Which Sources Does ChatGPT Cite for Company Financial Data?
ChatGPT financial data citation sources are URLs that ChatGPT links when answering questions about public company figures. This 2026 study maps ChatGPT citations to SEC EDGAR, company investor relations sites, Yahoo Finance, Nasdaq, Macrotrends, and Reuters. It does not measure Google AI Overviews, Google AI Mode, or Microsoft Copilot. The findings apply only to this sample.
Quick answers
- The sample covers 1,574 ChatGPT fan-out sub-searches collected from 2026-10-18 through 2026-11-02.
- The citation records include SEC pages, company IR sites, finance aggregators, and news wires.
- One result can contain several citations. The file records each cited URL separately.
- Citation frequency does not prove that a source or answer is accurate.
- The findings cannot be applied to an engine that the study did not measure.
Download the ChatGPT financial data citation sample (CSV)
Which websites did ChatGPT cite most for public company financial data?
The downloadable CSV contains the domain-level citation records for this ChatGPT sample. The research summary does not include verified totals for each domain. This page therefore does not name a leading website or publish unsupported counts.
The records fall into four source groups:
| Source group | Examples in the sample | Typical page type |
|---|---|---|
| Regulatory source | SEC EDGAR | Form 10-Q, Form 10-K, and filing exhibits |
| Company source | Company investor relations sites | Earnings releases, presentations, and results pages |
| Finance aggregator | Yahoo Finance, Nasdaq, and Macrotrends | Financial tables and historical series |
| News source | Reuters | Dated reporting and company context |
The table describes source types. It does not rank their accuracy or authority.
Where can you see the per-domain citation counts?
The downloadable CSV contains the records needed to calculate per-domain totals. Each row identifies a cited URL, its domain, its source group, and the result that contained it.
The file covers 1,574 fan-out sub-searches collected from 2026-10-18 through 2026-11-02. Readers can group the rows by domain without relying on an unsupported summary figure.
How many citations can one ChatGPT result contain?
One ChatGPT result can contain several citations. The counting method treats each cited URL as a separate citation record, even when several URLs appear in the same result.
Citation totals therefore measure links, not unique results. The CSV includes a result identifier for counting citations by domain or distinct results. Multiple citations do not show that a result is more accurate.
Does the most cited website provide the best financial data?
No. Citation frequency measures retrieval, not accuracy, authority, or freshness. A frequently cited page may contain an old period, a different accounting measure, or a figure shown in different units.
An IR team should treat the domain ranking as a map of what ChatGPT retrieved. The team must still compare each answer with the company’s filing and official IR materials.
Why do official sources and third-party pages both appear?
Official and third-party pages appear because the measured ChatGPT results cited several kinds of documents. A result might link to a filing, an IR page, an aggregator table, or a news report.
Each source type serves a different purpose. An SEC filing can provide the reported figure and its accounting context. A company IR page can provide the earnings release or presentation. An aggregator can present a historical series. A news report can provide dated context.
What is an official source for a public company figure?
An official source is the company’s SEC filing or a document on its investor relations site. These sources provide the company’s reported information and should anchor a financial fact check.
Use the SEC forms index to identify the filing type behind a cited figure. The index covers Form 10-Q, Form 10-K, and other SEC forms. Last checked: 2026-11-02.
Official does not mean simple. A filing may contain several versions of a metric, including GAAP and adjusted measures. The checker must match the answer to the correct label and period.
What can a third-party financial page add?
A third-party page can make a figure easier to find or compare across periods. Yahoo Finance, Nasdaq, Macrotrends, and Reuters appeared among the domains classified in this sample.
A third-party page should not replace the filing during verification. Its table may use a different period, scale, currency, or accounting basis. The key test is whether its claim matches the company’s filed disclosure.
Does a citation show that ChatGPT used the page correctly?
No. A citation shows that ChatGPT linked a page. It does not prove that the answer copied the right figure or kept the correct context.
A cited page can contain several periods and measures. ChatGPT might select the right number but attach the wrong period. It might also combine a filing figure with context from another page. Citation review must cover both the source and the final answer.
How should IR teams check a cited financial figure?
IR teams should compare the ChatGPT answer, cited page, and primary disclosure side by side. A citation alone is not enough to confirm a financial claim.
- Save the full ChatGPT answer and its capture date.
- Open the cited URL rather than relying on the answer summary.
- Find the exact figure on the cited page.
- Open the related filing or official IR document.
- Match the period, units, currency, and accounting basis.
- Record the source text and the filing text.
- Assign a clear verdict and owner.
This process keeps each decision traceable to a dated answer and a primary document.
Which filing should an IR team use?
The correct filing depends on the claim. Use the filing or exhibit that contains the company’s current disclosure for that figure.
A quarterly result may appear in a Form 10-Q or an earnings release attached to Form 8-K. Annual figures may appear in Form 10-K. Guidance may appear in an earnings release or another official company document. Cite the exact filing page or exhibit, not only the filing type.
Which details should be compared?
IR teams should compare the value, fiscal period, units, currency, and accounting basis. These details can change the meaning of a figure without changing its digits.
A revenue figure in millions is not the same as a table expressed in thousands. Adjusted EPS is not the same as GAAP EPS. A trailing period is not the same as a reported quarter. Each audit row should make these labels visible.
What should an IR team record for each check?
An IR team should record enough evidence for another reader to repeat the check. The record should include the question, answer text, capture date, cited URL, filing URL, filing quote, verdict, and fix owner.
Useful verdicts include:
- Matches the filing
- Wrong figure
- Outdated figure
- Wrong period
- Wrong units or currency
- Wrong accounting basis
- Citation does not support the answer
Use the methodology when a domain count leads to a specific answer review. It explains how to compare an investor answer with company filings instead of treating citation frequency as proof.
What if the cited page is right but the answer is wrong?
The IR team should record an answer error when the source is correct but ChatGPT presents it incorrectly. A correct source does not excuse a wrong answer.
The answer may pair a correct value with the wrong period or metric. It may also omit a qualifier found on the cited page. See why AI chatbots get earnings figures wrong for common failure patterns and filing-based checks.
What should an IR team do when ChatGPT cites an old page?
An IR team should first confirm that the cited page is old and still accessible. The team can then improve or redirect pages it controls while preserving required records.
A practical response is:
- Save the cited URL and capture date.
- Confirm which period the page covers.
- Identify the current official page for that metric.
- Update the old page when appropriate.
- Add a clear link to the current disclosure.
- Use a 301 redirect only when removing or moving the old page is appropriate.
- Recheck the same question after search systems revisit the page.
Should an IR team remove every old earnings page?
No. Old earnings pages can serve investors, analysts, and regulatory recordkeeping. Removing all historical pages may make research harder.
The better fix depends on the page. Add clear period labels, preserve stable archive links, and point readers to current results. If two URLs duplicate the same content, select one canonical page. The company’s legal and records teams should review removal decisions.
Should current financial information appear in HTML?
Current financial information should appear in accessible HTML when the IR site can provide it accurately. HTML gives readers and crawlers text they can search without opening a separate file.
PDFs can still provide the official formatted document. The IR page should identify the company, period, metric, units, and links to the filing. Google states that its AI search features use standard search requirements and need no special AI markup. Source: Google Search Central. Last checked: 2026-11-02.
How can an IR team tell whether a correction worked?
An IR team can recheck the same question and compare the new answer with the saved baseline. The comparison should cover the answer text, cited URL, figure, period, and capture date.
One later result does not establish a lasting change. Save repeated observations and look for a stable correction. The Investor Answer Index provides a broader framework for tracking what AI engines tell investors over time.
Can these findings be applied to Google AI Overviews or Microsoft Copilot?
No. The citation findings apply only to the measured ChatGPT sample. The study did not collect results from Google AI Overviews, Google AI Mode, or Microsoft Copilot.
A domain that appears frequently in ChatGPT may appear less often in another engine. Each engine needs its own dated result set. The same verification standard can apply, but the ChatGPT domain ranking cannot be transferred.
What can carry over to Google AI Overviews?
The checking method can carry over to Google AI Overviews. IR teams can save the answer, capture every cited page, and compare each financial claim with the filing.
The domain findings cannot carry over because Google AI Overviews was not part of this sample. See the separate guide to Google AI Overviews financial data sources for engine-specific analysis.
What can carry over to Microsoft Copilot?
The filing-based verification process can carry over to Microsoft Copilot. The citation results cannot.
An IR team should collect a separate Microsoft Copilot sample before making claims about its most cited domains. The sample should preserve the question, full answer, citations, capture date, and filing evidence. Without that evidence, a comparison would be speculation.
Do crawler controls guarantee a citation?
No. Allowing a crawler does not guarantee that ChatGPT will retrieve, use, or cite a page.
OpenAI documents the crawlers associated with search and model access. The documentation explains crawler controls, not citation placement. Source: OpenAI crawler documentation. Last checked: 2026-11-02.
Which AI engine gives the most accurate public company financial answers?
This study cannot determine whether ChatGPT, Google AI Overviews, or Microsoft Copilot gives the most accurate financial answers. It measured ChatGPT citation sources, not answer accuracy across engines.
A valid comparison would need a separate dated answer set from each engine. Each answer would then need a check against the same company filing, period, units, currency, and accounting basis. Citation frequency alone cannot support an accuracy ranking.
Does this study show whether investors use ChatGPT for earnings research?
No. This study does not measure how many investors or analysts use ChatGPT for earnings research. It measures cited URLs from a collected set of ChatGPT fan-out sub-searches.
The sample can show which source types appeared in those results. It cannot establish user adoption, investor behavior, or research frequency. Those questions need separate usage data that this study does not contain.
How was the citation sample collected?
The sample was collected by saving cited URLs from 1,574 ChatGPT fan-out sub-searches between 2026-10-18 and 2026-11-02. The analysis covers ChatGPT only.
The collection process used these steps:
- Start with the buyer question about websites cited for public company financial data.
- Collect the related fan-out sub-searches.
- Save each ChatGPT result with its capture timestamp.
- Extract every cited URL.
- Store one row for each citation.
- Classify each URL by domain and source group.
- Retain the result identifier for distinct-result counts.
The downloadable CSV is the source for the sample size, collection window, and citation records.
What does one row in the CSV represent?
One CSV row represents one cited URL in one captured ChatGPT result. Several URLs from the same result appear as separate rows.
The row structure lets readers count citations by domain while retaining the connection to the original result. It also prevents a result with several citations from being mistaken for several separate results.
What does the sample measure?
The sample measures which URLs and domains appeared as citations in the collected ChatGPT results. It does not measure whether each answer was accurate.
Citation presence, citation frequency, and factual accuracy are separate measures. A domain can appear often while supporting an incorrect or outdated answer. Accuracy requires a second step that compares the answer with a primary filing.
What does the sample not show?
The sample does not show how Google AI Overviews, Google AI Mode, or Microsoft Copilot select sources. It also does not establish a permanent ChatGPT source ranking.
The collection window ended on 2026-11-02. Retrieval can change after that date. The findings are a dated sample, not a forecast.
What should an IR team do with this research?
An IR team should use this research to identify source types that need review. The team should then check actual investor answers against primary disclosures. The domain list is a starting point, not a verdict.
Focus first on errors that could change an investor’s understanding. Examples include wrong guidance, wrong period labels, mixed GAAP and adjusted figures, or stale leadership information. Give each issue a source, filing quote, severity, and owner.
An Investor Answer Audit checks what AI engines tell investors against the company’s own SEC filings. It returns the captured answer, cited page, filing evidence, and a ranked fix list.
See the sample audit for fictional Acme Inc.
Frequently asked questions
What are ChatGPT financial data citation sources?
ChatGPT financial data citation sources are the URLs that ChatGPT links when answering questions about public company figures. The measured sample includes SEC filing pages, company investor relations sites, finance aggregators, and news wires.
Does the sample cover Google AI Overviews and Microsoft Copilot?
No. The sample measured ChatGPT only from 2026-10-18 through 2026-11-02. It did not measure Google AI Overviews, Google AI Mode, or Microsoft Copilot.
Does a citation mean the financial figure is correct?
No. A citation identifies a retrieved page. It does not prove that the answer copied the figure, period, units, or accounting basis correctly.
Can one ChatGPT result cite several sources?
Yes. One result can contain several citations. The CSV records each cited URL as a separate row, so citation totals may differ from the number of results.
Where can I see the domain counts?
The downloadable CSV contains the citation records used for the domain analysis. It includes the collection window, sample size, cited URL, cited domain, and source group.