How I Analyzed a Telegram Channel Export with Codex

I analyzed nearly three years of a public Telegram channel without converting the export to JSON, writing a parser, or uploading files one by one. I exported only the channel’s text, gave the Codex desktop app the resulting directory path, and described the result I wanted.

Codex inspected the HTML files and produced a detailed Markdown report that I could use as a starting point for a repeatable investment process. The whole run took 9 minutes and 9 seconds in my case.

The workflow was only three steps:

1. Export the channel as text without unnecessary media.
2. Give Codex the path to the export directory.
3. Ask one concrete question about the result I wanted.

This was a public broadcast channel, not a private conversation. A local file path should not be treated as proof that processing is entirely local. Before using private messages or other sensitive material, review what Codex is allowed to access and do as well as the privacy and data-use settings for your account and product.

This article describes a technical workflow, not investment advice.

Why I analyzed the channel

The channel belonged to an investment expert and contained several years of posts. Reading individual posts was useful, but it did not give me a concise view of the complete process.

I wanted Codex to answer three connected questions:

• What is the channel mainly about?
• How does the author research and select investments?
• Which parts of that process could become a repeatable workflow with my broker?

The goal was not to copy individual trades or ask AI for a stock tip. I wanted a structured first draft that I could refine later.

Exporting only the text from Telegram Lite

I used the chat-level export described in my guide to exporting Telegram chat history on a Mac. In Telegram Lite, I opened the channel, selected Export chat history, and set the range from October 2023 through August 2026.

I disabled photos, videos, voice messages, files, stickers, GIFs, and other media. Only the text was needed for this experiment. Telegram Lite exported a directory; the chat-level dialog I used did not offer a JSON format selector.

The final directory occupied 4.3 MB. Two of its HTML files were about 1.9 MB each. During the analysis, Codex reported three HTML files containing 2,779 records, including 1,484 text messages.

Excluding media kept the input small, but it created an important limitation: Codex could not inspect screenshots of trades or verify claims that appeared only in images.

Giving the directory to Codex

I used the Codex desktop app and supplied the path to the exported directory. The original prompt was short. Translated to English and with my username removed, it was essentially:

/Users/your-name/Downloads/Telegram Lite/ChatExport_YYYY-MM-DD

Analyze this Telegram chat export. It belongs to an investment expert.
Study how the expert works so I can reproduce the process with my broker.

That was the entire setup. I did not split the files, convert the HTML, create embeddings, or prepare a separate index.

OpenAI’s general guidance for analyzing datasets and producing reports recommends starting with a concrete question, inspecting the source files, and keeping generated reports separate from the inputs. In this case, Codex handled those steps from the single request and left the Telegram export unchanged.

What Codex produced

Codex first described what it had found: the number of files and records, the date range, and the fact that media was unavailable. It then prepared both a plain-language article and a more detailed Markdown analysis.

The report covered:

• the main subjects discussed in the channel;
• the author’s broad investment model;
• the recurring research and trade-planning process;
• portfolio and risk-management rules mentioned in the posts;
• how the publishing rhythm changed over time;
• where educational material ended and promotion began;
• which parts could become a workflow I could use with my broker;
• links to important messages that were worth reading in full; and
• which details were missing or could not be verified from public text.

The most useful part was not any individual investment idea. It was the attempt to turn scattered posts into an ordered process: research an idea, check the available evidence, define a plan, record the decision, and review the result.

The report linked its main findings back to specific messages in the export. That let me open the most relevant posts, read their surrounding context, and study important details instead of relying only on the generated summary.

The report also resisted presenting the channel as a complete trading system. The exported text did not contain a full history of trades, consistently stated risk limits, or enough evidence to verify performance claims. Because the media was excluded, screenshots could not fill those gaps either.

How I plan to use the result

The Markdown report is a starting point, not a finished investment strategy. I can use its structure to define my own process and return to Codex later with more specific questions.

For example, future iterations could clarify which rules are explicit, which were inferred from repeated posts, and which still need independent research. That is more useful to me than trying to imitate every opinion or market call made by the channel author.

Nothing in the generated report proves that the reconstructed approach is profitable or suitable for another person. Public posts may omit losing trades, private decisions, costs, taxes, and changes made after publication. Any financial claim still needs to be checked against its original source.

Takeaway

A Telegram channel export does not always need a custom data pipeline before AI can make it useful. For this text-only, 4.3 MB archive, giving Codex the directory path and asking one concrete question was enough to produce a useful first report in a little over nine minutes.

The important choices were simple: limit the export to the relevant public channel and date range, exclude media I did not need, describe the desired outcome, and treat the generated analysis as a starting point rather than a verified investment strategy.