How the ranking is calculated.

What the Index measures

The Public Company AI Adoption Index is a composite 0–100 ranking of S&P 500 companies by their disclosed AI economic exposure, scored from the AI Adoption Tracker dataset. Each company gets one composite that summarizes three dimensions: how mature the company’s AI deployment is (Depth), how willing management is to put numbers on AI’s financial impact (Disclosure), and how pervasive AI is across the company’s internal operations (Breadth).

The Index counts both internal AI deployment and AI-driven revenue from existing products as legitimate AI exposure. A software firm that monetizes AI features inside its product is captured. So is a real estate firm that earns revenue from leasing data centers to AI buyers, an industrial that sells cooling equipment into the AI buildout, or an investor that holds AI-infrastructure assets in its portfolio. The investor’s question is “is this company’s economics being shaped by AI,” and the answer is yes in all of those cases — they differ in mechanism, not in materiality. The leaderboard tags each company as Adopter, Hybrid, or Beneficiary so the mechanism is visible at a glance.

The three pillars

Depth — 40% of composite

How mature the company’s AI deployment actually is. Two inputs: the company’s maximum AI maturity stage observed on the call, and the maximum specificity score reached in any mention.

• silent (base 0) — no substantive AI mentions on the call
• exploring (20) — vague intent or capability assessment
• piloting (45) — named use case with active pilot
• scaling (70) — operational deployment with measurable scope
• monetizing (90) — AI generating attributable revenue

Specificity adds up to a 10-point bonus on top of the stage base: bonus = (max_spec / 5) × 10. A company at scaling with max specificity 5 scores 70 + 10 = 80. A company at monetizing with max specificity 5 scores 100.

Disclosure — 40% of composite

How willing management is to quantify AI’s financial impact. The pillar credits any directly AI-driven revenue, whether the revenue comes from the company’s own AI products or from existing products and services sold into AI-driven demand. Two parallel tracks; the higher wins.

Track A — Revenue disclosure: a ladder over the disclosure method (not_disclosed 0 → qualitative_only 20 → arr / run_rate / bookings 65 → explicit_segment 80), with up to +20 bonus for amount and growth rate.

Track B — Quantified non-revenue outcomes: counts structurally quantified outcomes (cost savings, productivity, margin, customer metrics) where management gave a number with a unit. 1–2 outcomes → 40, 3–4 → 55, 5+ → 70.

The two-track design captures companies that disclose AI impact operationally without breaking out revenue — major banks, mass-market retailers, healthcare insurers — which a revenue-only score would zero out.

Breadth — 20% of composite

How pervasive AI is across the company’s own internal operations. Counts distinct adoption scopes appearing in mentions: product_embedded, internal_use, product_standalone, infrastructure_build. The two context scopes (customer_demand_signal, vendor_supply) deliberately don’t count because Breadth measures internal pervasiveness, distinct from Disclosure’s broader revenue framing.

Scoring: 0 scopes → 0, 1 → 35, 2 → 65, 3 → 85, 4 → 100. The curve is concave because the jump from zero to one is the most informative; three vs four matters less.

The composite

composite = (Depth × 0.40) + (Disclosure × 0.40) + (Breadth × 0.20)

Internally calculated on the unrounded weighted average so tie-breaking is precise; rounded to integer for display. Companies with identical unrounded composite scores share a rank with a T- prefix (competition ranking).

Views and tech-filter

• Non-Tech(default) — excludes the Information Technology sector and the four Big Tech reclassifications (AMZN, GOOG, META, TSLA). The analytically useful cut for stock-pickers — AI exposure inside the major tech vendors is already well-covered; the harder question is who’s adopting outside the tech sector.
• Overall — all scored companies, tech included, for cross-sector context.
• Per-sector — rankings within each GICS sector with reclassifications applied.

Exposure type tag

Every company also carries an exposure-type tag — Adopter / Hybrid / Beneficiary — based on the ratio of adoption-scope mentions to buildout-context mentions in its disclosures. The tag is informational only; it does not affect the composite. It exists so a reader can distinguish “company is adopting AI internally” from “company is benefiting from AI demand” at a glance.

• Adopter— adoption share ≥ 80%. AI discussion is dominated by the company’s own deployment, products, or operations.
• Hybrid — adoption share 30–80%. Both internal deployment and buildout-driven demand are real parts of the story.
• Beneficiary — adoption share ≤ 30%. AI exposure dominated by demand from AI customers or capital into AI infrastructure.

Validation

Before launch, the Index was audited with a four-test pipeline that re-runs in ~15 minutes on every snapshot refresh. Each test produces a versioned report kept in the project’s audit trail.

• Test 1 — Quote integrity: 99.31% of quotes pass the verbatim contract (exact or fuzzy-match with conservative punctuation normalization). The remaining ~1% are analyst-question fields, openly disclosed as paraphrased for readability. Zero misleading content in management quotes.
• Test 2 — Extraction fit: 20 of 30 sampled companies show internally consistent extraction; 10 minor drift; 0 significant drift. Drift cases are mostly summary phrasing, not substantive errors.
• Test 3 — Categorical accuracy: Per-field agreement with an Opus 4.6 judge — stage 82%, disclosure method 80%, outcomes-type 88%, scope 54%, specificity 52%. The lower scope and specificity numbers reflect gray-area judgment calls and a known systematic +1-tier specificity bias respectively, both bounded in effect on the ranking and documented for a v0.2 prompt-tightening pass.
• Test 4 — Pairwise ranking:98% alignment on meaningful-gap pairs. Of 54 adjacent and sector-top pairs tested, only one has the judge picking a ranking opposite to the Index when the composite delta is ≥ 3 points. The other “raw disagreements” are within rubric noise (companies with composite scores within 2 points of each other).

Known limitations

• Single-quarter snapshot. Q1 2026 only. Momentum and trend rankings (improvers, decliners) arrive when 2+ quarters are extracted.
• No Talent pillar. Companies investing heavily in AI talent without disclosing it on calls are under-represented.
• Disclosure quality is not adoption itself. The Index measures what management will quantify on the record. For an internal capability benchmark, this is the wrong tool.
• Self-disclosure bias. Silence on a call does not prove the absence of AI; it proves the absence of disclosure.
• Methodology evolution.Weights and rubric tables are revisited periodically against validation results. Rankings will shift as the methodology iterates; the validation infrastructure documented above is re-run on every snapshot refresh and any material changes are recorded in the methodology’s public changelog.

What the Index is not

• Not a stock recommendation or buy/sell signal
• Not a measure of actual AI capability or competitive position
• Not an AI vendor or capability benchmark — it ranks adopters, not enablers
• Not a forecast of future AI revenue
• Not a substitute for fundamental analysis

Underlying dataset

The Index is built on the WireSift AI Adoption Tracker — a structured-extraction pipeline that reads every S&P 500 earnings call transcript and produces a comparable, auditable dataset of every AI claim management makes on the record. The full Tracker methodology — source tiering, extraction prompts, quality controls — is documented at /methodology.

How to verify any number

Every aggregate on the leaderboard is clickable. Row-expand on the leaderboard surfaces the raw scoring inputs for that company. Click a ticker to land on the per-company page, where every verbatim quote that informed the score is rendered with speaker, role, and section labels. If you suspect a number is wrong, click through to the source quote and verify it against the original transcript.