What passage optimization actually means in 2026

Passage optimization is the discipline of writing pages where any one section can stand alone as the cited answer. AI engines extract passages, not pages, so each H2 must answer one query cleanly. Google's May 2026 guidance calls this still SEO. The tactics that compound are the same ones that compound for traditional ranking: named entities, attributed numbers, parallel structure, fresh dates.

Passage optimization is the discipline of writing pages where any one section can stand alone as the cited answer to a specific query. It is not chunking. It is not paragraph SEO. It is the practice of structuring each section so that a search engine or an AI engine can lift a 40 to 60 word block and serve it as the answer without the surrounding page for context. The term overlaps with generative engine optimization (GEO), answer engine optimization (AEO), and AI SEO, but no consensus definition distinguishes these terms in the academic literature as of early 2026 (see Wikipedia's entry on generative engine optimization for the term's current scope).

How AI engines actually extract passages

AI engines do not read a page top to bottom. They run a query fan-out, retrieve passages from many pages, and assemble a synthesized answer with citations. Google's own documentation describes this as retrieval-augmented generation (RAG), a technique that "rel[ies] on our core Search ranking systems to retrieve relevant, up-to-date web pages from our Search index" and uses those retrieved passages to generate the response (Google for Developers, 2026).

Query fan-out is the second mechanism. When a user asks "how to fix a lawn that's full of weeds", Google's own example shows the model generating concurrent sub-queries: "best herbicides for lawns", "remove weeds without chemicals", "how to prevent weeds in lawn" (Google for Developers, 2026). Each sub-query competes for citations independently. A page that answers only the parent query loses the sub-query citations to pages that answer each one cleanly.

The implication is structural, not semantic. A 3,000 word essay built as one continuous argument is harder to extract than a 1,500 word page built as twelve sections, each with a tight first paragraph that stands alone as the answer to one sub-query.

The Aggarwal study: where the 40 percent number comes from

Research from Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, and Deshpande at Princeton, published in the proceedings of KDD 2024, found that GEO techniques can boost visibility by up to 40 percent in generative engine responses, per the Aggarwal et al. KDD 2024 paper on generative engine optimization. The paper introduced GEO-bench, a benchmark of diverse user queries across multiple domains, and tested a range of optimization strategies against it.

Two findings from the study matter most for working writers. First, the techniques that produced the largest visibility lift were not keyword tactics. They were citing statistics, quoting authoritative sources, and adding factually dense original claims. Second, the efficacy varied substantially by domain, "underscoring the need for domain-specific optimization methods" (Aggarwal et al., 2024). A tactic that lifts visibility for a debate-style query may do nothing for a tutorial query.

The takeaway is that passage optimization is not one move. It is a set of moves that compound when matched to the right query type.

What Google says about passage optimization (and what to ignore)

Google's official guidance, updated May 15, 2026, takes a position that contradicts a lot of the GEO industry advice circulating right now: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO" (Google for Developers, 2026). The same document explicitly disclaims four common tactics that practitioners frequently recommend:

• LLMS.txt files and other "special" markup. Google says: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search."
• "Chunking" content into tiny pieces. Google says: "There's no requirement to break your content into tiny pieces for AI to better understand it. Google systems are able to understand the nuance of multiple topics on a page."
• Rewriting content just for AI systems. Google says: "You don't need to write in a specific way just for generative AI search."
• Special schema.org markup for AI. Google says: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add."

This matters for how passage discipline is framed. Passage optimization is not a new AI-specific tactic. It is a writing discipline. Sections with a clean first-paragraph answer are easier to extract for AI engines AND easier to scan for human readers. The reason it works is the same reason good editorial writing has always worked: a reader, human or model, should be able to lift the substance of a section without lifting the whole article.

The five traits of an extractable passage

Across the GEO research literature and current AI citation tests, five traits separate passages that get cited from passages that do not. Each trait works on both human readers and AI engines because each addresses a common failure mode of weak writing.

1. The answer comes first. The first 40 to 60 words of the section state the answer. The rest of the section earns the depth. Search Engine Land's analysis of AI citation patterns describes this directly: "Put the main answer in the first 1-2 sentences" (Search Engine Land, 2026).
2. The entity is named, not implied. The passage uses the full noun phrase, not "we" or "it" or "this approach". The passage says "Jardine Studio's eight-week SEO engagement combines technical audits, content strategy, and Next.js implementation" instead of "our process." Research on generative AI search engines found that consistent naming across independent sources makes it more likely a model will surface that entity accurately, per Li and Sinnamon's 2024 study of generative AI search engines.
3. The claim has a number or a date. Specific beats general. "Performance improved" is invisible. "Email campaigns with a single focused call to action improved click-through rates by 18 percent in a 2026 test" is extractable. The Aggarwal study found that adding citations, statistics, and quotations to a passage was the single largest visibility lever they tested in the same KDD 2024 paper.
4. The citation is inline and attributed. "Studies show" is hedged. "Semrush analyzed 10 million keywords in 2026 and found AI Overviews appear in approximately 16 percent of queries" is auditable (Semrush, 2026). AI engines surface attributed claims more reliably because the citation pattern signals reliability.
5. The structure is parallel. When a page has six H2 sections that all use the same shape (answer-first paragraph, supporting prose, occasional list or table), the model can extract from any one of them with the same algorithm. Inconsistent structure is harder to extract.

The contrast between weak and strong passages is sharp enough that it shows up at the sentence level. The table below shows the same claim written two ways. The right column is what AI engines extract; the left column is what gets ignored.

The pattern is the same in each row: specific noun, specific number, specific source. That is what extractability looks like at the sentence level.

Passage optimization vs GEO vs AEO vs SEO

The terms overlap. Most working writers do not need to draw bright lines between them, but it helps to see what each one actually emphasizes.

Google's position is that all of these are still SEO. The honest version is that they share the same foundations (good content, technical accessibility, authority signals, structured answers) and differ in which surface they prioritize. Passage optimization is the writing discipline that pays off across all of them.

How to get cited by ChatGPT

ChatGPT cites pages that combine three traits: extractable structure, attributable specificity, and topical authority. The fastest path to ChatGPT citation is to first rank in Google's top 10 for the underlying query, because ChatGPT's retrieval pipeline draws heavily from the same web index Google uses. In testing run on May 16, 2026, Perplexity (which mirrors much of ChatGPT's retrieval pattern) cited 8 of Google's top 10 organic results for the query "what is passage optimization in SEO": the same SERP, the same priority order, with minimal additional sources.

A second live test on the same date confirmed the pattern. For the query "generative engine optimization how to rank", Perplexity cited 9 sources, and the highest-weighted source was the arXiv paper from Aggarwal et al. (cited 3 separate times across the answer). The Aggarwal paper does not rank in the top 5 of the Google SERP for that query, but it is cited everywhere across the GEO topic cluster, which lifts it in AI retrieval. A third test for "how to get cited by ChatGPT" itself returned 9 sources, with Search Engine Land's November 2025 piece cited twice and a niche AI-search specialist (Alhena) cited twice. The pattern is consistent: a small number of authoritative sources get cited repeatedly, and the citation set overlaps heavily with Google's top 10 organic.

To raise the probability of being cited:

1. Rank in Google's top 10 for the underlying query. This is the gate. Pages outside the top 10 are rarely retrieved.
2. Lead each H2 with a 40 to 60 word standalone answer. This is the unit AI engines extract.
3. Cite primary research and name the source inline. "Aggarwal et al. (2024) found a 40 percent visibility lift" is citable. "Studies show GEO works" is not.
4. Keep the page fresh. Semrush's AI Visibility Index, which tracked 2,500 prompts across Google AI Mode and ChatGPT, found "between 40 and 60 percent of cited sources change from month to month" (Search Engine Land, 2026). Recently updated pages compound an advantage.
5. Build presence beyond your own site. AI engines weigh mentions across Reddit, YouTube, LinkedIn, and industry publications. The Aggarwal paper itself was cited 124 times by mid-2025 because it was discussed everywhere, not because of one publisher's authority.

How to get cited by Perplexity

Perplexity's citation pipeline overlaps with ChatGPT's but tilts further toward freshness and structural clarity. Perplexity's answers consistently follow a recognizable shape: BLUF opener, "what helps" bulleted list, numbered workflow, "what does not work" list, practical example, honest caveat. Pages that match that shape get cited more often, because the model can extract directly.

The five-checkpoint model from ZipTie's analysis describes Perplexity's citation pipeline as a sequence: semantic relevance to the query, document accessibility (crawlable, indexed), extractability of self-contained statements, topical authority signals, and freshness. A page that passes the first three but fails on authority will be near-cited and excluded; a page that passes all five becomes one of the 5 to 10 sources Perplexity attributes in the response panel.

In practice, the four moves that matter most for Perplexity:

1. Write in self-contained statements. Each sentence in the answer-first paragraph should make sense if pulled into another document.
2. Use bolded action verbs in the opener. "GEO is about making your content easy for AI search tools to understand, trust, and quote" is the pattern Perplexity itself uses in its own answers.
3. Include a "what does not work" section. Negation lists are extracted at higher rates than affirmation lists in current Perplexity output.
4. Date the page and update it. Perplexity surfaces last-updated dates in its citation chips when present.

What it looks like in practice: a Jardine case

When Jardine Studio built the Black Salt Room website, every page was structured around the passage discipline: each H2 opened with a 40 to 60 word answer to the implicit local-service query (e.g., "What is the cost of a breathwork session in Hamilton?"). The site shipped with Lighthouse 100 scores across Performance, Accessibility, Best Practices, and SEO on first run, and now ranks in the top 5 for several Hamilton-area breathwork queries within 90 days of launch. The structural discipline did the work, not a content volume push. Passage discipline is the writing-side anchor of the studio's broader SEO and growth work.

What to measure

Two metric families matter, and most teams only track the first.

Traditional SEO metrics still apply: rankings for target keywords, organic clicks, impressions, click-through rate. These are visible in Google Search Console and remain the foundation for AI search visibility (per Google's AI optimization guide for Search, AI features draw from the same Search index).

AI visibility metrics are newer and noisier:

• Citation rate. How often your URL or brand appears as a cited source in ChatGPT, Perplexity, Claude, or Google AI Overview answers for your target queries.
• Share of voice in AI answers. Your citation count relative to competitor citation counts on the same query set.
• Citation churn. How often you appear month-over-month. The Semrush AI Visibility Index tracking 2,500 prompts found 40 to 60 percent of cited sources change month to month, per the Semrush AI Visibility Index summary on Search Engine Land. Profound's analysis of 240 million ChatGPT citations found similar volatility, with 70 to 90 percent of cited domains completely different after six months.
• Referral traffic from AI engines. Visible in analytics under the chat.openai.com, perplexity.ai, and gemini.google.com referrer domains.

The volatility numbers matter because they set realistic expectations. A page that gets cited in month one may not be cited in month three. Monthly tracking is the minimum cadence; weekly is better if the topic is fast-moving.

What to do this month

A five-step plan that compounds:

1. Audit one cornerstone page. Pick the most important page on your site. Check whether every H2 opens with a 40 to 60 word standalone answer. If not, rewrite the opening paragraphs.
2. Add inline citations to two factual claims per section. Numbers without attribution read like opinion. Numbers with attribution read like research.
3. Build a comparison table somewhere on the page. Tables get extracted by AI engines at high rates. One table per long-form article is the rule of thumb.
4. Add a visible last-updated date. Both in metadata and in the rendered page. Update it when you actually update the content.
5. Submit the URL to Google Search Console for reindexing. Then check the AI engines (ChatGPT with search, Perplexity, Google AI Mode) two weeks later to see whether the page surfaces.

If none of this is producing results in 60 days, the issue is rarely the writing. It is usually either the underlying Google ranking (passage optimization compounds on existing rank, it does not create it) or the topical authority on the domain. When the work needs to run as a structured engagement instead of a self-serve audit, the studio offers it as a dedicated AI search service: passage optimization, visible FAQ content, entity consistency, and citation monitoring on a defined prompt set.

FAQ