phidea
← GEO build guide
Plain series · page 2 / 3

02 — What the data says wins

Part 2 of 6 · ← What is GEO · Index · Next → Where to publish

This page is the short version of what the observation tool tells us. 11 levers tested across home, auto, and commercial cyber. Numbers come from the actual per-LLM run arrays committed to the Phidea repository.

The five levers that clearly move placement

1. Property-type naming (home insurance)

The single cleanest finding. LLMs have per-property-type editorial winners and the pattern replicates cross-LLM at near-maximum stability.

  • Luxury home → Chubb: 5/5 on Perplexity, 5/5 on Gemini.
  • Historic home → Chubb: 4/5 + 5/5.
  • Condo → Nationwide: 3/5 + 5/5.

Implication: each property type behaves like a niche. A carrier should pick one property-type editorial surface to claim, and defend it. Trying to win all five property types at once dilutes editorial focus.

2. Category editorial depth on a specialty use-case

The most counter-intuitive finding. The editorial winner is not the actual market specialist — it's the generalist with the deepest consumer-facing content.

  • Rideshare insurance → State Farm (3/5 + 5/5). Progressive is the actual TNC-endorsement pioneer; doesn't place first.
  • SR-22 → GEICO (5/5 + 3/5). Dairyland is the largest US SR-22 monoline writer; doesn't place first.

Implication: brand-specialty alone doesn't surface. Specialists must close the editorial-depth gap with comparison-site presence. Generalists can annex specialty surfaces by publishing depth.

3. Emerging-category presence

Open categories are claimable by whoever publishes editorial first. The carrier who got there before insurtechs got their content engine running takes the surface.

  • Electric vehicle insurance → Travelers (4/5 + 5/5). Tesla Insurance, Root, Lemonade do not place first.

Implication: scan for the next emerging category (PHEV, hydrogen, fleet EV, EV charging-station coverage). Whoever publishes a comprehensive editorial guide first wins it for years.

4. Bundled-policy prompt = its own retrieval surface

Bundles do not inherit from single-line winners. Major surprise.

  • Boston bundle → Farmers (5/5 + 5/5). MAPFRE owns Boston home but does not extend to bundle.
  • Los Angeles bundle → AAA (5/5 + 4/5). Mercury owns CA single-line auto, AAA owns the bundle prompt.

Implication: if you sell bundles, content optimized for "best home insurance" does NOT automatically win "best home and auto bundle." Bundle queries route to bundlers (Farmers, AAA, State Farm). Plan a separate editorial investment.

5. Regional monopoly — strictly line-specific

Regional dominance carries, but only for the line where the carrier is dominant.

  • NJM → Newark NJ auto (4/5 + 4/5). Writes ~1 in 4 NJ auto policies.
  • MAPFRE → Boston MA home (5/5 + 3/5). MAPFRE does NOT win Boston MA auto — that's State Farm + Plymouth Rock.

Implication: don't assume your home market share transfers across lines. Audit your top-3 states' citation share by line before assuming carry-over.

Two levers that work but are conditional

6. Specialty-peril coverage — saturation-dependent

Works only when the peril category itself has saturated editorial coverage.

  • High-value jewelry → Chubb (5/5 + 5/5). High-net-worth = Chubb's editorial home turf, lots of consumer guides cite Chubb.
  • Wildfire / flood → generalists by default. Specialists do not surface; comparison sites haven't built specialty-peril editorial depth.

Implication: specialty peril alone is not enough. The category-depth check matters more than peril-specificity.

7. Comparison-site citation share

Not really a "lever" you control directly, but a structural fact.

  • ~58% of Perplexity citation hosts on US insurance queries are consumer-comparison sites: NerdWallet, The Zebra, Bankrate, Policygenius, Insurify.
  • Every winning carrier in observations 1-5 above appears on at least one of these sites for the matching category.

Implication: comparison-site presence is table-stakes. Verify category-by-category coverage. Missing the consumer-comparison surface caps your retrieval ceiling regardless of how good your owned content is.

Two levers that DO NOT move placement

8. HQ-city presence

Only 1 of 10 HQ-city predictions matched in our test.

  • Boston (Liberty Mutual HQ) → MAPFRE on both LLMs.
  • Columbus OH (Nationwide HQ) → Cincinnati and Erie.
  • Bloomington IL (State Farm HQ) → State Farm only on Gemini, not Perplexity.

Implication: drop the HQ-city assumption from any planning. Carrier headquarters does not surface in geo queries — regional volume share or editorial depth does.

9. Being the market specialist alone

  • Hagerty (largest US classic-car insurer) does not win classic-car LLM queries.
  • Dairyland (largest SR-22 monoline writer) does not win SR-22.
  • Progressive (rideshare-endorsement pioneer) does not win rideshare.

Implication: brand-specialty positioning alone does not surface. Specialists can recover the surface only by closing the editorial-coverage gap.

What this means for content strategy

  • Pick one lever per quarter. "Property-type ownership" is one lever. "Bundle-prompt presence" is a different one. Don't try to win all at once.
  • Comparison-site presence is mandatory. If you're not on NerdWallet / Bankrate / The Zebra / Policygenius / Insurify in your category, every owned-domain investment compounds against a low ceiling.
  • The actual data is on `/distribution/llm-observations`. That page is the live measurement surface for everything on this page. Re-read it when you start a new investment to see what changed.
Cross-LLM, cross-week stability is the bar. A finding that holds at 5/5 on one LLM but not another is a single-LLM editorial accident. A finding that holds 4-5/5 on both Perplexity AND Gemini is a real lever. That's the threshold the observation tool uses to assign impact labels.