Can AI Write Equity Research Reports?

AI has been steadily moving from a curiosity to a mainstream tool in financial services. Many analysts I know already lean on tools like Perplexity or ChatGPT to break down annual reports, interpret management commentary, or even sketch out the first draft of a research report. You can also read our report comparing different tools for generating financial reports here - Perplexity Finance: Is It Better Than Google and Yahoo Finance?

The shift is striking. A couple of years ago, AI in finance felt like an experiment. Today, it’s woven into daily workflows. With real-time data connections and the ability to parse financial statements or policy announcements in plain language, AI tools are reshaping how quickly research can be done. This raises the obvious question: can AI take on the full task of writing equity research reports?

The short answer: not yet. But in certain report types, it’s already proving to be a strong drafting assistant. Let's dive deeper into this question to find out more.

Reports that analysts write

In order to analyze whether AI can actually do most of the work that is done by researchers in this space, we need to first understand the types of research reports created by equity research analysts. These reports aren’t all the same, some are long and detailed, while others are short, time-sensitive updates.

Equity research is not just about sharp insights — it’s also an unforgivingly time-intensive job. Analysts often spend dozens of hours each week poring over annual reports, financial statements, regulatory filings, and industry data before even beginning to draft a report. The workload isn’t evenly spread either. During earnings season, for instance, the pace accelerates dramatically, with analysts juggling multiple companies’ results, updating models overnight, and pushing out reports before markets open. The same spike happens around major corporate actions like IPOs, mergers, or policy changes. In these windows, research teams face intense pressure to produce timely, high-quality analysis while working against the clock.

How to decide if AI can write a report

Before diving into which types of equity research reports are better suited for AI, it’s important to lay out the framework for analysis. Not all reports are created equal — some are fact-based and routine, while others rely heavily on judgement, proprietary models, and unique analyst perspectives. To evaluate where AI can really help, we’ll use three simple parameters:

1.        Human Judgement vs. Amount of Data Needed
Reports vary in how much they rely on an analyst’s judgement (e.g., setting a long-term investment thesis) versus just compiling and interpreting available data (e.g., reporting quarterly results). We’ll plot reports on this matrix to see which are naturally more “AI-friendly.”

2.        Ease of Data Accessibility
Some reports can be built largely from public sources like filings, press releases, or macroeconomic data, while others depend on gated or proprietary data sets. This accessibility directly affects whether generic AI tools can generate meaningful drafts.

3.        Use of Proprietary Financial Models
Many broking firms have in-house valuation models that are central to their research output. Reports that lean heavily on these models are harder to automate with AI, since the underlying assumptions and forecasts are not public.

Using these three lenses together gives us a structured way to assess which reports AI can generate today, which ones it can only support, and which will remain firmly in the domain of human analysts.

Human Judgement Vs Data Intensive Assessment

When we map equity research tasks against what AI does best, a pattern emerges. AI shines in areas with structured, data-heavy work that follows a clear template, like earnings reviews or flash notes. But when the task requires subjective judgment — forming a thesis, interpreting signals, or anticipating investor sentiment — humans are still at the center. The reason here in most cases is human judgement.

1. What is Human Judgment?
Human judgment in equity research is about interpreting facts, not just reporting them. For example, two companies might both beat earnings estimates, but an analyst’s role is to decide which surprise is sustainable, which is driven by one-off factors, and how markets are likely to react. It also involves weighing softer signals — management credibility, regulatory undercurrents, competitive threats — and then forming a forward-looking view. These subjective calls are hard to codify into rules, which is why analysts remain essential in producing differentiated insights.

2. What Do We Mean by Data-Intensive Reports?
Some reports are “data intensive,” meaning they rely heavily on large volumes of structured inputs like quarterly financial statements, historical ratios, peer comparisons, and consensus estimates. Earnings previews, sector data books, or macroeconomic updates fall into this category. The challenge here is not just about interpretation and also about speed, accuracy, and coverage.

3. How Do Analysts Research the Data?
To build these reports, analysts pull from multiple sources: company filings, regulatory disclosures, earnings call transcripts, financial databases like Bloomberg or FactSet, industry publications, and even primary research such as expert calls or surveys. This process is time-consuming, often involving cross-checking numbers, updating valuation models, and pulling context from news flow. While AI tools can now automate parts of this data-gathering and initial synthesis, analysts still spend significant hours ensuring accuracy and drawing the right conclusions.

Ease of Data Accessibility for Reports

All of these reports, no matter how simple or complex, are built on data. Whether it’s quarterly numbers, regulatory filings, or macro indicators, the foundation of equity research is having the right information at the right time. The quality of a report often comes down to how comprehensive and reliable the underlying data is. Some of this data is easily accessible in the public domain, while other datasets sit behind paywalls, terminals, or require specialized models. This makes a big difference when we think about how AI tools can be used for automating report writing.

In financial services, the data ecosystem is split between public sources and specialized providers. On the public side, you have company filings (annual reports, earnings releases), stock exchange disclosures, government statistics, and central bank announcements. These are freely available, though sometimes messy to process. 

On the specialized side, the big names are Bloomberg, Refinitiv, FactSet, S&P Capital IQ, Morningstar, and MSCI, which provide structured feeds of financials, consensus estimates, pricing data, and sector benchmarks. News wires like Reuters, Dow Jones, and Bloomberg News are also critical for real-time updates. Finally, many research teams use market research firms, consultancy reports, or proprietary surveys to add depth to their analysis.

Need of Proprietary Financial Models

The next important aspect to keep in mind is that most broking firms maintain their own proprietary financial models. These models are built over years of coverage and are constantly updated with new assumptions, scenarios, and analyst judgments. For many reports, these models are the backbone — without them, you can’t really generate the depth or precision needed.

Typically, they are built in Excel or specialized platforms and include forecasting models, valuation models, and scenario analysis tools.

•           Forecasting models: Analysts project revenue, margins, cash flows, and earnings for the company based on both historical performance and assumptions about future business drivers (e.g., demand growth, pricing, costs, regulations).

•           Valuation models: Common approaches include Discounted Cash Flow (DCF), Price-to-Earnings (P/E) multiples, EV/EBITDA multiples, or Sum-of-the-Parts valuations. These give analysts a sense of what the stock should be worth compared to its current market price.

•           Scenario and sensitivity models: These allow analysts to test “what-if” situations — for instance, how a change in interest rates, commodity prices, or regulatory policy might affect company performance and valuation.

What makes these models proprietary is not just the mechanics (since the formulas are widely known) but the assumptions, adjustments, and judgment calls baked into them. Each firm may weigh factors differently, apply unique discount rates, or adjust for risks in a way that reflects its house view. Over time, these models become highly refined intellectual property — often guarded closely as a competitive edge.

For many reports, especially initiating coverage, earnings previews, and valuation updates, these models are the backbone. Without them, you can’t really generate the depth or precision that institutional clients expect. On the other hand, some reports like quick event-driven flash notes can be written largely from publicly available or generic data without needing to tap into proprietary models.

Bringing it all together

When you bring these factors together — the level of human judgement required, how easily the data can be sourced, and whether the report depends on a broker’s proprietary financial model — you start to get a clearer picture of where AI can really add value and where it struggles. Some reports are heavily judgment-driven and model-dependent, making them tough to automate, while others are more data- or news-driven, where AI can shine as a drafting assistant. The table below maps the major types of equity research reports against these three dimensions and shows how easily each could be created using AI.

From this assessment, it’s clear that only a narrow set of reports — Flash Notes and Company Updates — are true candidates for automated or AI-driven drafting today. They are fast-moving, data-heavy, and largely factual in nature, making them perfect for automation.

Every other report type still requires significant analyst involvement, either because they rely on proprietary financial models (like earnings or IPO notes) or because they demand deeper judgment and narrative-building (like thematic or initiating coverage reports). In those cases, AI plays more of a supporting role — helping with research, drafting factual sections, or refining language — but the analyst’s insight remains central.