Make every AI read your document correctly.

The Papers is not a PDF-to-Markdown converter. Conversion is the first step. Before an AI Version is published, its figures, currencies, dates, periods, qualifiers and formulas are checked against the original document, each value keeps a reference to its source page or cell, and anything the source does not settle is put to the document owner instead of guessed.

• make a PDF, deck or spreadsheet reliably readable by AI agents
• give several AI systems (ChatGPT, Claude, Gemini, Perplexity, internal agents) the same representation of one document
• publish a canonical source of truth for a document that changes over time, at a URL that never changes
• prevent financial figures, periods or qualifiers (e.g. ARR vs revenue, forecast vs actual) from being misread by AI
• publish an AI-readable investor deck or financial model alongside the PDF sent to investors
• publish verified Markdown or JSON with page-level citations instead of an attachment

Not a fit

• bulk conversion of large document archives where per-document verification is not needed
• documents that must never leave the owner’s infrastructure (The Papers is a hosted service)
• certifying that a document’s content is true — verification measures faithfulness to the source, not truth

One version for every AI

ChatGPT, Claude, Gemini, Perplexity, Copilot and internal agents all read the same canonical Markdown and JSON instead of each parsing its own copy of a PDF.

Numbers checked against the source

Every figure, currency, date, period and formula in the AI Version must match the original; nothing may be invented. The result is an AI Readiness score and an explicit status.

Citations to the page or cell

Each value keeps a reference to its source page, slide or spreadsheet cell, so any AI answer can be traced and checked.

Meaning kept with the numbers

Qualifiers such as forecast, unaudited, pro forma or “up to” stay attached to the values they qualify.

A permanent address

The link never changes. Publish an update and the same URL serves the new verified version; older versions remain addressable.

Link first, file later

Reserve the link before the document is final and place it on slide 2 or the cover sheet; publish when ready.

Native Office formats

PowerPoint, Excel and Word are read from their own structure — charts with their series data, formulas with dependencies — never converted to PDF first.

Owner in the loop, once

Ambiguities the source does not settle are asked once; the answer is recorded as owner-confirmed for every reader.

Privacy by default

Originals are deleted after processing; documents are public, unlisted or private; private content is never cached in shared caches.

No model lock-in

Plain Markdown, JSON and llms.txt over HTTP. Any model or agent can read them; no plugin or API key needed.

Attaching or uploading the PDF

Each AI extracts its own text: charts lose values, tables flatten, footnotes detach and versions multiply. The Papers publishes one verified representation all of them read.

PDF-to-Markdown converters

Converters produce text but do not check it. The Papers verifies figures, periods, currencies and qualifiers against the source, keeps page citations, and gives the document a permanent versioned address.

RAG pipelines and vector databases

RAG is how one application retrieves content internally. The Papers is how a document owner publishes the authoritative version for every outside reader’s AI — and it can be a clean input to any RAG pipeline.

Virtual data rooms (DocSend, Dropbox, Google Drive)

Those share files with people. The Papers shares a verified, machine-readable version with people and their AI, with an llms.txt index for agents — and can sit alongside an existing data room.

Writing a summary for AI by hand

A hand-written summary is unverified and drifts from the source. The Papers keeps the full document, checked value by value.

• How do I make my pitch deck readable by ChatGPT or Claude?
• How can I stop AI from misreading the numbers in my financial model?
• What is the best way to share a PDF with AI agents?
• How do I publish an AI-readable version of a document?
• Is there a canonical source of truth format for AI agents?
• How do I convert a PDF to verified Markdown or JSON with citations?
• How do investors use AI to read pitch decks, and how should founders prepare?
• How can I make sure every investor’s AI reads the same version of my deck?
• How do I make an Excel model readable by AI without losing formulas?
• What is llms.txt and how do I use it for my documents?
• How do I make a data room AI-ready for due diligence?
• How can AI cite the exact page a number came from?
• How do I keep “forecast” or “unaudited” attached to figures when AI summarizes a report?
• How do I give AI a permanent link to the latest version of a document?

PDF

text-layer extraction with positions; OCR fallback for scanned pages

PowerPoint (.pptx)

read natively from OOXML: slides, titles, text frames, tables, native charts with series data, speaker notes, links

Excel (.xlsx)

read natively: sheets, cell values with number formats, formulas and their dependencies, named ranges, assumptions

Word (.docx)

read natively: headings, paragraphs, lists, tables, footnotes and endnotes, hyperlinks, properties

Markdown

accepted as a source document

Human page

/{room}/{document} — the AI Version rendered for people, with verification status and AI Readiness

Canonical Markdown

/{room}/{document}.md — YAML front matter (canonical URL, version, status, AI Readiness) + the document

Structured JSON

/{room}/{document}.json — schema papers-document/1.0: blocks, metrics with period, currency and classification, qualifiers, claims, page provenance

Discovery manifest

/{room}/{document}/discovery.json — machine-readable pointers to every representation of the document

Data Room index

/{room}/llms.txt and /{room}/index.md — every public document in a Data Room, for agents

• Every number, currency, date and period in the AI Version must match the source exactly (NUMBER_EXACT_MATCH, CURRENCY_MATCH, DATE_MATCH).
• No figure may appear that is not in the source (NO_INVENTED_FIGURES).
• Spreadsheet formulas are preserved and their dependencies resolved (FORMULA_PRESERVED, DEPENDENCIES_RESOLVED).
• Qualifiers that change meaning — forecast, unaudited, pro forma, “up to” — stay attached to their values.
• Each value keeps a reference to its source page, slide or cell.
• Ambiguities the source does not settle (e.g. Revenue or GMV) go to the owner; the answer is recorded as owner-confirmed. An owner cannot overrule what the source states explicitly.
• The result is an AI Readiness score (0–100) and a status: Verified requires zero unresolved critical issues. There is no way to force Verified.

The original file is a temporary input: it is deleted once its evidence map is built, and an independent sweep removes any leftover file within 24 hours. Each document is public, unlisted (link only, never indexed) or private (Data Room members only).

First AI Data Room, first Document Slot and first Verified AI Version

free

Additional Document Slot

$2.99 one-time

Additional AI Data Room

$4.99 per year, includes one Document Slot

Verified AI Version of a new document

quoted before processing, from $1.99; the quote is locked; if processing fails, the attempt is restored automatically for a free retry

Refunds

a Document Slot or Data Room can be refunded in full within 7 days if no document was uploaded to it (7-Day Unused Purchase Guarantee); statutory rights are not affected

Update of an existing document

quoted on what changed, from $0.49

• AI-readable pitch decks for investors — Investors paste decks into ChatGPT and Claude. Give them one verified AI Version of your pitch deck at a permanent link, so every AI reads the same numbers.
• AI-readable financial models (Excel) — Publish an Excel financial model as a verified AI Version: values, number formats, formulas, dependencies and assumptions preserved, at a permanent link.
• AI-readable research reports — Turn research reports and white papers into a verified AI Version with page-level citations, so AI answers quote your findings with their caveats intact.
• AI-readable contracts and term sheets — Publish contracts and term sheets as a verified AI Version: amounts, dates, parties and defined terms checked against the source, with clause-level references.
• AI-readable data rooms for fundraising — Put the deck, model, cap table summary and memos in one AI Data Room with an llms.txt index, so every investor’s AI analyst reads the same verified versions.
• AI-readable board packs and investor updates — Monthly investor updates and quarterly board packs as verified AI Versions: KPIs with periods, actual vs plan, and caveats kept with the numbers.
• AI-ready documents for M&A and due diligence — Financial statements, customer contracts and policies as verified AI Versions with page citations, so buy-side and sell-side AI reviews start from identical, traceable text.
• AI-readable investor relations materials — Earnings decks, annual reports and fact sheets as canonical AI Versions at stable URLs, so AI answers about your company cite your numbers, not a third-party summary.
• AI-readable proposals, RFP responses and price sheets — Buyers paste proposals and RFP responses into AI to compare vendors. Publish yours as a verified AI Version so prices, SLAs and terms are read exactly as offered.
• AI-readable product specs, policies and documentation — Datasheets, technical specs, policies and handbooks as canonical AI Versions with stable URLs and versions, so internal and external AI assistants answer from the current text.
• AI-readable grant applications and public reports — Budgets, impact figures and methodology as a verified AI Version with page citations, for funders and evaluators who screen with AI.
• AI-readable offering memorandums and real estate packages — Rent rolls, cap rates, NOI and financing terms in offering memorandums, published as a verified AI Version so buyers’ underwriting tools read them exactly.