CSV, Excel, and Google Sheets Ingestion

What gets ingested

• One row = one record.

• The first non-empty row is treated as the header (column names).

• Only visible (non-hidden) Google Sheets tabs are processed.

• For Google Sheets, the displayed (formatted) values are ingested, not the raw formulas.

• All content is processed as UTF-8 text.

Tip: Each row is converted to a simple, readable block for AI. Make headers self-explanatory because they become the labels seen by the AI:

{
Product name: Widget A
Price amount: 19.99
Price currency: USD
URL: https://example.com/widget-a
Tags: outdoor; waterproof
}

Best-practices checklist (quick)

• Use a single header row with clear, human-readable column names.

• Keep exactly one record per row; avoid merged cells.

• Make spreadsheet and sheet/tab names meaningful.

• Ensure consistent types in each column (all dates in one format, all prices as numbers, etc.).

• Prefer one column per concept (split amounts, currencies, statuses, units).

• Use visible sheets for data; hide or remove non-data sheets.

• Share/permission the file so the crawler account can view it.

Designing great column names

• Prefer self-explanatory, natural language headers:

  • Good: “Product name”, “Return policy”, “Support email”

  • Avoid: “PN”, “ret_pol”, “email1”

• Use a single header row (Row 1) with no blank cells.

• Avoid duplicate headers across the same sheet.

• Keep names short but specific. If needed, use “Title Case” or “snake_case”; just be consistent.

• Avoid punctuation that looks like data (e.g., “Price ($)”); instead use two columns “Price amount”, “Price currency”.

Layout rules

• No merged cells, multi-row headers, or pivot tables in the data area.

• Avoid blank top rows or columns; put headers on the first non-empty row, data starts immediately below.

• Do not place totals/subtotals within the data table; keep those on a separate sheet.

• Keep one dataset per sheet/tab. If you have different entities, use separate tabs.

• Name spreadsheet and sheet tabs descriptively; these names are surfaced in document titles.

Google Sheets specifics

• To ingest a specific sheet, copy the tab URL (it contains “#gid=…”). If you share the document’s root URL without “gid”, the first visible sheet is used by default.

• Hidden sheets are skipped. Hide any sheets you do not want ingested.

• Ensure the crawler/service ([e-mail]) account has at least Viewer access to the file.

• Avoid protected ranges that block read access to header or data cells.

Large files and performance

• Very large files can increase processing time and cost. Consider:

  • Splitting large datasets into multiple files or tabs (e.g., by time period or category).

  • Removing unused columns before upload.

  • Keeping only the latest, relevant rows if historical data is not needed.

Examples: good vs needs fixing

• Column names

  • Good: “Customer name”, “Order date”, “Price amount”, “Price currency”

  • Needs Fix: “CN”, “date?”, “$price”, “cur”

• Values

  • Good: “19.99” and “USD”

  • Needs Fix: “$19.99 USD” (mixes amount and currency into one cell)

• Dates

  • Good: “2025-08-26”

  • Needs Fix: “Aug 26th, ‘25” (ambiguous)

• Lists

  • Good: “outdoor; waterproof”

  • Needs Fix: “outdoor, waterproof / best” (multiple inconsistent separators)

Common mistakes to avoid

• Missing or unclear headers; duplicate column names.

• Multi-row or merged headers; blank header cells.

• Merged cells anywhere in the data area.

• Embedding totals/subtotals within the data table.

• Hyperlinks with display text instead of a plain “URL” column.

• Mixed data types in the same column (e.g., numbers and “N/A” and dates).

• Hidden sheets that you intended to ingest, or visible sheets you didn’t intend to ingest.

• Very wide tables with many unused columns; remove noise.

• Clear, descriptive headers and consistent values are the single biggest factor in good results.

• Use one dataset per visible sheet, keep layout simple, and avoid merged cells or complex formatting.