One topic per page: naming what your page is about

Before an AI answer engine can cite a page, it has to retrieve a passage from it that fits the query. The citation itself is carried by the page's URL and title, not by the passage: OpenAI's web search, for example, attaches the URL, title and location of each cited source. The passage instead determines whether a model lifting it out of your page can tell what it is about. Two habits govern that: naming your subject, and holding one topic.

The evidence behind both habits is real, but it sits upstream of citation. What the cited researchers measured is passage retrieval, document retrieval and QA accuracy after coreference or chunking interventions, not citation counts.

How engines find you, then name you

Retrieval systems generally work below page level: they pull the few blocks of text that best match a query, then generate an answer bound to the sources those passages came from. Google documents a passage ranking system that identifies individual sections of a page; OpenAI and Perplexity do not document their retrieval unit, so treat the passage-level picture as how these systems are commonly built rather than as vendor fact. For your page, two things have to go right. The passage has to surface for the query, and it has to make sense once it is separated from everything around it. Attribution is handled by the source URL and title, so an unnamed passage still gets credited to your page; what suffers is the model's ability to interpret it and write an accurate sentence around it. Google's Cloud Natural Language API scores every entity for salience, "the importance or centrality of that entity to the entire document text". That is a text-analysis product rather than anything Search documents, but it is a useful name for the property. A focused page about a named subject maximizes salience; a page that mentions its subject only in passing does not.

Both properties have been measured, in information retrieval rather than in citation studies. CLAP (CIKM 2025 paper), a framework that segments passages, resolves coreference chains and generates localized pseudo-queries, reported "up to 20.68% absolute nDCG@10 improvement" for that combination rather than for entity naming alone, and observed that "only a portion of a passage is typically relevant to a query, while the rest introduces noise". A separate study of document chunking (NAACL 2025 Findings) worked on documents the authors artificially stitched together from different topics, and found that a chunker splitting on topic boundaries retrieved about 12 F1 points better than a fixed-size one, because it better preserved topic integrity. That is a result about where you cut a document, not a measured penalty for mixing topics on a page.

Hold the frame, though. Both numbers are retrieval results, one step upstream of citation, and the chunking authors hedge their own finding: "These results suggest that in real life, the topics in a document may not be as diverse as in our artificially noisy, stitched data, and hence semantic chunkers may not have an edge over fixed-size chunker there."

Four adjacent topics have their own articles in this series: the knowledge-graph side of naming (schema markup, sameAs), the craft of making each section stand alone, breadth of query coverage, and title length.

Name your subject, and name it consistently

The classic failure is a page that is plainly about a specific thing yet never says its name. An About page, a product page, a service page, or a home page that runs entirely on generic first person: "we help teams ship faster", "our platform automates this", "this tool saves hours". Every sentence assumes you already know who "we" is. A human visitor who arrived through the navigation does know. A passage lifted out of the page by a retrieval system does not. The engine can still cite the page by URL, but a passage built entirely of unexplained "we" and "it" gives a model nothing to work with when it writes the sentence around that citation. Coreference research supports that narrower point; it does not show that attribution becomes impossible.

The fix is mechanical. State the named subject in the main content, near the top, in plain prose, and then keep using the same name. "Acme is a deployment platform for small teams" does in one sentence what a page of unattributed "we" cannot. Consistency matters as much as presence: if the page drifts between "Acme", "Acme Corp", and "the Acme platform" as if they were different things, reconcile the variants at first mention and pick one form for the rest. Consistent naming is a signal engines can act on, and inconsistent naming actively manufactures ambiguity. OpenAI's GPT-5 system card documents a real-time router that picks a model "based on conversation type, complexity, tool needs, and explicit intent", so not every query is answered by the largest model available. A study of ambiguity in retrieval-augmented systems (ACL 2025 paper) found that "smaller language models are likely to benefit more from coreference resolution compared to larger models, indicating that coreferential complexity poses a greater challenge for models with limited capacity." The result suggests explicit referents help smaller models most. Which models a commercial answer engine actually runs on your page is not documented.

One honest gate: never force a name where there is none. A general explainer about a concept legitimately has no brand subject, and it is not deficient for that; its subject is the concept, which its title and body already name. First-person "we" is also fine on any page, so long as the subject is named somewhere salient. The rule targets pages that are about a specific nameable thing and refuse to name it, not every page that says "we". And naming your subject in the prose is only the internal half of disambiguation; connecting that name to an entity graph through schema markup, sameAs links, and consistent profiles is the external half, covered elsewhere in this series.

Keep one topic per page

The second habit is focus. A page that holds one topic retrieves more cleanly than one that sprawls, because every passage in it reinforces the same match instead of diluting it. Keep the evidence in proportion, though. The chunking study above compared two ways of cutting a document, not a coherent page against a mixed one, and its authors caution that the topic-aware chunker may lose its edge on realistic documents. Reading it as support for one topic per page is an inference, not a measured page-editing effect. So the working rule is "no substantial off-topic section", not "one page may only make one point". A page has to be genuinely mixed before this matters.

This is where a common misreading needs heading off: focus is not narrowness. Covering many subtopics of one subject is a good thing. That is breadth, a separate craft in its own right, and a comprehensive guide that works through every facet of a single subject is the ideal on both counts: broad in coverage, single in topic. Nothing about topic focus argues for splitting a thorough guide into fragments. Only genuinely unrelated subject matter dilutes a page: the agency services page that ends in three paragraphs about a company retreat, the product page with an appended essay on industry history that belongs in its own article.

The practice follows directly. Put the page's one subject in the first block, where the reader and the retrieval system meet it first. Let every section serve that subject. When a section starts serving a different subject, that section has found its own page; move it there and link it.

Make the title reflect the body

A page declares its topic before anyone reads it: the <title>, the meta description, and the H1 are the promise a searcher and an engine see first. That promise should match what the body delivers. Google's Search Quality Rater Guidelines treat a misleading title, one that has little to do with the actual content, as a low-quality signal, and Google's helpful content guidance asks publishers to self-assess whether the page title gives a descriptive, helpful summary of the content and avoids exaggerating or shocking. In principle a title that promises what the body does not deliver also wastes the match, though no cited source measures that.

The honest scope here is clear mismatch. A title that promises a comparison and delivers a sales pitch, or names a topic the body never treats, is the failure. A title that is merely broader or narrower than the body is not; the quality guidance targets misleading, not imprecise. And this is the semantic half of the title question, whether the promise matches the content. Whether the title is present, unique, and the right length is a structural concern with its own article in this series.

This is about the page, not the site

One clarification, because most advice blurs it. Everything in this article is intra-page: one page naming its subject and holding one topic. It is not "topical authority" or "site radius", the site-level idea that off-topic pages dilute a whole domain's standing on a subject. That is a different unit of analysis and a different, more contested debate. A single well-named, single-topic page can be retrieved and attributed on its own merits, whatever else the site publishes. Fix the page in front of you; the site-level question is separate.

What this does not do

Said plainly: naming your subject and holding one topic are retrieval and identifiability levers, not a measured citation lift. No controlled experiment isolates either property's effect on citations. The canonical GEO experiment (KDD 2024) tested presentation levers like adding statistics and quotations, but had no entity-clarity or topical-focus method in its design. A 2026 study (Vishwakarma et al.) of which source a model cites anonymized the brands in its data, so it cannot speak to naming at all. The measured numbers in this piece are retrieval results, and one of them was produced on artificially mixed documents with the authors' own hedge attached.

The practical case holds anyway. Naming your subject costs one sentence. Holding one topic costs a moment of editorial discipline. Both help human readers before any machine, and the retrieval research makes a machine benefit plausible. Do the cheap, honest work; just do not let anyone sell it to you with certainty the evidence does not support.

Frequently asked questions

Does naming my subject get me cited more by AI?

It helps identifiability, not a proven citation count. The measured evidence sits upstream: one retrieval framework that resolves coreference among other steps reported up to a 20.68% nDCG gain. Attribution itself runs on the source URL and title, so an unnamed passage is still credited to your page. No controlled study shows that naming your entity lifts citations, so treat it as a low-cost clarity practice, not a proven requirement or a multiplier.

Should every page make only one point?

No. Keep one topic, not one point. Covering many subtopics of a single subject is good breadth, and a comprehensive single-subject guide is the ideal. Only genuinely unrelated topics dilute a page, and the measured retrieval penalty shrinks on realistic, coherent pages. Aim for no substantial off-topic section, not artificial narrowness.

My page is about a concept, not a brand. Do I still need to name a subject?

No. A general explainer legitimately has no brand subject and is not deficient for it; its subject is the concept itself. The naming rule targets pages that are about a specific nameable thing, an About, product, or service page, yet refer to it only as "we" or "our platform". Do not bolt a name onto a topic page.

Does the title really need to match the content?

Yes, for clear cases. A title that misleads, or has little to do with the body, is a documented quality demerit in Google's rater guidelines. The bar is obvious mismatch, not a title that is simply broader or narrower than the page. Write the title as a descriptive summary of what the page delivers.

Is this the same as topical authority?

No. Topical authority is a site-level idea: whether your whole domain covers a subject deeply. This article is intra-page: whether this one page names its subject and holds one topic. A single well-named, single-topic page can be retrieved and attributed on its own, regardless of what the rest of the site covers.