---
title: "The first investor to read your earnings release isn't human · ARC"
description: "AI agents read earnings releases before human analysts do, and as of September 2026 both Anthropic and OpenAI sell that capability to Wall Street off the shelf. What that means for how public companies write, structure, and test their earnings communications before the next reporting cycle."
canonical: "https://arc.silba.ai/insights/earnings-release-agentic-ai/"
markdown: "https://arc.silba.ai/insights/earnings-release-agentic-ai/index.md"
author: "Rachel Carroll"
published: "2026-10-06"
modified: "2026-10-20"
language: "en"
---

# The first investor to read your earnings release isn't human

Your earnings release is read, scored and compared against every prior quarter before a human analyst opens it. Here is what the machine is looking for.

Rachel Carroll

- Number Insight 01
- Published 6 Oct 2026
- Revised 20 Oct 2026 · v1.2
- Reading 30 s findings · 4 min full
- Sources 8, all public
- Series Agentic markets

## Key findings

1. Reading your release by machine is now a product banks buy, not an edge a few funds built. *Anthropic, 5 May 2026 · OpenAI, 10 Sep 2026*
2. Most US institutions already use AI in their research. *68% · CAQ, Jul 2026, n = 101*
3. An agent does four things with a release: parse, compare, weight, connect. *§2*
4. Retrieval systems confuse adjacent quarters, sections and peers, and answer confidently anyway. *−13 to −20 pts · FinRank, Aug 2026*
5. Six disciplines fix this, and none is technically difficult. *Table 1*

Within seconds of your earnings release hitting the wire, it has been read, parsed, and scored by systems that never sleep, never miss a filing, and have no patience for ambiguity. They have compared the language to every previous quarter, assessed the tone, flagged the inconsistencies, and begun generating the signals that feed the research and trading workflows of the institutions that move your stock.

> The human analysts are still making their coffee.

In May 2026, this was a description of infrastructure that a handful of quantitative funds had built for themselves. On May 5, 2026, Anthropic made it a product: [ten pre-built financial-services agents](https://www.anthropic.com/news/finance-agents), one dedicated to reviewing earnings, launched with its models [already in production at JPMorgan, Goldman Sachs, and Citi](https://www.fortune.com/2026/05/05/anthropic-wall-street-financial-services-agents-jamie-dimon/). On September 10, 2026, OpenAI followed with [ChatGPT for Financial Services](https://openai.com/index/introducing-chatgpt-financial-services/), built for investment bankers and equity researchers, with Daloopa, [Quartr](https://quartr.com/newsroom/press-release/quartr-becomes-a-data-partner-to-openai-s-chatgpt-for-financial-services), LSEG News, and PitchBook data indexed inside it. The capability to machine-read your release the moment it publishes is no longer a competitive edge for the few. **It is a line item, from two suppliers.**

Most investor relations programs were not designed for this reader. The next reporting cycle is the window in which to fix it.

## Who is actually reading

The evidence is no longer anecdotal. A [Center for Audit Quality and KRC Research survey](https://thecaq.wpenginepowered.com/wp-content/uploads/2026/07/caq_institutional-investor-survey-ai_2026-07.pdf) of 101 US institutional investors, published in July 2026, found that 68% make extensive or moderate use of AI in investment research. Sixty percent regularly analyze earnings-call transcripts with it, and 48% use it to spot inconsistencies between the narrative and the financials. [SimCorp's January 2026 survey](https://www.deutsche-boerse.com/dbg-en/media/news-stories/press-releases/More-than-Two-thirds-of-Investment-Managers-Prominently-Using-AI-to-Support-Front-Office-SimCorp-Study-Reveals-4910704) of 200 senior executives at buy-side firms managing $10 billion or more found 70% using AI to support the front office, meaning the portfolio managers, analysts, and traders who make investment decisions.

Fig. 1 Nearly half of US institutions use AI to check your narrative against your numbers

- Extensive or moderate use of AI in research 68%
- Analyze earnings-call transcripts 60%
- Extract metrics from disclosures 60%
- Compare peer filings 50%
- Spot inconsistencies, narrative against financials 48%
- Compare filings across periods 47%
- Expect their use to increase within two years 83%

**Share of 101 US institutional investors** (director level and above, firms managing at least $500 million) using AI for each task. The last bar is an expectation, not current use. Source: Center for Audit Quality and KRC Research, July 2026 \[5\].

One caution belongs here. In June 2026, [Morningstar Manager Research](https://www.morningstar.com/en-gb/business/insights/research/AI-in-active-fund-management) found fewer than 20 asset managers with meaningful AI integration in portfolio construction. The decision layer at most institutions is still human. But the read layer is not. **Your release does not have to persuade a machine to buy. It has to survive a machine's first pass to reach the human who might.**

## What agentic systems do with your earnings release

When an agentic system processes an earnings release, it performs four operations simultaneously, and most IR teams have never been briefed on any of them.

**It parses language for tone against a baseline.** Not the way a human reads for mood, but systematically, comparing word choice and sentence construction against every prior release on record.

**It checks consistency across quarters.** What a company regards as normal evolution of its story registers to the system as narrative drift, which is scored negatively regardless of the business rationale behind it. The CAQ survey's 47% who use AI to compare filings across time periods for consistency are doing exactly this.

**It weights by structure.** Extraction models assign higher weight to opening sections. Companies that bury their strongest result in the fourth paragraph are underweighting their own best news.

**It tests whether the quarter connects to the thesis.** A strong result not anchored to the stated investment thesis is processed as a data point. Connected to it, the same result is processed as evidence.

## The six most common errors in the agentic era

In reviewing earnings communications across sectors and market caps, I see six patterns that consistently degrade signal quality, usually without the company or its IR team realizing it. The fourth is new to most IR teams: the [FinRank benchmark](https://arxiv.org/abs/2608.07400) found the best system recovered less than half the supporting evidence in its top ten results.

Table 1 The six errors, what the machine does with each, and the fix

| No. | Error | What the machine does | The discipline that fixes it |
| --- | --- | --- | --- |
| 01 | Narrative drift | Reads changed wording as changed conviction. | Restate the thesis, in the same substance, every quarter. |
| 02 | Ambiguous guidance | Defaults to neutral or slightly negative weighting. | Ranges, drivers and assumptions, same format each time. |
| 03 | Inconsistent terminology | Sees two businesses, or an unannounced strategy change. | One controlled vocabulary across every channel. |
| 04 | Ambiguous provenance | Answers confidently from the wrong period or entity. | Company, period, currency and definition with every figure. |
| 05 | Buried lead | Gives the best result less weight. | The three most important facts at the top. |
| 06 | Missing thesis anchors | Supplies a generic frame of its own. | Connect each result to the investment case. |

From §3. The pairing of error and discipline is ARC's.

> Your last release has already been scored by the systems described here. The only question is whether you have seen the score.

## Sources

1. Anthropic. [Announcement of agents for financial services](https://www.anthropic.com/news/finance-agents). 5 May 2026.
2. Fortune. [Report on the Wall Street launch](https://www.fortune.com/2026/05/05/anthropic-wall-street-financial-services-agents-jamie-dimon/). 5 May 2026.
3. OpenAI. [Introducing ChatGPT for Financial Services](https://openai.com/index/introducing-chatgpt-financial-services/). 10 Sep 2026.
4. Quartr. [Press release on becoming a data partner to ChatGPT for Financial Services](https://quartr.com/newsroom/press-release/quartr-becomes-a-data-partner-to-openai-s-chatgpt-for-financial-services). Sep 2026.
5. Center for Audit Quality, KRC Research. [Institutional investor survey on AI](https://thecaq.wpenginepowered.com/wp-content/uploads/2026/07/caq_institutional-investor-survey-ai_2026-07.pdf). Jul 2026.
6. SimCorp, via Deutsche Börse. [Survey of investment managers on AI in the front office](https://www.deutsche-boerse.com/dbg-en/media/news-stories/press-releases/More-than-Two-thirds-of-Investment-Managers-Prominently-Using-AI-to-Support-Front-Office-SimCorp-Study-Reveals-4910704). Jan 2026.
7. Morningstar Manager Research. [AI in active fund management](https://www.morningstar.com/en-gb/business/insights/research/AI-in-active-fund-management). Jun 2026.
8. Mansouri et al., arXiv. [FinRank benchmark: retrieval over 10-K and 10-Q filings](https://arxiv.org/abs/2608.07400). Aug 2026.

## ARC tests how AI research tools read your company.

In a workshop, we present our current findings and apply them to your own materials.

[Request a workshop](https://arc.silba.ai/contact/)

## Keep reading

### [Why your IR website is the most undervalued asset in your capital markets program](https://arc.silba.ai/insights/ir-website-agentic-optimization/)

### [The agentic investor: What the new capital allocator actually looks like](https://arc.silba.ai/insights/agentic-investor-capital-allocator/)

[All insights](https://arc.silba.ai/insights/)