Howtoresearchregulatoryfilingsacross50countrieswithAI

Key takeaways

• Check filing availability for each company and year before you start extracting figures.

• Keep the original label, units, reporting period and source location with every result.

• Flag missing disclosures and documents the agent cannot read clearly.

• Open the filing before using a figure to support a conclusion.

What makes filings from different countries harder to research?

Finding the right document is the first job. For a US company, that usually means the Form 10-K on the SEC’s EDGAR system. In other markets, the annual filing may have a different name, language or format. Apple’s 2025 Form 10-K is on EDGAR; Unilever publishes its Annual Report and Accounts 2025 as a PDF and as a tagged (ESEF) HTML file, and also files a Form 20-F; Toyota publishes an annual securities report in Japanese and a Form 20-F in English. Each is an identifiable document, but a search for “2025 annual report” will not tell you which one the agent actually found.

This matters more as the company list grows, because checking each regulator and investor relations site by hand gets slow. Bigdata.com’s data catalog brings regulatory filings from more than 90,000 companies in over 50 countries into one searchable source, with history back to 2010. Its International Filings package, built with the data partner Captide, adds annual reports, earnings materials and governance documents from companies outside the US. Coverage still varies by issuer and year, so confirm that the specific filings you need are there before you start extracting.

You can run this workflow from the agent you already use. We offer a remote MCP (Model Context Protocol) server, so Claude, ChatGPT, Cursor or any other MCP-compatible client can search our filings directly, connecting through OAuth or an API key. Our MCP setup guide covers each option.

Before extraction, make a coverage list: company, country, document type, fiscal year end, document language and source link. Mark any filing you cannot find. Also check whether you have the consolidated group report or a subsidiary’s accounts. That small step prevents a missing document or the wrong issuer from quietly becoming a row in your comparison.

What can an AI agent pull out of a filing?

Give the agent a specific disclosure to find. For example: "Find total revenue in the consolidated annual filing for the fiscal year ended in 2025. Return the figure exactly as printed, its label, currency and units, and the page or passage where it appears." That request gives you something you can check. "Analyze the company’s financial health" leaves too many choices unstated.

Use the same fields for every result: issuer, filing title, fiscal year end, filing date, original line label, reported figure, currency, units and source location. Keep the passage or a short extract alongside the number. If the agent cannot locate the disclosure, the result should say not found rather than substitute a figure from a press release or news story.

An agent can help collect those rows, but the table is still a first pass. Scanned pages, translated terms and tightly packed financial statements all need inspection in the original document. So do tables with several year columns: the right number from the wrong year is still a wrong result. When a page or column is unclear, flag it for a person to check instead of treating the extraction as complete.

How do you compare the same disclosure across countries?

Before collecting the numbers, be clear about what you want to compare. If you’re looking at revenue for the whole group, ask for the consolidated figure from each company’s annual filing. Keep the name each company uses for that figure in your first table. Calling everything “revenue” can make the numbers look more comparable than they really are.

The reporting period matters too. Each company reports a number for its fiscal 2025, but the years cover different months: Apple’s year ended on September 27, 2025, Unilever’s on December 31, 2025, and Toyota’s on March 31, 2025 (Toyota calls the year from April 2024 to March 2025 “FY2025”). Keep those dates next to the original currency and units so you can see what you are comparing. Then check what each figure includes. Unilever’s 2025 turnover covers continuing operations only: the Ice Cream business, demerged on December 6, 2025, is reported as a discontinued operation. Toyota’s total includes ¥4,437,827 million of financial services revenue. Without that context, the numbers are easy to misread.

You may not need to convert the figures into one currency at all. If you do, keep the reported amount alongside the converted figure and explain which exchange rate you used, the date it relates to and how you applied it. Add a short note wherever the businesses or measures differ. That gives the analyst enough context to judge whether the comparison is useful.

How do you check that the agent read the right document and year?

Once you have the filing open, start with the cover page. Check that you have the right company, document type and reporting period. Then find the figure the agent cited and read the table around it. What is the line called? Which year does the column cover? Are the amounts in thousands or millions? These details tell you whether the agent has picked the right number.

It is also easy to confuse the reporting period with the filing date. Apple’s 2025 Form 10-K covers the year ended September 27, 2025, but was filed on October 31, 2025. October tells you when the document was filed, not when the financial year ended.

Before using a figure in published research, check whether a later amendment or report has revised it. If the citation opens the full filing, search for the exact line label and read the surrounding table. A good agent makes this easier by taking you as close as possible to the evidence, with enough context to understand what you are looking at.

A three-market example: one question, three reported figures

Suppose you ask an agent for a consolidated 2025 revenue-related line from one company in each market. A checked first pass would look like this:

The figures and page references above come from the companies’ own filings. Toyota’s English Form 20-F reports the same ¥48,036,704 million under the label “Total sales revenues,” made up of sales of products (¥43,598,877 million) and financial services (¥4,437,827 million). The Japanese report’s label translates as total operating revenues. Unilever’s €50,503 million turnover reflects continuing operations; the demerged Ice Cream business is presented separately as a discontinued operation.

Note: this table checks where each figure comes from, it does not rank the three companies like for like: the year ends and currencies differ, and the selected lines need more work before they can answer a comparative investment question. Toyota has also since published results for the year ended March 31, 2026, which overlaps more with calendar 2025. Whichever year you use, state it in the row. Across a larger company list, keep these differences beside each row and send exceptions for review before drawing a conclusion.

Sources

• Apple 2025 Form 10-K

• Unilever Annual Report and Accounts 2025

• Toyota annual securities report for the year ended March 2025, Japanese

• Toyota Form 20-F for the year ended March 2025, English

• Bigdata.com data catalog

• Introducing the International Filings data package