Do Investors Use AI for Earnings Research?
The question “do investors use AI for earnings research?” is an evidence question about whether investors use ChatGPT, Google AI Overviews and Google AI Mode, or Microsoft Copilot to study company results. Public research does not provide a reliable adoption rate for this task or show which engine is most accurate.
The available evidence supports a narrower conclusion. Investors can use these products to find financial information and review records in SEC EDGAR. A dated retail survey measures interest in AI-assisted investment decisions. Academic benchmarks test selected financial questions. Neither source type measures live investor use after a specific earnings release.
Do investors and analysts use AI engines for earnings research?
Public evidence does not show how many investors or analysts use AI engines for earnings research. The sources reviewed here measure stated interest, available product functions, or performance on selected financial documents.
A sound adoption study would separate retail investors, asset managers, equity analysts, and other finance professionals. It would report the sample size, collection dates, recruitment method, and exact meaning of “use.”
Without those details, an IR team should not place an adoption percentage in a board presentation. The supported conclusion is simpler. Investors can access AI tools for research, but the cited evidence does not show how often they use those tools for earnings work.
Do institutional investors use ChatGPT for earnings research?
No reliable public figure identified in this review measures institutional use of ChatGPT for earnings research. Access to ChatGPT does not prove regular use by asset managers or equity analysts.
ChatGPT documentation states that the service can search the web and provide source links. Those functions can support searches for earnings releases, filings, and investor relations pages.
The documentation is not an investor study. It has no investor sample, collection period, recruitment method, or observed research task. It does not show how often analysts use ChatGPT or whether they verify answers against filings.
Do asset managers use AI to review earnings calls?
The public sources cited here do not establish how often asset managers use AI to review earnings calls. A reliable answer would require primary research about asset managers and their work.
A study would also need a precise definition of use. Searching a transcript, summarizing comments, comparing periods, and drafting questions are different tasks. A broad question about AI use cannot isolate earnings research.
A useful institutional study would separate buy-side and sell-side respondents. It would also distinguish internal research systems from public AI engines. These details are necessary before an IR team can apply a result to board discussions.
How many investors use ChatGPT for earnings research?
No reliable public figure identified in this review shows how many investors use ChatGPT for earnings research. A useful figure would need a defined investor population and a specific research method.
The figure would also need collection dates, sample size, recruitment details, and a clear definition of earnings research. Stated interest in AI-assisted investing would not prove actual use of ChatGPT for reviewing company results.
Product documentation cannot supply an adoption rate. A benchmark score also cannot show how many people use a product. IR teams should describe these sources according to what they measure.
What primary research measures this use?
Primary research can measure AI use through surveys, interviews, observed tasks, or system records. Each method answers a different question and has different limits.
A survey measures reported behavior, stated interest, or both within a defined sample. Interviews describe how participants work. Controlled tasks measure behavior under set conditions. System records measure activity within the system that produced those records.
Product documentation does not measure investor behavior. A financial benchmark does not measure adoption. IR teams should identify the research type before using a finding as evidence.
What does the retail investor survey show?
The eToro Retail Investor Beat for the first quarter of 2023 measured stated willingness to use AI for investment decisions. It did not measure earnings research or observed AI engine use.
Opinium conducted the online survey from February 17 through March 9, 2023. The sample included 10,000 self-directed retail investors across 13 countries. Respondents held at least one investment product.
The report states that 35% of respondents were willing to use AI to guide or alter their portfolios. The figure measures stated willingness within that sample.
What are the limits of the retail survey?
The 2023 retail survey does not show that 35% of investors used ChatGPT, researched earnings, or acted on an AI answer. It measured willingness within a defined population.
The study came from an investment platform. IR teams should apply the result only to the surveyed population and question. A careful board statement would say that one 2023 retail survey found stated interest in AI-assisted investment decisions.
A retail survey also cannot establish behavior among institutional analysts. The populations and research tasks are different.
What information should an adoption study report?
A credible adoption study should report its population, sample size, collection dates, recruitment method, and definition of use. These fields show whether the result answers the board’s question.
The population must be precise. “Finance professionals” can include people with many different roles. That group cannot stand in for shareholders, asset managers, or equity research analysts without supporting data.
The method should state whether respondents reported behavior or completed an observed task. Reported use and observed use are different measures. Interest in possible future use does not prove current use.
Is there a study that measures how often AI engines get earnings figures wrong?
No cited study measures how often live AI engines get a company’s earnings figures wrong immediately after a release. Existing financial benchmarks test selected questions and documents under defined conditions.
A benchmark can show whether a system answers its test questions correctly. It does not establish an error rate for ChatGPT, Google AI Overviews, Google AI Mode, or Microsoft Copilot in live investor searches.
A live answer also depends on retrieval, source selection, timing, question wording, and response generation. A study must test those parts together before its results can describe live earnings answers.
Is there research on Fortune 500 answer accuracy?
The cited sources do not provide a live accuracy rate for AI answers about Fortune 500 companies. FinanceBench and FinQA test financial question answering, but their measured populations are questions and documents.
The Investor Answer Index covers Fortune 500 companies each earnings season. Its purpose differs from an adoption survey. An accuracy claim still requires a dated answer, the tested engine, the question, and the filing used for the check.
IR teams should not apply a benchmark score to a live engine without evidence that the benchmark tested the same task.
What do financial question-answering benchmarks measure?
Financial question-answering benchmarks measure performance on selected questions and documents. They do not measure investor adoption or live answers after an earnings release.
A benchmark supplies a defined test set and reference answers. This structure helps researchers compare performance under controlled conditions.
The benchmark population is not a group of investors. Its results cannot show how many people use an engine, which sources a live engine selects, or what the engine says at another time.
What does FinanceBench measure?
FinanceBench is an academic benchmark first submitted on November 20, 2023. Its authors created 10,231 questions about public companies.
The dataset includes answers and supporting evidence from financial documents. Its measured population is a constructed set of financial questions and documents, not investors or analysts.
FinanceBench can help researchers evaluate retrieval and question answering against reference material. It cannot show how many investors use ChatGPT, Google AI Overviews, Google AI Mode, or Microsoft Copilot.
What does FinQA measure?
FinQA is an academic dataset for numerical reasoning over financial reports. The authors first submitted the paper on September 1, 2021.
The dataset contains 8,281 question-and-answer pairs based on 2,789 financial reports. It combines written questions, financial text, tables, answers, and reasoning programs.
The measured population is a set of financial documents and constructed questions, not investors or analysts. FinQA tests reasoning over supplied material. It does not measure what investors ask after a release or whether a live engine finds the newest filing.
Can a filing benchmark predict live answer accuracy?
A filing benchmark cannot predict live answer accuracy by itself. Live answers also depend on retrieval, source selection, timing, question wording, and response generation.
A system may answer correctly when researchers supply the relevant filing. A live search product may retrieve a different source. These tests measure different parts of the process.
IR teams should separate document reasoning from live retrieval. Why AI chatbots get earnings wrong describes how stale material and retrieval timing can affect an answer.
Which AI engine gives the most accurate public company answers?
The cited evidence does not establish whether ChatGPT, Google AI Overviews, Google AI Mode, or Microsoft Copilot gives the most accurate answers about public company financials.
A valid comparison would ask each engine the same company-specific questions during the same period. It would then check every material claim against the same SEC filings.
A result from one question would not establish overall accuracy. Each engine can return different answers, citations, or sources across questions and capture times.
IR teams should record each engine separately. They should not apply a single accuracy label without dated captures and filing evidence.
Which websites do AI engines cite for public company financial data?
The cited studies do not rank the websites that ChatGPT or Google AI Overviews use most for public company financial data. A source ranking would require a defined sample, collection date, question set, and capture method.
An IR team can still record the source used for each company answer. The record should include the engine, capture time, cited page, complete answer, and relevant filing text.
A source record shows whether an engine used an SEC filing, an official investor relations page, or another publisher. It also identifies the page that may need attention.
How do institutional and retail use differ?
Current public evidence does not support a measured comparison of institutional and retail AI use for earnings research. A valid comparison would need to study both groups under the same method.
The research should ask both groups the same questions. It should report the sample size for each group, collection dates, recruitment method, tools included, and tasks observed.
The study should also measure verification. An answer checked against an SEC filing represents a different research process from an answer accepted without review.
Why does investor type matter to IR teams?
Investor type matters because an adoption rate applies only to the population that a study sampled. A result from self-directed retail investors cannot establish behavior among institutional analysts.
IR teams usually cannot identify who received a specific engine answer. They can still inspect the answer, its cited source, and the relevant filing. That evidence supports a board discussion without guessing about the reader.
The observable issue is what the engine returned during a recorded test. An IR team should not claim that a particular investor relied on that answer without direct evidence.
What questions do investors ask after a release?
No public dataset identified in this review ranks the exact questions investors ask after every earnings release. A useful question set must reflect the company, its disclosures, and current search behavior.
Relevant topics can include reported results, guidance, cash flow, business segments, management, and material events. These are broad topic areas, not a fixed public list of audit questions.
The tested topics should change when company disclosures and investor interests change. Stable facts may need less revision than results or major events.
How does MS House create the question set?
MS House does not reuse one fixed list for every company. The tested topics vary by company and quarter.
A prediction model forms a synthetic set from question patterns found across Google, Bing, DuckDuckGo, ChatGPT, Reddit, and Quora. Academic research about investor search behavior also informs the process.
Stable company facts may produce similar topics over time. Results and material events can change the topics selected for a later quarter.
Which questions should IR teams test first?
IR teams should start with topics where a wrong or stale answer could affect understanding of the company. The company’s filing and the importance of each disclosure should guide the order.
A practical review can begin with current guidance, reported results, material business changes, and leadership facts. Every finding should point to an SEC filing or the official investor relations site.
MS House does not publish its audit question wording. The public Investor Answer Audit methodology explains how MS House selects topics, captures answers, and checks claims.
Which AI engines should an IR team check?
An IR team should check every active engine in its measurement process. A result from one engine does not establish what another engine returns.
MS House currently checks ChatGPT, Google AI Overviews, Google AI Mode, and Microsoft Copilot. MS House records each result separately.
For each test, record the engine, capture time, answer, cited page, and relevant filing text. This evidence makes the finding easier to review and reproduce.
A source record also points to the page that may need attention. That page could belong to the company or another publisher.
What are the limits of current studies?
Current studies often answer narrower questions than readers assume. A benchmark tests selected documents. A survey measures reported behavior or interest. Product documentation explains available functions.
None of the cited sources measures live investor answers after a specific earnings release. None provides a reliable market-wide rate for investors using AI engines to study earnings.
Timing and population also limit each finding. A result describes only the information, systems, and participants included during the stated period.
Why are news summaries weak board evidence?
News summaries may omit the population, sample size, collection dates, and research method. The original report gives board readers a better basis for evaluating a claim.
IR teams should cite the original survey, academic paper, regulator publication, or official documentation. Board readers should be able to inspect the method behind each material claim.
This review does not use news coverage as evidence. It also does not present a benchmark score as the error rate for live investor answers.
Does Regulation FD cover AI engine answers?
Regulation FD governs selective disclosure by issuers and people acting on their behalf. It does not govern the independent output of a third-party AI engine.
The SEC adopted Regulation FD in August 2000. The rule took effect on October 23, 2000. The SEC’s adopting release provides both dates and explains the rule’s purpose.
The SEC published guidance on company websites on August 1, 2008. The guidance addresses issuer websites and disclosure. It does not state that an independent AI-generated answer becomes an issuer statement.
IR and legal teams should evaluate their own disclosures and duties. MS House does not provide legal or investment advice.
What should IR teams measure instead?
IR teams should measure the answers available about their own company. The core evidence is the captured answer, the source used, and the filing text that confirms or corrects each material claim.
This method does not claim that every investor saw an answer. It measures what the tested engine returned at the recorded time.
Repeated captures can show whether a stale claim remains, disappears, or returns. Each finding should keep the answer and filing evidence together.
How should an IR team build a baseline?
An IR team should use a company-specific question set and update it each quarter. Stable company facts may require fewer changes than questions about results or major events.
For every capture, record these items:
- The engine and capture time.
- The question or topic tested.
- The complete answer.
- The cited or linked source.
- The relevant SEC filing text.
- The finding and proposed owner.
Compare the captures over time. The comparison shows whether stale claims remain, disappear, or return.
What errors should IR teams record?
IR teams should record wrong figures, stale periods, unsupported claims, source mismatches, and missing context. Each finding should include evidence rather than a general accuracy label.
A stale answer may rely on an older company page. A source mismatch may cite a secondary page when an SEC filing contains the controlling disclosure. Missing context may combine measures that the company reports separately.
Each finding should name the source or page that needs attention. This gives the IR, finance, legal, or communications team a clear next step.
Should an IR team add an AI search investor perception study?
An IR team can add an AI search investor perception study when it needs evidence about what AI engines tell investors. The study should complement, not replace, analyst interviews or other perception work.
A useful study should capture each engine answer, record its cited source, and compare every material claim with the company’s SEC filing. Repeated captures can show changes over time.
The AI search investor perception study explains this approach. The study should not claim that every investor saw or relied on a tested answer.
What should the board see about AI and earnings research?
The board should see verified company findings, not an unsupported adoption rate. Each material finding should show the engine answer, cited source, filing evidence, and proposed fix.
Rank findings by their effect on understanding. Put wrong or outdated material claims before minor wording issues. Keep the opening summary short so the CFO can see the main issue first.
The sample audit for fictional Acme Inc. shows how MS House presents evidence and assigns an owner. It also shows the report format before a company buys an audit.
Frequently asked questions
Do investors use AI engines for earnings research?
Public evidence does not provide a reliable adoption rate for this specific task. IR teams can still measure what active engines say about their company and check each answer against SEC filings.
Is there a study that measures how often AI engines get earnings figures wrong?
Public benchmarks test selected questions over financial documents. They do not measure live answers about one company immediately after an earnings release.
How many investors use ChatGPT for earnings research?
No reliable public figure identified in this review answers that question. A useful figure would need a defined investor population, collection date, sample size, and research method.
What should IR teams measure instead?
IR teams should capture what each active engine says, record the cited source, and compare every material claim with the company’s SEC filing.
Does Regulation FD cover AI engine answers?
Regulation FD governs selective disclosure by issuers and people acting on their behalf. It does not govern the independent output of a third-party AI engine.