Guide
How to make an earnings report or investor update readable by AI
Publish the numbers as text and real tables, each with its period, currency, basis (GAAP, non-GAAP, pro forma) and status (actual, guidance, unaudited), and keep definitions and reconciliations next to the figures. Then publish one fidelity-checked, AI-readable version at a permanent link, so analysts’ screeners and assistants read checked figures instead of re-extracting them from slides and PDFs.
How your report is read today
Analysts, investors and their tools increasingly run earnings releases, decks and shareholder letters through AI before a human reads them. The screener extracts headline figures, compares them with the prior period and with guidance, and summarizes the story.
Errors compound quickly: a non-GAAP figure taken as GAAP, a quarterly number compared with an annual one, guidance read as a result, or a figure from a chart image misread by a digit. Once that summary circulates, the correction rarely catches up.
Label every number completely
- Period: “Q3 FY2026 (three months ended 30 September 2026)”, not “the quarter”.
- Currency and unit: “USD millions” in every table header.
- Basis: GAAP, non-GAAP, adjusted, pro forma, constant currency — stated next to the figure.
- Status: actual, unaudited, preliminary, guidance, outlook range.
- Comparison: “up 12% year over year” should say against which period and on which basis.
Tables, charts and reconciliations
| Element | What to do |
|---|---|
| Headline metrics | A table in text, not only styled numbers on a slide |
| Charts | Data labels or the underlying table next to the chart |
| Non-GAAP measures | Reconciliation table in the same document, near the figure |
| Guidance | Separate section, labeled as forward-looking, with the range and period |
| KPIs with custom definitions | Definition stated once, used consistently across quarters |
For startups sending investor updates
- Use the same metric names and definitions every month or quarter.
- Put a short metrics block at the top: MRR or ARR, burn, runway, headcount — each with its date.
- Mark anything unaudited or estimated.
- Send a link rather than an attachment so readers always open the latest version, and keep earlier updates in one place.
Publishing one canonical version
The release, the deck, the letter and the transcript all contain the same numbers, and each reader’s tool extracts them independently. A canonical AI Version gives every tool the same checked figures.
The Papers turns each document into an AI Version at a permanent URL, with Markdown and JSON in which every metric carries its period, currency and classification, plus qualifiers such as forecast, unaudited or pro forma. Every value must match the source exactly, and each one links back to the page, slide or cell it came from. A Data Room index (llms.txt) lets AI tools find all your reports in one place, and analytics show which AI agents and crawlers read them.
Public documents can be listed and indexed; unlisted ones are reachable only by link. With a custom domain, the room can live at an address such as ai.yourcompany.com.
FAQ
- Does this replace our regulatory filings?
- No. Filings remain the official record. An AI Version is a faithful, machine-readable copy of documents you already publish.
- How do we handle guidance so AI does not report it as a result?
- Put it in its own section, labeled as forward-looking, with period and range. The Papers keeps qualifiers such as forecast attached to each figure.
- What happens when we restate a figure?
- Update the document; the link stays the same and the version history shows what changed.
- Can we see whether AI tools actually read the report?
- With The Papers, analytics show reads by people and by AI readers — which agent or crawler, and in which format.