How ChatGPT Search Citations Work for Investor Relations Teams
ChatGPT Search citations for investor relations are source links that help teams inspect the pages behind an answer. ChatGPT Search, OpenAI, Google AI Overviews, Google AI Mode, Microsoft Copilot, and SEC EDGAR are the main entities in this guide. The review compares cited pages with Form 8-K, Form 10-Q, and Form 10-K filings.
A citation identifies a source. It does not prove that the answer matches the source or the company’s filing.
How does ChatGPT Search show citations?
ChatGPT Search shows inline citations in answers that use web search. OpenAI also describes a Sources button below the response that opens a list of cited links.
The official reference is OpenAI’s Searching the web with ChatGPT help page. Last checked: 2026-10-01.
Where can an IR team find the source links?
An IR team can open the inline citations beside a statement and the Sources button below the answer. These links provide the starting point for source review.
Open each citation beside a financial figure, executive name, transaction term, guidance statement, or strategic claim. Review every relevant link rather than stopping after the first source.
Record which citation appears beside each claim. A full source list may include several pages. Connect each page to the exact statement under review.
What does the official OpenAI help page explain?
The OpenAI help page explains how ChatGPT searches the web and displays citations. The page also explains how users can open the source list.
The help page does not replace a financial review. Investor relations teams still need to inspect each cited page. They must then compare the claim with SEC EDGAR.
Use the OpenAI help page to understand the source display. Use the applicable filing and official investor relations materials to reach a verdict.
How should the team classify the sources?
An IR team should classify every cited page before using it as evidence. Citation order should not decide which source controls the review.
Use these source types:
- SEC EDGAR filing or exhibit
- Official company investor relations page
- Government or regulatory page
- Third-party report, article, or data page
This classification helps the team find the primary record. It does not prove that the answer is correct.
What should the team do when an answer has no citations?
An answer without visible citations should be logged as unsupported, not automatically labeled false. The team still needs a filing comparison before reaching a verdict.
Save the prompt, answer, engine, and review time. Then search SEC EDGAR and the official investor relations site for the relevant disclosure.
Use narrow verdicts such as supported, contradicted, incomplete, stale, or not verifiable. These labels describe the evidence better than a broad label such as “bad answer.”
How can an IR team open and record the cited sources?
An IR team should open every relevant citation and record it beside the exact claim it appears to support. The log should preserve both the answer and the filing evidence.
Record the access date to show when the team completed its review. Keep enough detail for another reviewer to repeat the check.
What steps should the team follow?
An IR team can verify ChatGPT citations with this process:
- Save the exact prompt and complete answer.
- Record the engine, date, and review time.
- Open each inline citation tied to a relevant claim.
- Open the full Sources list.
- Record each URL, page title, publisher, and page date.
- Copy the passage related to the claim.
- Find the corresponding SEC filing passage.
- Assign a verdict, owner, and next action.
Keep one row for each claim and source pair. Separate rows prevent one supported claim from hiding another claim that lacks evidence.
What should a source log include?
A source log should connect the answer, cited page, filing evidence, and final verdict. Another reviewer should be able to repeat the check without guessing.
| Field | What to record |
|---|---|
| Review record | Date, time, engine, and reviewer role |
| Prompt | Exact question submitted |
| Answer | Complete answer as displayed |
| Claim | Exact figure or statement under review |
| Citation | URL, title, publisher, and page date |
| Source passage | Text related to the claim |
| Filing evidence | Form, item, accession number, URL, and excerpt |
| Result | Supported, contradicted, incomplete, stale, or not verifiable |
| Action | Owner, correction route, and status |
The same record structure can cover ChatGPT Search and the other active engines. Keep each engine’s answer and source list separate.
Should the team save screenshots?
Screenshots can support the review, but they should not replace text, URLs, and filing identifiers. A screenshot can show where a citation appeared beside a claim.
Save the answer as text so reviewers can search and compare it. Keep a screenshot when the citation placement matters. Record the access time and source URL in the same review file.
For each filing, save the accession number and direct SEC URL. These identifiers help another reviewer find the same filing package.
Does a citation prove that an answer is correct?
No. A citation identifies a source displayed with an answer. It does not prove that the answer matches the source, preserves its context, or uses the latest disclosure.
A cited answer can omit a condition, mix reporting periods, or repeat an old figure. The citation shows where the source review should start. It is not a financial accuracy certificate.
What does a citation identify?
A citation identifies a page presented beside or below a ChatGPT Search answer. It gives the IR team a source to inspect.
The reviewer should answer 4 separate questions:
- Does the page contain the stated claim?
- Does the page cover the correct reporting period?
- Is the page an SEC or official company source?
- Does the answer preserve the source’s conditions and context?
A source may contain the same figure while covering an older period. Matching digits alone do not prove that the full answer is supported.
How can a cited answer still be wrong?
A cited answer can be wrong when the source is stale or the answer misreads it. A later disclosure may also replace the relevant fact.
For example, an answer may state a guidance figure without its range, currency basis, or continuing-operations condition. The figure may appear in the source while the answer remains incomplete.
Compare the full claim rather than only the digits. Include units, periods, ranges, qualifiers, and the stated accounting basis.
What verdicts should an IR team use?
An IR team should use narrow verdicts that describe the evidence. Avoid an accuracy score that lacks a source trail.
| Verdict | Meaning |
|---|---|
| Supported | The answer matches the applicable disclosure and keeps material context |
| Contradicted | The answer conflicts with the applicable disclosure |
| Incomplete | The answer omits context needed to understand the claim |
| Stale | A later applicable disclosure replaced the cited fact |
| Not verifiable | The available answer and sources do not support a firm conclusion |
The MS House methodology follows the same core rule. A citation identifies a source but never replaces a filing check.
Which websites do AI engines cite most for public company financial data?
This page has no citation-frequency sample, so it cannot rank the websites cited most by ChatGPT, Google AI Overviews, or Microsoft Copilot. Such a ranking would require dated answers, a stated sample size, and a clear collection method.
For each answer, classify the visible sources as SEC EDGAR, an official investor relations page, a government page, or a third-party page. Record every URL and the review date. This shows which sources appeared in that review without making an unsupported claim about all public company answers.
The AI engine citation sources guide explains how to inspect these source types.
Which AI engine gives the most accurate public company financial answers?
This page does not provide evidence that ChatGPT, Google AI Overviews, or Microsoft Copilot is the most accurate engine. Accuracy must be checked claim by claim against the applicable company filing.
Use the same prompt, evidence fields, and verdict rules for each engine. Compare the figure, period, unit, scope, basis, status, and context. Keep each answer and source list in the review file. A citation count alone cannot show which engine produced the most accurate answer.
For related measurement, see the guide to measuring Google AI features in Search Console.
Who can check AI answers against SEC filings?
The investor relations team can coordinate checks of ChatGPT, Google AI Overviews, Google AI Mode, and Microsoft Copilot answers. The reviewer should connect each claim to the exact SEC filing passage.
Accounting can review financial presentation and reporting details. Legal or the disclosure committee can review an unclear conflict between filings. Communications can own corrections to company pages. Assign every finding to a role, and use not verifiable when the available record does not settle the issue.
Which tools help IR teams monitor what ChatGPT says?
This page does not rank commercial monitoring tools. The review described here uses ChatGPT source controls, SEC EDGAR, the official investor relations site, and a structured evidence log.
A useful process must save the prompt, raw answer, visible citations, source passages, filing excerpt, verdict, owner, and re-check date. Tool selection does not replace those records. The evidence should let another reviewer repeat the check and see why the team reached its verdict.
Which source should control when figures conflict?
The latest applicable company filing or exhibit on SEC EDGAR should control a check of filed financial information. Review amendments and later filings before deciding that an answer conflicts with the company’s record.
The official investor relations site can provide release materials and supporting context. A third-party page should not override the applicable company filing.
How should an IR team rank the sources?
An IR team should start with SEC EDGAR and then review the official investor relations site. Other cited pages can explain where an answer came from. They should not replace the primary company record.
Use this review order:
- Latest applicable SEC filing and any amendment
- Relevant exhibit or official release included with the submission
- Official investor relations page
- Government or regulatory source
- Other cited source
The applicable document depends on the claim. A Form 10-Q may cover quarterly results. A later Form 8-K may contain a newer event or update.
What if ChatGPT cites the company’s earnings release?
A company earnings release is an official source. The reviewer should also check whether the release appears as an exhibit in an SEC submission.
The EDGAR filing page provides the form type, filing date, documents, and accession number. Record these details when the release appears in the filing package.
Form 8-K Item 2.02 covers results of operations and financial condition. The SEC Form 8-K instructions explain how Item 2.02 information is treated. The instructions also cover an accompanying exhibit. Last checked: 2026-10-01.
The team should base its verdict on the actual disclosure text and any later applicable filing.
What if 2 SEC documents appear to conflict?
Two SEC documents may cover different dates, reporting periods, amendments, or purposes. An IR team should check that context before selecting the applicable record.
Compare the filing dates and reporting periods. Then check for a Form 10-Q/A, Form 10-K/A, amended Form 8-K, recast financial statements, or later disclosure.
If the conflict remains unclear, assign the item to accounting and legal review. Use not verifiable until the company identifies the applicable record.
How should an IR team compare a cited answer with an 8-K, 10-Q, or 10-K?
An IR team should divide the answer into individual claims and find the applicable SEC passage. Each comparison should cover the figure, period, unit, scope, status, and qualifiers.
Use the SEC’s EDGAR search page to find company filings. Last checked: 2026-10-01.
Which filing should the reviewer open?
The reviewer should open the filing that covers the claim and its date. A form label alone does not settle which document applies.
Start with Form 8-K when the claim concerns reported results or another current event. Review Part I of Form 10-Q for quarterly statements and related management discussion. Use Part II of Form 10-K for annual statements and the related discussion.
The SEC publishes the official Form 10-Q and Form 10-K instructions. Last checked: 2026-10-01.
Check the EDGAR filing page for exhibits, amendments, and incorporated documents.
How does the reviewer complete an SEC EDGAR filing check?
The reviewer should open the filing page, select the main document or relevant exhibit, and search for the metric or phrase. Save enough surrounding text to explain the verdict.
Record:
- Form type
- Filing date
- Reporting period
- Item or section
- Accession number
- Direct SEC URL
- Exact filing excerpt
An accession number identifies a specific EDGAR submission. Keeping it in the log helps another reviewer return to the same filing package.
What should a line-by-line comparison test?
A line-by-line comparison should test more than whether the same number appears in both places. It should check every material part of the claim.
| Test | Review question |
|---|---|
| Value | Does the number match? |
| Unit | Is it dollars, a percentage, shares, or another unit? |
| Period | Does the answer use the correct reporting period? |
| Scope | Does the figure cover the company, a segment, or continuing operations? |
| Basis | Is the figure reported or adjusted? |
| Status | Is it a result, target, estimate, or guidance range? |
| Context | Did the answer keep the stated conditions and qualifiers? |
The verdict should cite the filing excerpt that resolves each failed test.
How should the team document a mismatch?
The team should write a short mismatch statement that names the answer, cited source, filing, and specific difference. Avoid broad claims about ChatGPT Search.
Use this format:
The answer states [claim]. It cites [source]. The applicable [form and item], filed under accession number [number], states [filing evidence]. The answer is [verdict] because [specific difference].
Do not fill evidence gaps with assumptions. If the filing does not settle the issue, use not verifiable and assign further review.
What evidence should the team keep after each release?
The team should keep the raw answer, complete source list, relevant source passages, filing evidence, verdict, and action record. Organize the evidence by release, engine, prompt, and claim.
The record should show what investors could see. It should also show which sources appeared and what the company had disclosed during the review.
Which evidence belongs in the review file?
The review file should preserve both the visible answer and the evidence used to judge it.
Keep:
- Exact prompt
- Raw answer text
- Screenshot of the visible answer
- Inline citations and full source list
- Source URLs, titles, dates, and relevant passages
- SEC form, item, excerpt, accession number, and URL
- Verdict and reason
- Fix owner and status
- Review date and re-check date
This structure gives the IR team a short result with supporting evidence. Other company teams can open the underlying record when needed.
How should the team organize the evidence?
The team should keep one record for each claim under review. Each record should connect the prompt, answer, citation, source passage, filing excerpt, and verdict.
Use stable file names or record identifiers. Include the company, release, engine, review date, and claim topic. Do not place confidential information in a public location.
Keep source files separate from the summary. The summary should point to the evidence rather than copy every passage into one long document.
How should the team show change over time?
The team should compare repeated topics and verdict categories across review periods. Do not claim a trend from one answer.
Track whether each recurring topic is supported, contradicted, incomplete, stale, or not verifiable. Record whether each source is an SEC filing, an official investor relations page, or another source type.
Keep the prompt, answer, source passage, and filing excerpt beside each result. A summary without the underlying evidence is difficult to review.
What should happen after a source is corrected?
The team should save the correction request, owner, date, and updated source passage. The team should then repeat the same prompt and record the new answer.
Do not assume that a source correction will change the next answer. Complete the same citation and filing review again.
The financial figure correction guide explains how to document a correction and re-check the answer.
For timing after an update, see the IR website correction timing guide.
Which ChatGPT crawlers matter for an investor relations website?
Source citations and crawler access are different. A citation links to a source shown in an answer. Crawler access determines whether an OpenAI user agent can fetch a page. Access alone does not guarantee a citation.
For robots.txt and OAI-SearchBot checks, see the OAI-SearchBot guide for investor relations websites.
OAI-SearchBot, GPTBot, and ChatGPT-User are OpenAI user agents with different stated purposes. OpenAI explains each agent in its official crawler documentation. Last checked: 2026-10-01.
What is OAI-SearchBot?
OAI-SearchBot is the crawler an IR team should review when checking whether public investor relations pages can appear as sources in ChatGPT Search. OpenAI says this user agent supports search discovery and can link to and surface websites in ChatGPT search results.
For this citation review, record the site’s robots.txt rule for OAI-SearchBot, the final HTTP status, the canonical URL, and whether current disclosure appears in the HTML. Allowing the crawler may help OpenAI discover the page. It does not guarantee that ChatGPT will index the page, cite it, correct an answer, or update within a set time.
What is GPTBot?
GPTBot crawls content that may help improve OpenAI’s generative models. OpenAI treats GPTBot separately from OAI-SearchBot.
Blocking GPTBot does not determine whether a page can appear in ChatGPT search results. An IR team can set separate rules for search discovery and model training.
What is ChatGPT-User?
ChatGPT-User visits a page after a user asks ChatGPT or a custom GPT to open it. OpenAI states that robots.txt rules may not apply to these user-requested visits.
ChatGPT-User is not an automatic search crawler. Its access does not promise that ChatGPT will cite or retain information from the page.
How should an IR team verify crawler access?
An IR team should check the live investor relations website rather than rely on a saved configuration.
- robots.txt: Confirm the rules for OAI-SearchBot and GPTBot match the company’s policy. Note that robots.txt may not control ChatGPT-User visits.
- HTTP status: Confirm the final IR page returns status 200 without requiring a login.
- Canonical URL: Confirm the page names its live URL in the canonical link.
- Current HTML: Confirm current figures, guidance, and executive names appear in the page’s HTML.
Record the review date because access and page content can change. The generative engine optimization checklist for investor relations explains the page work that follows this access review.
For crawler-specific steps, see the OAI-SearchBot access guide.
How can an IR web team confirm OAI-SearchBot access?
An IR web team can confirm OAI-SearchBot access by testing the live robots.txt file and the affected page before investigating stale ChatGPT citations.
- Confirm robots.txt does not block OAI-SearchBot.
- Confirm the page returns status 200 without a login.
- Confirm the canonical link points to the live URL.
- Confirm the current disclosure appears in the page HTML.
OpenAI describes OAI-SearchBot in its official crawler documentation. Last checked: 2026-10-20. Record the test date because access rules can change.
Access does not guarantee that ChatGPT will select or cite the page. The OAI-SearchBot guide for investor relations websites explains the checks in more detail.
Frequently asked questions
How do I see the sources behind a ChatGPT Search answer?
Open each inline citation and the Sources button below the answer. Record the URL, page title, relevant passage, prompt, answer, and review date.
Can I trust a cited source in an answer about our results?
Not without a filing check. A citation identifies a source, but it does not prove the answer matches the source or the company’s filing.
Which document controls when a cited source and a filing disagree?
Use the latest applicable Form 8-K, Form 10-Q, or Form 10-K on SEC EDGAR. Check amendments and later filings before reaching a verdict.
What evidence should an IR team keep after each release?
Keep the prompt, raw answer, source list, relevant source passages, filing excerpt, accession number, review date, verdict, and assigned owner.
See the fictional Acme Inc. sample audit to review how findings, filing evidence, cited sources, and owners can appear in one report.