Topical Depth vs Breadth: The LLM Tradeoff
For traditional search, breadth pays. A property covering 1,000 query intents at acceptable depth ranks more often than a property covering 100 at excellent depth. For LLM retrieval, the inverse holds for most categories. A property with deep, primary-source coverage of 50 to 150 entities gets cited more often than a property with shallow coverage of 1,000. The two systems read content differently, weigh evidence differently, and reward structure differently. This piece sets out why the depth-versus-breadth tradeoff is real, what depth looks like operationally, and how to decide which side to invest in for a given category in 2026.
Why the Two Systems Diverge
Google’s classic ranking system distributes opportunity across millions of queries. A query receives ten organic links, the engine ranks pages with partial topical overlap, and a property that covers many adjacent queries with reasonable depth captures a long tail of traffic. The unit of competition is the query.
LLM retrieval works at the level of a passage that grounds a generated answer. When a user asks a question, the retrieval layer pulls a handful of candidate documents, the language model drafts an answer, then citations are attached to the spans that match retrieved passages. The unit of competition is the cite-able passage. A property with 100 deeply researched pages, each containing several quotable spans, generates many more citation opportunities than a property with 1,000 shallow pages that each contain one templated answer.
Two structural mechanics explain the divergence. First, retrieval chunking favours longer, evidence-dense passages because they survive the embedding-based similarity scoring better than thin generic statements. Second, the LLM’s trust prior weights primary sources higher than aggregator content, and primary sources tend to be deep on a narrower set of entities. The result is a category where the long tail strategy that built large publishers in 2018 produces diminishing returns on LLM citation surface in 2026.
Position: Depth Is the Default, Breadth Is the Exception
Across audits in BFSI, fintech, industrial materials, and healthcare, the brands gaining LLM citation share are running depth-first strategies. The properties losing citation share to lighter-footprint competitors are those that built breadth assuming the rules from a decade ago still applied.
For the BFSI lender audit, the property’s 25,216 sitemap URLs and 94,100 ranking keywords represented a breadth-first investment. The result on AI surfaces was an 8 percent ChatGPT mention rate, 15.6 percent on Google AI Overview, and 19 percent on AI Mode. Despite being a category-defining property by every classical metric, the brand was being out-cited by smaller competitors with deeper coverage of narrower topic clusters. The recommendation that came out of the 35-section audit was not to publish more. It was to deepen the 980 live 200-OK pages with primary data, named-author commentary, and structural-data-validated entity attribution, while clearing the 24,236 sitemap-waste URLs that were diluting the property-level quality signal.
For the multi-LOB wealth platform RFP, the same pattern surfaced at a larger scale. We classified 11,920 keywords across 25 AI batches and shipped 10 page-level content recommendations totalling 27,818 lines of JSON. Six of the recommendations were ANALYZE-existing rewrites and four were CREATE-new pages, which represents a roughly 60:40 split in favour of deepening existing pages over adding new ones. The breadth was already there. The depth was the missing input.
The exception is greenfield categories. For the loan-aggregator DSA we shipped from scaffold to launch in one day, breadth came first. Six verticals with six bespoke landing surfaces, 16 metro programmatic pages, and a 22-section IA spanning ~622 v1 pages. Until a category presence exists at all, depth on a single page does not produce citations because the property is not yet a candidate source. Breadth establishes candidacy. Depth wins citation.
The Depth Anatomy
Six Depth Markers in a Cite-Worthy Page
| Marker | What it looks like | Why retrieval rewards it |
|---|---|---|
| D1 Primary data | Numbers the property generated or licensed exclusively | No higher-trust source to defer to |
| D2 Named author | Real practitioner with verifiable credentials | Lifts source prior, satisfies E-E-A-T |
| D3 Methodology disclosure | How the numbers were produced, what the sample size was | Allows downstream writers to cite responsibly |
| D4 Counter-examples | Cases where the thesis does not hold | Signals expertise; thin content rarely carries them |
| D5 Entity attribution | Brands, people, places named with sameAs anchors | Resolvable entities feed citation pipelines |
| D6 Structured-data block | Article, FAQPage, HowTo schema valid and matching content | Reduces parsing ambiguity for retrieval layers |
Three of six is a strong page. Five of six is a category-defining one. Audit your top 20 pages for depth markers before commissioning new breadth.
When Breadth Still Wins
Breadth strategies retain their advantage in three situations. First, programmatic SEO for high-volume head terms where each page genuinely serves a distinct query and the underlying data is structurally different per URL. A coworking marketplace covering 12,000 to 18,000 metro-area URLs is not running shallow content. It is running structurally deep content at programmatic scale. Each URL contains location-specific availability, pricing, and amenity data that does not exist elsewhere.
Second, local pack and map results, where the engine surfaces the nearest matching entity rather than the deepest. A healthcare chain we audited surfaced 14 local-pack appearances across 30 priority Chennai queries by maintaining structurally complete location pages with primary address, doctor names, and procedure inventory. Depth on each page was high, and the breadth of locations was a force multiplier.
Third, AI Overview citation patterns that pull from multiple sources per answer. Where the answer is consolidated from many citations, a property covering more of the relevant subset wins more individual citation slots than a property with one deep page on each topic. The math depends on the category’s citation density.
The decision rule we use: if the category leader is winning on cite-share with deep pages, deepen. If the category leader is winning on coverage with structurally distinct programmatic pages, broaden. Pulling the answer from category data takes one weekend with the right retrieval-layer test panel.
The Operational Tradeoff
The throughput math favours depth more than most content teams assume. A deep page (D1 through D6, 1,800 to 2,400 words, primary data block, named author, methodology disclosure) takes roughly five times the effort of a shallow page. A property running one deep page per week instead of five shallow ones publishes 20 percent of the content volume. The cite-share data we have logged shows that the deep approach generates roughly three to five times the citation surface per piece on LLM surfaces, and two to three times the dwell time on Google. The break-even is around 60 percent of comparable volume, and most categories sit comfortably below that break-even.
For the steel exporter we audited, the property’s 648 pages were carrying 0.51 percent Share of Voice with 6,470 ranking keywords. Seventy-seven percent of organic traffic flowed through approximately 20 pages. The fifteenth-lowest performer was contributing 0.04 percent of traffic. The recommendation was to retire roughly 200 pages, deepen the top 80, and add 15 new pages in clear coverage gaps. Net page count dropped from 648 to around 540. Investment shifted toward depth on the survivors.
The full method is in our content strategy service and connects to the foundation work in technical SEO. For category-specific depth-versus-breadth calls, the manufacturing growth engineering writeup documents the retire-and-deepen approach in detail.
Practitioner Takeaway
- Score your top 20 pages against D1 through D6. Most properties find three of six on their best pages and one of six on their average page. The gap is the depth deficit.
- Identify retire candidates before commissioning new pages. Pages with no clicks in 180 days, no inbound links, and no depth markers are draining property-level quality. Clear them with 410 responses before adding new breadth.
- Commission one primary-data asset per quarter. The asset that other publishers cite, that Wikipedia editors quote, that gives the property an unmatched source position on a narrow topic.
- Name and credential the authors. Two named editors with topic-specific credentials outweigh ten generic Editorial Team bylines on cite-share over 12 months.
- Test depth-versus-breadth assumptions quarterly. Run the 20 priority queries against ChatGPT, Claude, AI Overview, AI Mode, and Perplexity. Read the citation list. The leader’s depth profile and coverage profile are visible from the citations alone.
Frequently Asked Questions
How many entities should a depth-first property cover?
Depends on category. For BFSI India, 80 to 150 entities (products, regulators, branches, named experts) is a defensible coverage scope where each entity gets a deep canonical page. For local healthcare, 20 to 40 procedures plus location entities is typical. For programmatic categories like real estate or coworking, the entity count is structurally larger because each location is an entity.
Does breadth still help with branded search?
Yes. Breadth on branded query terms (variants, sub-product names, internal navigation) helps with site-search-style queries and protects the SERP from competitor encroachment. The depth-versus-breadth tradeoff applies to non-branded category queries, not to branded coverage.
How long does a depth-first investment take to show in citations?
Perplexity reflects depth changes within weeks because of its freshness preference. AI Overview and AI Mode update on Google’s indexing cadence, typically four to eight weeks for material changes on established URLs. ChatGPT and Claude lag further, often two to four months, because their retrieval indices refresh less frequently for non-news content.
Can a small property out-cite a category leader?
Yes, in narrow topics with high depth differential. The healthcare chain we audited out-ranked competitors four to thirty-three times its domain index on Chennai kidney and urology queries because location and procedure pages carried D1, D2, D3, and D5 markers that competitor pages did not.
Where does AI-generated content fit in a depth-first strategy?
AI generation accelerates the writing stage but not the depth stage. Primary data, methodology disclosure, named-author commentary, and counter-examples cannot be generated. They have to be sourced and authored by people inside the function. AI assists with draft assembly and structural variance, not with depth.
Want a precise read on whether your category rewards depth, breadth, or a hybrid, and where your property currently sits on the depth markers? Request the audit that scores your top 50 commercial pages against the D1 to D6 anatomy.