# AI readiness: how to check whether AI reads your document correctly

> A practical self-check to see whether ChatGPT, Claude or other AI tools read your document correctly, and how The Papers scores AI Readiness.

To check whether AI reads your document correctly, ask a few assistants for the key figures, periods and conditions and compare every answer with the source, line by line. A document is AI-ready when its numbers exist as text with their qualifiers attached, can be traced to a source page, and come back identical no matter which tool reads them.

## A 15-minute self-check

- List the ten facts that must never be wrong: key metrics, dates, amounts, parties, conditions.
- Upload the document to two or three different AI assistants in a fresh conversation.
- Ask for each fact with its period, currency and status: "What was revenue, for which period, actual or forecast?"
- Ask where each answer came from: "Which page or slide?"
- Mark every answer as correct, wrong, missing a qualifier, or invented.
- Repeat after any edit to the document; results change with layout.

## What usually fails

| Symptom | Typical cause | Fix in the source |
|---|---|---|
| Number missing | It exists only inside a chart or image | Add the value as text or a small table |
| Wrong period | Period is in a column header or axis far from the number | Write the period next to the figure |
| Forecast read as actual | Label is in a footnote or legend | Keep "forecast" or "unaudited" beside the value |
| Wrong scale | "in thousands" stated once at the top | Repeat the unit in table headers |
| Mixed terms | Revenue, sales and GMV used interchangeably | Use one term per metric |
| Scanned pages unread | No text layer | Export from the source app instead of scanning |

## Signals of an AI-ready document

- Text-based, not image-based; tables are real tables.
- Every metric has a period, currency and classification nearby.
- Qualifiers that change meaning ("up to", "pro forma", "estimated") sit next to the value.
- Headings make sections easy to locate and cite.
- One version, at one address, so no reader is working from an outdated copy.

## How The Papers scores AI Readiness

When you upload a document, The Papers builds its AI Version and checks it against the source: every number, currency, date and period must match exactly; no figure may appear that is not in the source; spreadsheet formulas are preserved with dependencies resolved; meaning-changing qualifiers stay attached; each value keeps a reference to its page, slide or cell.

Issues found along the way are reported, and ambiguities the source cannot settle, such as whether a figure is Revenue or GMV, are sent to you and recorded as owner-confirmed. The outcome is an AI Readiness score and a status. "Fidelity checked" requires zero unresolved critical issues and cannot be forced, so the status means something to the reader. The score and status also appear in the Markdown front matter, where agents can read them.

## Keep checking after you publish

Readiness is not a one-time event. Documents get updated, and readers keep using new tools. In The Papers, per-document analytics show who read the AI Version, people or AI, which agent or crawler, and in which format, so you can see whether readers reach the fidelity-checked version or are still working from attachments.

## FAQ

### Is a high AI Readiness score a guarantee AI answers correctly?

No tool can guarantee how another assistant answers. A fidelity-checked version removes the extraction step where most errors start and gives agents checked figures to quote.

### Can I force a document to "Fidelity checked"?

No. In The Papers, "Fidelity checked" requires zero unresolved critical issues; you resolve them, it cannot be overridden.

### Which file types can be scored?

PDF, PowerPoint, Excel, Word and Markdown. Scanned PDF pages are read with OCR as a fallback.

### How often should I re-run the self-check?

After every material edit and before sending the document to an important reader.

Run the self-check on your most important document, then upload it to see its AI Readiness. → https://thepapers.link/

## Related

- [Why does ChatGPT get the numbers in my PDF wrong?](https://thepapers.link/blog/why-chatgpt-gets-pdf-numbers-wrong.md)
- [How to make AI cite the exact page a number came from](https://thepapers.link/blog/make-ai-cite-the-source-page.md)
- [What is a canonical AI version of a document?](https://thepapers.link/blog/what-is-a-canonical-ai-version.md)
- [What is an AI Data Room?](https://thepapers.link/blog/what-is-an-ai-data-room.md)

Updated 2026-10-27. Canonical: https://thepapers.link/blog/ai-readiness-score
