Dattva Research · July 2026
Paytm Published a Markdown File for AI Agents. Here Is What That Means for Every B2B Company.
Paytm filed an Agent Datapack — a plain markdown file — alongside its Q1 FY2027 results, becoming the first major Indian public company to structure financial disclosures for AI agents to read directly. The same logic now applies to every B2B company whose buyers research vendors through AI.
When Paytm filed its Q1 FY2027 results, it published three documents: the usual PDF, the Excel, and a new one — an "Agent Datapack" in plain markdown format. The markdown file contained the same financial data as the PDF, structured specifically for AI agents to parse without rendering. It is the first major Indian public company to do this. It will not be the last.
What the Agent Datapack Actually Is
A markdown file is plain text with minimal formatting signals — headers marked with hash symbols, bold with asterisks, tables with pipes. No images, no fonts, no layout. It renders visually in a browser or markdown editor, but it does not need to. An AI agent reading the raw text gets the same information whether it renders or not.
Paytm structured its financial disclosures this way deliberately. The Agent Datapack contains the same quarterly numbers — revenue, GMV, merchant count, loan disbursements — that appear in the PDF investor presentation. The difference is not the data. The difference is the intended first reader.
The PDF is written for an analyst who will open it, scroll it, and highlight numbers. The Agent Datapack is written for a model that will parse it, extract structured facts, and use them to answer questions. Same numbers. Different architecture.
What It Is Not
It is not a replacement for the PDF or the investor deck. Paytm published all three simultaneously. The Agent Datapack is not simpler or shorter — it contains the same level of detail as the PDF. It is not a summary for lazy readers. It is an additional format for a different class of reader: automated systems that need clean, unambiguous structure to extract facts reliably.
It is also not a marketing document. The markdown file does not have brand colours, an investor relations header, or a forward-looking statements disclaimer formatted for visual scan. It has data, headers, and tables. The absence of presentation layer is the point.
What It Signals
Three things, in order of immediacy.
First: Paytm's investor relations team is operating on the assumption that a non-trivial portion of their financial data will be read by AI agents before it reaches a human analyst. This assumption is not speculative — buy-side analysts at large institutions are already using AI tools to process earnings releases before the analyst call. Paytm is optimising for that reader.
Second: the decision to publish a markdown file alongside a PDF is a structural statement about what "publishing" means. Historically, a company published a report and readers adapted to its format. The Agent Datapack inverts this — the company is adapting its format to how its readers now work. Investor relations teams that do not make this adaptation are publishing into an increasingly asymmetric landscape: their competitors' numbers will be cleanly extracted; theirs will be approximated from PDF parsing.
Third: this pattern will not stay in investor relations. The same logic that drives a finance team to publish a markdown Agent Datapack drives a product team to publish an llms.txt. It drives a content team to restructure blog posts so that the answer appears in sentence one rather than paragraph four. It drives a developer to ensure pricing pages render in the base HTML rather than after JavaScript executes. The Agent Datapack is one expression of a broader shift in how information is prepared for AI consumption.
Financial analysis used to be humans reading reports, then software trying to catch up. That order is flipping. The report is now being written for the software first. Paytm did not announce this as a strategic initiative. They just published a markdown file. That is how infrastructure shifts tend to start.
The B2B Parallel
Most B2B companies will not publish quarterly earnings. But the same principle applies to every piece of content that describes what they do, who they serve, and why a buyer should choose them over a competitor.
When a buyer's AI agent researches vendors on their behalf — visiting websites, comparing features, extracting pricing — it is doing to B2B product pages what AI analyst tools are doing to Paytm's earnings release. It is trying to extract structured, attributable facts from content that was built for a different reader. The product pages that are structured for this extraction get cited. The ones that are not get paraphrased at best and skipped at worst.
The B2B equivalent of the Agent Datapack is an llms.txt file — a plain-text document at the root of a company's domain that describes the company, its products, its target customers, and its key credentials in unambiguous language. The format is deliberately simple: no HTML, no styling, no metadata. Just text that a model can read and trust. A company that publishes a well-written llms.txt has done the same thing Paytm did with its markdown file: it has adapted its format to how its readers now work.
llms.txt is to B2B product content what the Agent Datapack is to financial disclosures. One is written for AI agents researching vendors. The other is written for AI agents analysing public companies. The structural principle is identical: plain text, direct statements, no decoration, written for a model before it is written for a human.
What Changes and What Does Not
Publishing structured content for AI agents does not replace writing for humans. Paytm still published the PDF. The llms.txt file sits alongside a company website, not instead of it. The Agent Datapack and the investor presentation cover the same material for different readers.
What changes is the assumption underlying content preparation. The implicit assumption in most corporate content — investor disclosures, product pages, service descriptions — has been that a human is the first reader and formatting serves that human. The Agent Datapack signals a different assumption: that an AI agent may be the first reader, and formatting should serve that agent's parsing requirements as well.
Companies that make this assumption explicit produce better AI-readable content as a byproduct. The discipline of writing for a model — direct statements, named data points, no preamble before the answer — also produces clearer content for the human who reads it second.
Small file. Bigger shift.
Frequently Asked Questions
What is an Agent Datapack?
An Agent Datapack is a structured plain-text file — in Paytm's case, a markdown file — that contains the same information as a formal disclosure document but formatted for AI agents to parse directly rather than for humans to read visually. It strips away layout, design, and presentation elements and retains only data, headers, and tables. Paytm published one alongside its Q1 FY2027 results — the first major Indian public company to do so.
What is the difference between an Agent Datapack and an llms.txt file?
An Agent Datapack contains financial or operational disclosures formatted for AI agent consumption in an investor relations context. An llms.txt file contains a company's product and service description formatted for AI agent consumption in a vendor research context. Both are plain-text files structured for model parsing rather than human reading. The content differs; the structural principle is identical.
Does publishing an llms.txt file mean rewriting all existing website content?
No. The llms.txt file sits alongside existing content at the root domain. It does not replace any existing page. It is an additional file that AI crawlers read before they read page content. A well-written llms.txt describes what the company does, what its products are, who its customers are, and what credentials it holds — in plain text, without HTML or formatting. Writing one typically takes a few hours and requires no changes to the existing website.
Will every company eventually publish AI-readable structured content?
The trajectory suggests yes, in the same way that every company eventually published a website and eventually built a mobile-responsive version. Each adaptation was driven by a shift in where the audience was. The audience is increasingly AI agents — in investor research, in vendor comparison, in procurement research. Companies that adapt their content format to this reader will have their information extracted accurately. The Paytm Agent Datapack is early evidence of this adaptation becoming deliberate rather than accidental.
See where your brand stands in AI answers today
Run a free AI Visibility diagnostic across ChatGPT, Perplexity, Gemini, and Claude — with prioritised, copy-paste fixes at no cost.