04 — Page shapes that get cited
Part 4 of 6 · ← Where to publish · Index · Next → Regulated content patterns
LLMs cite some pages and ignore others. The difference is structural, not stylistic. This page is the short list of patterns we've seen win across the observation tool data.
The five shapes that get cited most
1. Comparison tables with named carriers
The most consistent shape. A page that compares 6-12 named carriers in one table — with columns for premium, discounts, AM Best rating, J.D. Power score, available states — gets cited in roughly half of LLM answers to "best X insurance" queries.
Why it works: the LLM treats the table as a sourced shortlist. The comparison-site editorial team has already done the work of choosing which carriers belong on the page; the LLM borrows that choice.
What it looks like in the wild: - NerdWallet "Best home insurance companies" — 10 carriers in a table - Bankrate "Best car insurance" — 8 carriers compared on price and service - The Zebra "Cheapest auto insurance by state" — table per state
What does not get cited at the same rate: prose-only "How to choose home insurance" articles, infographics, video-only content.
2. Geo-segmented "Best X in [city/state]" pages
LLM queries are increasingly geo-specific. "Best home insurance in Seattle" gets a different answer than "best home insurance" — and the comparison sites that have a dedicated Seattle page win the Seattle answer.
Why it works: jurisdiction matters in insurance. State minimums, available carriers, regional weather risk — all change the answer materially. A page that says "for Washington state homeowners, the typical policy looks like X and the top carriers are Y" answers the actual question, not the general one.
What it looks like: - Bankrate's per-state "Best home insurance in Washington" pages - NerdWallet's "Best car insurance in Florida" pages - The carrier's own per-state "Coverage in California" pages
3. Specialty-context guides ("Best X for Y user")
The query shape that's growing fastest. "Best home insurance for a luxury home." "Best car insurance for a young driver." "Best cyber insurance for a fintech startup." Each of these is its own retrieval surface.
Why it works: the LLM resolves the user-context to a specialty editorial page when one exists, and to a generalist comparison page when one doesn't. Where the specialty page exists, the carriers on it are the carriers the LLM names.
This is the lever Phidea's observation tool tested most intensively. Property-type, vehicle-type, specialty use-case (rideshare, SR-22), commercial-vertical (cyber for SaaS) — all behave like their own retrieval segments. The comparison site that publishes "best car insurance for rideshare drivers" wins the rideshare query. The carriers it names win first-recommendation slots.
4. Carrier-own "How our coverage works in [state] for [risk]" pages
This is the owned-domain mirror of pattern 2 + 3. The carrier's own page that says "our condo policy in Washington covers X, costs Y, includes Z endorsements" gets cited as the source of truth for what THIS carrier offers.
Why it works: when the LLM has named a carrier from a comparison site, the next citation is often the carrier's own page explaining what they offer. Two citations: one for the recommendation, one for the verification.
What it looks like: - Chubb's "Masterpiece home insurance" detail pages - State Farm's "Auto insurance in Washington" page - Travelers' "Electric vehicle coverage" page
What does not work the same way: generic carrier marketing pages ("We've protected families since 1922") or product-brochure PDFs.
5. State-DOI-disclosure pages
Underrated. State insurance department websites publish required disclosures, complaint statistics, and licensed-carrier registries. LLMs cite these as the regulatory source of truth.
Why it works: the LLM treats state DOI pages as authoritative for "what's true in this state" — required limits, complaint ratios, licensure status. Carriers rarely link to their own state-DOI pages, but they should.
What it looks like: - California Insurance Department complaint-ratio rankings - Washington Office of the Insurance Commissioner consumer guides - Florida's OIR licensed-carrier registry
These aren't always pretty, but they're sourced and they're cited.
Three shapes that DO NOT get cited at the same rate
Generic thought-leadership
"The future of insurance in 2030." "Why insurtech matters." "Five trends shaping our industry." These get traffic on LinkedIn but rarely get cited in answers to buyer queries.
Why: they don't ladder to a specific buyer query. The LLM has nothing to do with them.
Press releases without substance
"Acme Insurance announces new partnership with Acme Tech." Often gets indexed (see comparison sites that aggregate them) but rarely gets cited in answers — the LLM has nothing to do with the relationship.
Exception: when the press release names a specific operational change ("Carrier X now covers EVs in California with this endorsement"), it CAN get cited as the source of that fact. So shape press releases around concrete coverage changes, not deal mechanics.
Long-form 5,000-word "ultimate guides"
The classic SEO move from 2018-2022. Doesn't get cited as well as a 1,200-word focused page on the same topic. The LLM extracts 200-400 words from a source; long-form pages bury the answer.
What this means for content design
Two operating rules:
1. Every owned-domain page should answer one buyer query. Not five. One. The query shape should be in the H1 or in the URL slug. If your page can't be summarised as "the answer to query X" in one sentence, it's mis-shaped for LLM citation.
2. Comparison-friendly structure beats narrative structure. Tables, named entities (carriers, states, products), explicit numbers (premium ranges, deductibles, J.D. Power scores), and clear sub-headings beat well-written long prose that lacks structural anchors. The LLM extracts the table, the named-entity list, or the heading — not the connective text between them.
Where to start auditing your own pages
Pick your top 10 owned-domain pages by traffic. Run each one through:
- Does the H1 answer a buyer query? If it's "Welcome to Acme Insurance," fail.
- Is there a comparison or named-entity list above the fold? If the first 400 words are prose, fail.
- Is there a state or specialty context tag? If it's pan-state, pan-customer-type, fail.
- Is the page <1,500 words and <2 scrolls deep? Long pages bury the answer.
Pages that fail all four: they're SEO-shaped, not GEO-shaped. Re-cut them.
The strongest GEO test: take your URL, paste it into ChatGPT or Perplexity with "summarize this page," and read what you get back. If the summary is a confused recap of marketing copy, an LLM citing your page will surface the same confusion. If the summary names specific entities, prices, and contexts, the LLM has something to cite.