overviews, ChatGPT, Perplexity, Claude, Bing Copilot<\/h2>\n
Content economics: depth, originality, and entity-clarity matter more<\/h2>\n
Measurement: rank tracking is no longer enough<\/h2>\n
Conclusion<\/h2>\n
Frequently Asked Questions<\/h2>\n
Is SEO dead because of AI?<\/summary>\n
How do I get my page cited in AI Overviews?<\/summary>\n
Do backlinks still matter for SEO?<\/summary>\n
What’s the difference between ranking and AI citation?<\/summary>\n
Should I optimise for ChatGPT and Perplexity, not just Google?<\/summary>\n
Has keyword research changed because of AI?<\/summary>\n
Does schema markup matter more now?<\/summary>\n