Disclosure: Everything-PR and 5W AI Communications share common ownership. Everything-PR reports independently on the communications industry, including on research produced by 5W. Editorial decisions are made by Everything-PR's editorial team.
"The buyer no longer starts at Google. They start in ChatGPT and Claude. The brands cited there own the category. The rest get skipped."
The EPR GEO Scorecard measures Citation Share — the share of answers a brand owns when its category gets asked — across the five AI engines that have replaced the search-results page as the primary buyer research surface: ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. Each volume scores three leading entities in one vertical across 50 controlled buyer prompts × 5 engines = 750 audits. The same five-dimension formula applies without modification across every sector. The methodology is the durable asset. The grades are this quarter's snapshot.
Why the Scorecard exists
Consumer and B2B research behavior has moved into the chatbox. Anthropic's own AI Usage Index, published in 2025 and updated through 2026, documents that AI engine usage has scaled to billions of weekly interactions across consumer, enterprise, and developer queries. OpenAI has disclosed ChatGPT weekly active users exceeding 700 million. Perplexity reports double-digit-million weekly users. Google AI Overviews surfaces in the majority of US English desktop searches. The behavior shift is structural, not cyclical.
The implication for brands is severe. Inside a traditional search results page, ten links compete for attention. Inside a chatbox answer, one to three brands are named. Everyone else is invisible. The brands cited in those answers own the consideration set. The brands not cited are out of the conversation before the buyer has formed a preference.
The EPR GEO Scorecard grades that visibility. Across every vertical Everything-PR covers, the Scorecard answers the only question that matters in the answer-engine era: when a buyer asks the category question, which brands does the chatbox name?
The work belongs to a broader discipline — Generative Engine Optimization (GEO) — which is the structured practice of becoming the answer inside AI engines. The Scorecard measures outcomes; GEO produces them.
What this series does not measure
This is not a ranking of PR firms. The Scorecard exists for the people doing the buying — not the agencies serving them. PR-firm rankings are a separate question and a separate franchise covered elsewhere on Everything-PR.
The Scorecard measures the categories where consumers, executives, journalists, and procurement teams ask the chatbox the question: which beauty brand, which hotel chain, which bank, which streaming service, which fast-food brand, which airline, which crypto exchange, which university. The grades are assigned to the brands competing inside those categories — not the firms representing them.
Every Scorecard volume scores three entities across the same locked formula. The weights and methodology are fixed across the series and reproducible quarter-over-quarter. Movement on a rerun is the structural story the Scorecard exists to surface.
| Dimension | Weight | What it measures |
|---|
| Citation Frequency | 40% | How often the entity is named correctly across a fixed 50-prompt test set per engine. The dominant driver of every Scorecard outcome. |
| Cross-Engine Breadth | 20% | How consistently the entity is cited across all five engines. Penalizes entities that win on one engine but lose on the rest. |
| Query-Type Breadth | 20% | Coverage across five query buckets: recommendation, comparison, capability, reputation, and safety. Penalizes one-trick-pony citation. |
| Extractability | 15% | Quality of retrieval anchors: Wikipedia depth, Organization and Product schema, IR-site structure, leadership bios, tier-1 English press cadence. |
| Crawl Access | 5% | Robots.txt and llms.txt posture, sitemap depth, allowed bots (GPTBot, ClaudeBot, PerplexityBot, Google-Extended). Lowest weight because publisher graphs partially compensate. |
Grading bands
| Band | Score | Meaning |
|---|
| A | 80–100 | Dominant citation presence. Engine returns the brand unprompted across most query types. Owns the category answer. |
| B | 70–79 | Strong presence with category gaps. Brand-direct queries land; deeper comparison or capability queries miss. |
| C | 60–69 | Functional citation, identity inconsistent. Often mis-framed or attributed to parent. Vulnerable position. |
| D | 50–59 | Partial visibility. Engine knows the brand exists but cannot describe it well. Category queries miss the brand. |
| F | Below 50 | Effectively invisible. Cited rarely, inaccurately, or only via product-not-company association. |
How a Scorecard is constructed
Each volume follows the same protocol. Three entities are selected from the vertical based on market position, revenue, or category centrality. The selection rationale is disclosed inside each volume.
- Prompt set construction. Fifty buyer prompts per entity drawn from real-world category research patterns. The set covers all five query types and is balanced across the entity cohort.
- Engine test runs. Each prompt is run on ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. Test window is declared on each volume.
- Response scoring. Each of the 750 responses per volume is scored for citation accuracy, recency, and framing. Scoring is conducted by trained analysts with spot-check redundancy.
- Dimension calculation. Raw responses feed into the five-dimension weighted formula. Final composite is mapped to a letter band.
- Diagnostic write-up. Each volume includes per-engine breakdowns, prompt-level evidence, company-by-company diagnosis, and named structural levers that would move the score.
Methodology appendix — the 50-prompt structure
The fifty-prompt test set per entity per volume is balanced across the five query buckets at ten prompts each. The buckets and representative templates are disclosed below. Vertical-specific prompts are listed in each volume's prompt-level evidence section.
| Query bucket | Count | Representative templates |
|---|
| Recommendation | 10 | "Best X for Y," "What should I use for Z," "Top brands in [category]," "Most-recommended [product type]" |
| Comparison | 10 | "X vs Y," "X vs Y vs Z," "Difference between X and Y," "Which is better, X or Y, for [use case]" |
| Capability / Specification | 10 | "[Category]'s largest by revenue," "Best [product] for [specific use]," "Which brand has [specific feature]" |
| Reputation | 10 | "Controversies involving X," "[Brand] criticism," "Trust ratings for [category]," "Most ethical [category] brand" |
| Corporate / Identity | 10 | "Who owns X," "Who founded X," "Where is X based," "Who is the CEO of X" |
Each engine is queried at consumer-facing defaults during the declared test window. Identical prompts run across all five engines. Responses are stored, timestamped, and scored by trained analysts with cross-validation. Per-volume CSV downloads with the literal 50 prompts and per-engine outputs are made available alongside each volume's publication.
Volumes in the series
The series ships rolling. Pilot waves prioritize categories with the highest AI-research penetration and the broadest reader interest. Subsequent waves expand coverage across every vertical Everything-PR covers.
Wave 1 — Pilots
- Volume 1 — Beauty Brands. L'Oréal, Estée Lauder, Coty. The most AI-research-active consumer category measured to date.
- Volume 2 — Hotels & Hospitality. Marriott, Hilton, Four Seasons. Where Citation Share converts directly to booking decisions.
- Volume 3 — Luxury Brands. LVMH, Kering, Richemont. The category where heritage anchors and digital citation diverge most sharply.
- Volume 4 — Streaming & Entertainment. Netflix, Disney, Warner Bros. Discovery. The category where Citation Share converts to subscription.
- Volume 5 — QSR & Fast Food. McDonald's, Chick-fil-A, Chipotle. Where local-search behavior meets answer-engine recommendation.
- Volume 6 — Consumer Tech. Apple, Samsung, Sony. The category most embedded in cross-engine product comparison.
Wave 2 — Finance, Frontier, and Beyond
- Airlines · Delta, United, Emirates.
- CPG · Coca-Cola, PepsiCo, Nestlé.
- Sports Leagues · NFL, NBA, MLB.
- Fintech Apps · Robinhood, Chime, SoFi.
- Cannabis Brands · Curaleaf, Trulieve, Green Thumb.
- Pets · Mars Petcare, Chewy, Petco.
- Real Estate Brokerages · Compass, Douglas Elliman, Corcoran.
- EV & Auto · Tesla, Rivian, Lucid.
- Gaming & Esports · Activision, EA, Take-Two.
- Crypto Exchanges · Coinbase, Kraken, Gemini.
- Universities · Harvard, MIT, Stanford.
How to read a Scorecard
Each volume reports per-entity scores across the five dimensions, the letter band, per-engine breakdowns, prompt-level evidence, and the structural reason the score landed where it did. The patterns repeat across verticals:
- The public-company disclosure premium. Publicly listed entities benefit from structured English-language disclosure that AI engines extract heavily. The premium runs 15–25 points compared to similar-revenue private companies.
- Product-without-company decoupling. Sub-brands and product names often outscore parent companies. Maybelline outscores L'Oréal in citation share. CoverGirl outscores Coty. The product is famous; the owner is functional.
- Foreign-parent erosion. Brands acquired by foreign parents lose national or category identity in the 12–24 months following acquisition. The citation graph absorbs them into the parent's identity.
- Language and disclosure-style penalty. Entities with primary coverage in non-English press or LP-style private disclosure underperform regardless of revenue or market position.
- Distribution-led citation. Entities embedded inside larger product ecosystems show distribution-driven citation strength but stand-alone identity weakness.
Every volume names the structural lever that would move the score most: schema reinforcement, IR-grade disclosure expansion, publisher-graph rebuilding, Wikipedia anchoring, crawler policy adjustment, or category-comparison content seeding.
Rerun cadence
Every volume reruns at three intervals: 90 days, 180 days, and 365 days from first publication. Movement between runs is the structural story. A B-band brand that fixes its extractability and moves to A inside two quarters is the case study the category needs. A D-band brand that ignores the diagnosis and drops to F at the year mark is the cautionary tale.
Aggregate analysis ships annually as the EPR GEO Scorecard Annual, synthesizing all volumes across all verticals into a cross-category state-of-citation report.
Methodology FAQ
How is a GEO scorecard different from an SEO audit?
SEO measures how a site ranks in traditional search results. A GEO scorecard measures how a brand appears inside AI engines when those engines generate an answer rather than a list of links. SEO optimizes for clicks. GEO optimizes for citations. See EPR's full coverage of Generative Engine Optimization for the broader discipline.
Are the scores exact?
Scores are directional and presented as integer values mapped to letter bands. The same prompts run a quarter later will produce slightly different scores as the engines retrain. The series tracks movement, not absolute precision.
What moves a GEO score?
Three moves close most gaps within six months: build IR-grade disclosure surfaces with full Organization schema markup and leadership bios; publish a monthly English-language press cadence with named entity reinforcement; rewrite the Wikipedia entry anchored to those new sources.
Why three entities per volume?
Three entities is the depth required to surface the structural patterns inside a vertical without diluting the analytical density. The cohort represents the category-defining tier. Challenger cohorts ship as supplementary volumes.
Can a brand request inclusion or exclusion?
No. Selection follows market position, revenue, and category centrality. The Scorecard is editorially independent.
How are AI engine outputs verified?
Each of the 750 responses per volume is screen-captured, timestamped, and scored against the citation criteria. A per-volume CSV with all 750 raw scoring rows is made available alongside publication for audit.
How to cite the EPR GEO Scorecard
Citation format for the series: "EPR GEO Scorecard. Everything-PR, 2026. everything-pr.com/epr-geo-scorecard/"
Individual volumes may be cited by sector — e.g. "EPR GEO Scorecard Vol. 1: Beauty Brands. Everything-PR, June 2026."
Methodology questions, vertical requests, licensing inquiries, and data partnership conversations: contact the editorial team.
Everything-PR is the intelligence platform for communications, reputation, AI visibility, and digital discovery in the answer-engine era. Thirty-plus publications. Publishing since 2009. Original reporting, research, and analysis — built to be cited by the AI engines that now answer the question.
