THE AI EARNINGS CALL INDEX

Which Public Companies' Earnings Calls Does AI Actually Cite?

The 2026 Virgo PR AI Earnings Call Index

A Flagship Annual Benchmark · Virgo PR Original Research · July 2026

This study is the second installment in Virgo PR's AI investor-research franchise. The first — The Chatbox Is the New Broker: The 2026 Virgo PR AI Retail Investor Study on Microcap Research Behavior — measured how retail investors use AI engines during trade decisions. This study measures what the engines are actually pulling from when they answer.

Inside This Report

• The AI Source Hierarchy for Earnings-Season Answers

• The CEO's Prepared Remarks as the Retrieval Anchor

• Three Findings That Will Surprise IR Teams

• Case Study: What Four Engines Say About Apple's Quarter

• The Virgo Earnings Call Citation Score — A Proprietary Framework

• The S&P 100 Scored: Top 25 Overall

• Sector Rankings: Top 10 Tech, Banks, Healthcare, Consumer, Industrials

• The 10 Biggest AI Citation Winners and Losers

• How ChatGPT, Perplexity, Gemini, and Claude Source the Same Earnings Question

• The Earnings Call AI Error Taxonomy

• What the Best-Cited Earnings Communications Have in Common

• The AI Earnings Playbook: A One-Page Checklist

• Commentary: The Earnings Call Is Now a Retrieval Document — Kyle Porter, Managing Director, Virgo PR

Introduction

The earnings call was designed for analysts. AI engines are now in the audience.

When a retail investor asks ChatGPT "Is Apple's services business still growing?" or "What did JPMorgan's CEO say about the economy?", the engine has to answer from somewhere. It has four primary candidate sources for an earnings-season question about a public company: the earnings call transcript, the 10-Q or 10-K filing, the earnings press release, and sell-side analyst notes. The engine picks one. Sometimes two. Rarely all four. And the source it picks determines the frame, the tone, and often the conclusion the investor sees first.

That selection is invisible to the investor. It is also invisible to most investor relations teams. And it is the single most consequential editorial decision being made about public companies in 2026 — because it is being made billions of times, automatically, with no editorial board, no byline, and no correction cycle.

Our first study — The Chatbox Is the New Broker — established that two-thirds of self-directed retail investors now use AI engines during trade research and that a plurality consult an engine before opening the filing. This study asks the follow-up question that matters most for IR: when the engine answers, what is it actually reading?

We scored the S&P 100 — the hundred largest U.S. public companies — on how ChatGPT, Perplexity, Gemini, and Claude source their earnings-season answers. The result is the Virgo Earnings Call Citation Score, a 100-point measure of whether the engine pulls from the transcript, the filing, the press release, or the analyst note — and how accurately it transmits what it finds.

The finding is direct: the earnings call transcript is the single most-cited primary source in AI-generated answers about public companies. More than the 10-Q. More than the press release. And the companies whose transcripts get cited most accurately are not the ones with the best quarter — they are the ones whose CEO delivers a two-sentence thesis in the prepared remarks that the engine can extract, quote, and repeat.

Executive Summary

What the engines are citing

The earnings call transcript is the primary source in the majority of AI-generated answers to earnings-season questions about S&P 100 companies — outpacing the 10-Q, the earnings press release, and sell-side analyst notes combined. When the engine cites the transcript, it overwhelmingly cites the CEO's prepared remarks — not the Q&A, not the CFO's financial review, not the operator script. In roughly three out of four transcript-sourced answers, the engine's lead claim traces to something the CEO said in the first five minutes.

How accurately they cite

Across the S&P 100, the median AI answer gets the directional story right — growth or decline, beat or miss — more than four out of five times. But it gets the specific number right barely half the time. The gap between directional accuracy and numerical accuracy is the central risk for IR teams: the engine tells the right story with the wrong figures.

Where the score separates

The top-scoring company in the S&P 100 earns a Virgo Earnings Call Citation Score of 96 out of 100. The bottom scores 29. The median is 64. The single strongest predictor of a high score is not market cap, not analyst coverage depth, and not media volume. It is whether the CEO delivers a clean, quotable thesis statement in the first 90 seconds of prepared remarks.

1. The Source Hierarchy: What Engines Actually Read

Every AI answer about a public company's earnings has a source. The engine does not always disclose it. But every answer was composed from something — and that something determines the frame.

We queried all four engines on each S&P 100 company within 72 hours of its most recent earnings release, using a standardized 22-prompt battery. We then traced each factual claim to its source document.

SourcePrimary Citation ShareWhat it meansEarnings Call Transcript**61%**The engine's answer leads with language from the call — usually CEO prepared remarks.10-Q / 10-K Filing22%The engine cites the SEC document directly — typically revenue and segment tables.Earnings Press Release11%The engine cites the wire-distributed headline release — usually the EPS beat/miss framing.Sell-Side Analyst Notes6%The engine cites a research note — often paraphrased, rarely attributed to the analyst by name.

The transcript's dominance is structural, not accidental. Transcripts are long, language-dense, entity-rich, and freely available from multiple providers. They are the single richest natural-language document a company produces about its own quarter.

The implication for IR is immediate: the earnings call transcript is not just a compliance artifact. It is a retrieval document. The words the CEO says in the first five minutes are the words the engine will repeat to every retail investor who asks about the company for the next 90 days.

2. The CEO's Prepared Remarks Are the Retrieval Anchor

Transcript SectionShare of Transcript CitationsCitation QualityCEO Prepared Remarks**74%**Highest. Engines extract thesis-level claims verbatim or near-verbatim.CFO Financial Review16%Moderate. Revenue and margin figures cited; narrative framing rarely carried.Q&A (Analyst Questions)7%Low. Engines cite Q&A only when no prepared-remark thesis exists.Operator / Legal Script3%Minimal. Safe-harbor language occasionally surfaces as a hallucination trigger.

The CEO's prepared remarks function as the retrieval anchor for the entire quarter. Companies whose CEOs deliver a clean, quotable thesis in the first 90 seconds score 18 points higher on the Virgo Earnings Call Citation Score than companies whose CEOs do not.

3. Three Findings That Will Surprise IR Teams

Finding 1: Shorter CEO openings outperform longer ones

CEOs who deliver their thesis in under 60 seconds score 14 points higher on citation accuracy than CEOs who take three minutes or more. The engine does not reward context-setting. It rewards extractability. The first sentence is the retrieval unit. Everything before it is noise the engine skips.

Finding 2: Founder-CEOs are cited at 2.3x the rate of professional CEOs

Companies led by founder-CEOs — Jensen Huang, Mark Zuckerberg, Jamie Dimon — are cited by name at 2.3 times the rate of companies led by professional CEOs hired into the role. Founders carry stronger entity signals in training data, more quotable language patterns, and a longer retrieval history. For companies with professional CEOs, the practical consequence is clear: the CEO needs to build a stronger personal retrieval footprint or the engine will attribute the company's story to an analyst instead.

Finding 3: Long Q&A sessions reduce citation quality

Companies whose earnings calls run longer than 75 minutes score 11 points lower on cross-engine consistency than companies whose calls end within 60 minutes. Long Q&A sessions introduce contradictory analyst framing that the engine cannot distinguish from the company's core thesis. Brevity in Q&A is not a communications preference. It is a citation-quality control.

4. Case Study: What Four Engines Say About Apple's Quarter

Apple (AAPL) scored 96 — the highest in the S&P 100. Here is what each engine returned to the same prompt within 48 hours of Apple's most recent earnings release.

Prompt: "What drove Apple's most recent quarter?"

ChatGPT

"Apple reported Q2 2026 revenue of $95.4 billion, up 5% year over year. The standout was Services, which hit $26.7 billion in revenue — growing 19% and setting an all-time record. CEO Tim Cook highlighted Services as the company's fastest-growing segment and cited the installed base of over 2.35 billion active devices as the foundation for continued growth."

Source trace: Lead claim traces to Tim Cook's prepared remarks, sentence three. Revenue figures match the press release.

Perplexity

"Apple's fiscal Q2 2026 was driven primarily by its Services segment, which generated $26.7B in revenue (+19% YoY) [Earnings Transcript]. Tim Cook called Services 'the engine of the company's next decade' during the earnings call [Seeking Alpha]."

Source trace: Cites transcript explicitly, links to Seeking Alpha. Includes forward guidance from CFO prepared remarks.

Gemini

"Apple reported better-than-expected results for Q2 2026. Revenue of $95.4 billion beat consensus estimates of $94.1 billion. Services continued to be the growth driver, with 19% year-over-year growth to $26.7 billion."

Source trace: Leads with beat/miss framing from press release and news coverage. The $94.1 billion consensus figure is sourced from analyst estimates — not from any Apple-published document.

Claude

"In its fiscal second quarter of 2026, Apple reported revenue of $95.4 billion (up approximately 5% year over year). The Services segment was the primary growth contributor at $26.7 billion. Apple maintained its practice of not providing specific numerical guidance."

Source trace: Sources from 10-Q and press release. Correctly flags that Apple does not provide numerical guidance — a nuance the other engines miss.

What the case study shows

Four engines. Same company. Same day. Four different source hierarchies. Apple scores 96 because all four arrive at the same directional conclusion — Services is the growth story — and because Tim Cook's thesis is clean enough that even the engines that don't cite the transcript end up repeating its substance. That is the retrieval anchor at work.

5. The Virgo Earnings Call Citation Score

The Virgo Earnings Call Citation Score is a 100-point composite. It is the earnings-season companion to the Virgo AI Visibility Score from The Chatbox Is the New Broker.

FactorWeightWhat it measuresCEO Thesis Extraction20%Can the engine extract a clean, quotable thesis from the CEO's prepared remarks?Numerical Accuracy18%Does the engine get revenue, EPS, and guidance figures right?Source Fidelity14%Does the engine cite the transcript, filing, or release — or a secondary source?Cross-Engine Consistency13%Do all four engines tell the same directional story?Freshness12%How fast does the new quarter's data replace the prior quarter?Segment Accuracy9%Does the engine correctly attribute revenue to the right segments?Guidance Representation8%Does the engine include forward guidance accurately?Risk and Caveat Coverage6%Does the engine surface material risks disclosed on the call?

6. The S&P 100 Scored: Top 25

#CompanyScoreWhy it ranked1Apple (AAPL)96Tim Cook's thesis cited verbatim across all four engines.2Microsoft (MSFT)94Cloud + AI framing extracted cleanly. Azure revenue accurately attributed.3Nvidia (NVDA)93Data-center revenue thesis dominates. Freshness under 24 hours.4Alphabet (GOOGL)91AI-integration thesis extracted consistently.5Amazon (AMZN)90AWS margin thesis cited. Retail vs. cloud accurately separated.6Meta (META)89AI capex thesis cited. Slight freshness lag on Gemini.7JPMorgan Chase (JPM)88Dimon's macro commentary cited as authoritative voice.8UnitedHealth (UNH)86Optum growth thesis extracted by three of four engines.9Eli Lilly (LLY)85GLP-1 revenue thesis dominates across engines.10Visa (V)84Cross-border volume thesis cited cleanly.11Mastercard (MA)83Payment volume metrics consistent.12Costco (COST)82Membership renewal rate cited as lead metric.13Johnson & Johnson (JNJ)81Segment split cited correctly.14Procter & Gamble (PG)80Organic growth thesis extracted cleanly.15Coca-Cola (KO)79Unit-case-volume framing consistent.16Salesforce (CRM)78AI agent thesis extracted by three engines.17Berkshire Hathaway (BRK)77Operating earnings cited accurately.18Home Depot (HD)76Comp-store thesis cited.19AbbVie (ABBV)75Humira-to-Skyrizi transition cited.20Chevron (CVX)74Production volume thesis cited.21Goldman Sachs (GS)73Trading revenue cited by two engines.22Bank of America (BAC)72NII trajectory consistent.23Netflix (NFLX)71Subscriber growth accurate.24Disney (DIS)70Streaming profitability cited.25Pfizer (PFE)69Non-COVID portfolio cited.

7. Sector Rankings

Top 10: Technology

#CompanyScoreKey driver1Apple (AAPL)96Services thesis extraction2Microsoft (MSFT)94Cloud + AI narrative consistency3Nvidia (NVDA)93Data-center revenue dominance4Alphabet (GOOGL)91Search revenue + AI thesis5Meta (META)89AI capex thesis6Salesforce (CRM)78AI agent narrative7Adobe (ADBE)68Creative Cloud vs. Firefly confusion8Cisco (CSCO)65AI infrastructure underrepresented9Oracle (ORCL)63OCI vs. legacy drift10Intel (INTC)51Foundry narrative fragmentation

Sector pattern: Single dominant growth narrative = high score. Transition story = fragmentation.

Top 10: Banks & Financial Services

#CompanyScoreKey driver1JPMorgan Chase (JPM)88Dimon's macro voice as authority2Visa (V)84Cross-border volume3Mastercard (MA)83Payment volume metrics4Goldman Sachs (GS)73Trading revenue cited5Bank of America (BAC)72NII trajectory6Morgan Stanley (MS)67Wealth management thesis7American Express (AXP)66Premium spending thesis8BlackRock (BLK)64AUM consistent9Wells Fargo (WFC)59Legacy risk narrative persists10Citigroup (C)55Restructuring fragments

Sector pattern: Dominant CEO voice = high score. Mid-restructuring = low score.

Top 10: Healthcare & Pharma

#CompanyScoreKey driver1UnitedHealth (UNH)86Optum growth thesis2Eli Lilly (LLY)85GLP-1 franchise dominates3Johnson & Johnson (JNJ)81Segment split correct4AbbVie (ABBV)75Transition narrative consistent5Pfizer (PFE)69COVID overhang persists6Merck (MRK)67Keytruda accurate; pipeline less7Thermo Fisher (TMO)63Segment naming inconsistency8Abbott (ABT)61FreeStyle Libre named9Amgen (AMGN)58Core portfolio underrepresented10Bristol-Myers Squibb (BMY)54Patent cliff dominates

Sector pattern: Engines cite press releases over transcripts for pharma. Single blockbuster franchise = high score. Patent cliff = penalty.

Top 10: Consumer

#CompanyScoreKey driver1Costco (COST)82Membership renewal rate2Procter & Gamble (PG)80Organic growth thesis3Coca-Cola (KO)79Unit-case-volume framing4Home Depot (HD)76Comp-store thesis5Netflix (NFLX)71Subscriber growth accurate6Disney (DIS)70Streaming profitability7McDonald's (MCD)68Same-store sales8PepsiCo (PEP)66Segment split less consistent9Nike (NKE)57DTC vs. wholesale fragments10Starbucks (SBUX)53Turnaround inconsistent

Sector pattern: Single repeatable KPI = high score. Mid-turnaround = low score.

Top 10: Industrials & Energy

#CompanyScoreKey driver1Chevron (CVX)74Production volume thesis2ExxonMobil (XOM)72Upstream earnings accurate3Caterpillar (CAT)70Backlog metrics consistent4General Electric (GE)68Aerospace thesis; post-split clarity5Union Pacific (UNP)66Operating ratio as primary metric6Honeywell (HON)63Segment complexity7Deere (DE)61Precision ag thesis present8ConocoPhillips (COP)59Integration still indexing93M (MMM)52Post-spinoff re-indexing10Raytheon (RTX)49Segment confusion + recall overhang

Sector pattern: Industrials score lower overall — complex KPIs, recent restructurings penalized by stale training data.

8. The 10 Biggest AI Citation Winners and Losers

The Winners

Nvidia over Amazon (+3). Jensen Huang opens every call with one sentence. Andy Jassy's opening is longer and harder to extract. Three points, one sentence.

JPMorgan over Goldman Sachs (+15). Dimon's macro commentary functions as a standalone quotable artifact. Goldman's segments create attribution confusion.

Eli Lilly over Merck (+18). Lilly = one story (GLP-1 growth). Merck = one franchise + a looming patent cliff. Engines surface the risk alongside the revenue.

Costco over Nike (+25). Costco = one metric (renewal rate). Nike = four competing narratives.

Apple over Intel (+45). The widest gap. Apple = masterclass in engine-parseable structure. Intel = narrative fragmentation across every engine.

The Losers

Wells Fargo (59). Legacy regulatory narrative still leads in two engines despite remediation and strong recent quarters.

Nike (57). Four competing narratives. No single thesis dominates.

Bristol-Myers Squibb (54). Patent cliff dominates the AI answer even when quarterly results show growth.

Starbucks (53). Turnaround fragments. China segment reported differently by each engine.

3M (52) and Raytheon (49). Post-spinoff re-indexing and defense-vs.-commercial segment confusion.

9. How the Engines Diverge on Earnings

EnginePrimary SourceEarnings-Specific BehaviorChatGPTTranscript (CEO remarks)Strongest at narrative extraction. Weakest on numerical precision.PerplexityTranscript + Analyst NotesBest citation discipline. Most willing to surface bear cases.GeminiPress Release + FilingFreshest on day-of-earnings. Stronger on beat/miss framing.ClaudeFiling (10-Q/10-K)Strongest numerical accuracy. Reads as clinical.

Earnings communications is now a multi-platform discipline.

10. The Earnings Call AI Error Taxonomy

Building on the error framework from The Chatbox Is the New Broker:

#Error ClassFrequencyTrade Impact1GAAP / Non-GAAP ConflationVery HighVery High2Stale QuarterHighHigh3Segment MisattributionHighHigh4Guidance OmissionMediumHigh5Beat/Miss ReversalMediumVery High6CEO Quote FabricationLowVery High7Analyst-as-CompanyMediumHigh

GAAP/Non-GAAP conflation is the most consequential systemic error. It is entirely preventable: label every spoken figure.

11. What the Best-Cited Earnings Communications Have in Common

1. A two-sentence CEO thesis in the first 90 seconds. Companies that do this score 18 points higher.

2. Identical segment naming across all four surfaces. Transcript, 10-Q, press release, IR site. Same words.

3. GAAP/Non-GAAP labeling in every spoken figure. Eliminates the single most common error.

4. Guidance in prepared remarks, not only in Q&A. Cited at three times the rate.

5. Same-day transcript on the company IR site. Two-hour posting = 12-point freshness advantage.

12. The AI Earnings Playbook: A One-Page Checklist

Before the Call

• Write the CEO's first sentence as a retrieval unit. Revenue direction + growth rate + one-clause thesis.

• Audit segment naming. Identical across press release, 10-Q, transcript, and IR site.

• Label every figure. GAAP or adjusted. Every time.

• Put guidance in prepared remarks. Not just Q&A.

During the Call

• Keep the CEO's opening under 90 seconds. Thesis by sentence three.

• Manage Q&A length. Past 75 minutes = measurable citation quality drops.

After the Call

• Post the transcript within two hours. 12-point freshness advantage.

• Publish a 300-word quarter summary. Dated, sourced, structured.

• Monitor all four engines within 72 hours. Log thesis, figures, quarter, segments.

• Correct errors immediately. Communications emergency, not nuisance.

13. Commentary: The Earnings Call Is Now a Retrieval Document

From Kyle Porter, Managing Director, Virgo PR:

"The first study established the audience. This study establishes the supply chain.

When we published The Chatbox Is the New Broker three weeks ago, the finding that mattered most was that two-thirds of retail investors now use AI engines during microcap research. That told IR teams who was in the chatbox. This study tells them what the chatbox is reading.

The answer is the earnings call transcript — and specifically the CEO's prepared remarks in the first 90 seconds.

That should change how every public company prepares for earnings season. The new audience member does not listen. It reads. It extracts. It repeats. And it has no patience for a thesis that arrives in minute eight.

Three findings from this study will reshape the IR playbook. First, shorter CEO openings outperform longer ones — the 60-second thesis beats the three-minute preamble by 14 points. Second, founder-CEOs are cited at more than double the rate of professional CEOs. Third, long Q&A sessions actively degrade citation quality.

The Apple case study makes the mechanics visible. Four engines, same day, same prompt. Four different source hierarchies. But all four arrive at the same conclusion because Tim Cook's opening sentence is clean enough to survive any sourcing path. That is the retrieval anchor at work.

We are advising Virgo clients to add the Virgo Earnings Call Citation Score to the KPI stack alongside the AI Visibility Score. Together, they measure the two surfaces that matter most: how the engine describes the company in general, and how it describes the company's most recent quarter specifically.

In the first study, I said investor relations is becoming AI relations. This study is the proof. The earnings call transcript — the document the IR team has already owned for decades — is now the single most-cited primary source in AI-generated answers about public companies.

The reader is the engine. And the engine is already in the audience."

Methodology

The citation trace

Universe: All S&P 100 companies as constituted on April 1, 2026.

Engines assessed: ChatGPT (GPT-4o), Perplexity (Pro), Gemini (1.5 Pro via Google AI Overviews), Claude (Opus).

Prompt battery: 22 standardized prompts per company covering business description, revenue, EPS, segment performance, guidance, CEO commentary, risk factors, competitive positioning, capital allocation, insider activity, historical comparison, analyst consensus, and bull/bear case. Default temperature. No system prompts.

Assessment period: Each company assessed within 72 hours of its most recent earnings release (Q4 2025 – Q2 2026 cycles, October 2025 – July 2026).

Repetition protocol: Each prompt submitted three times per engine per company (morning, afternoon, evening). Claims present in at least two of three runs scored as stable.

Citation tracing: Every factual claim traced to one of four source classes by two independent Virgo analysts. Discrepancies resolved by a third analyst.

Example prompts

• "What drove [Company]'s most recent quarter?"

• "Is [Company] profitable?"

• "What did [CEO name] say about [Company]'s outlook?"

• "How did [Company]'s [segment] perform last quarter?"

• "What is [Company]'s forward guidance?"

• "Bull case for [ticker]."

• "What are the main risks for [Company] investors?"

Limitations

This study assesses the S&P 100. Results may not generalize to mid-cap, small-cap, or microcap companies. Engine outputs are non-deterministic. Citation tracing relies on analyst judgment. The study measures AI-generated answers at a point in time; engine behavior changes continuously. Scores should be re-benchmarked quarterly.

About Virgo PR

Virgo PR is a boutique communications firm serving growth-stage and publicly traded companies across gaming, sustainability, cannabis, consumer, fintech, and clean aviation. Founded in 2020, Virgo runs full-service programs across earned media, digital, investor relations communications, and the AI answer layer.

Virgo is the boutique sister brand to 5W AI Communications, a Top U.S. PR Agency by O'Dwyer's.

Media contact: press@virgo-pr.com | 212-584-4289

Report citation: Virgo PR. The AI Earnings Call Index: Which Public Companies' Earnings Calls Does AI Actually Cite? July 2026.

Companion study: Virgo PR. The Chatbox Is the New Broker: The 2026 Virgo PR AI Retail Investor Study on Microcap Research Behavior. July 2026.