> For the complete documentation index, see [llms.txt](https://docs.reportpundit.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.reportpundit.com/knowledge-base/settings/mcp-connector/getting-accurate-results-from-large-reports-in-claude.md). # Getting accurate results from large reports in Claude Claude is good at analysing your report data. It is not a substitute for the report itself. This page explains where the line sits, so you can trust the answers you get and know which ones to double-check. None of this is complicated. It comes down to giving Claude a sensible amount of data and asking it to show its working. ### Why report size matters When Claude runs a report, it reads the columns and rows into your conversation. A conversation can hold a lot, but not an unlimited amount. A very large report can be cut short. That is the part worth understanding: a report that has been cut short still looks complete. Claude can total up the rows it received and give you a confident figure that is simply missing the rest. It is not wrong about the rows it has. It just does not have all of them. Everything below is about avoiding that situation rather than detecting it afterwards. > **TO CONFIRM:** the approximate row count at which results start getting truncated. A real number here, for example "reports under 5,000 rows analyse reliably," turns this whole page from advice into a rule. ### Eight practices that keep the answers right **1. Check the report before you run it.** Ask what fields and filters it uses. It takes a second and it stops you analysing the wrong report. > What fields and filters does my Sales by Product report use? **2. Narrow the date range.** A month of order line items analyses cleanly. Three years of them does not. If you need a long period, ask for a summary by month rather than every row. **3. Ask for aggregates, not row dumps.** "Total revenue and units by product for August, top 20" is both faster and more reliable than "show me the report." **4. Pull once, then keep asking.** The data stays in your conversation. Follow-up questions are answered from rows Claude already has, so they are instant and consistent with each other. Starting a new chat is what forces a re-run. **5. Ask Claude to show its working.** For any number you are going to act on: > How did you calculate that, and which columns did you use? Most surprises come from a date range, a filter, or a column that means something slightly different from what its name suggests. This question exposes all three. **6. Ask how many rows it got.** A quick sanity check when a total looks off: > How many rows did that report return, and what is the earliest and latest date in them? **7. Combine reports deliberately, not accidentally.** Two focused reports give better answers than one enormous one. See the pairings below. **8. Reconcile before you file or pay.** For tax, payouts and anything going to your accountant, use Claude for the analysis and your Report Pundit export as the record. Not because Claude is unreliable, but because the export is the artefact you can point at later. ### Two reports, better answers Some questions cannot be answered from a single report. Ask for both and Claude will run each in turn. **Sales plus inventory,** for reorder decisions: > Run my Sales by Product report for the last 60 days and my Inventory on Hand report, then tell me which products I should reorder this week. **Sales plus refunds,** for true profitability: > Run my Sales by Product and Refund reports for last quarter and rank products by net revenue after refunds. Each report on its own is small and analyses cleanly. Together they answer a question neither one can. ### When to export instead Claude is for analysis. It is not a data transfer tool. Export from Report Pundit when you need: * a full dataset for your accountant or bookkeeper * a file to load into a warehouse, BI tool or spreadsheet model * a permanent record of a period, for filing or audit * every row, because every row is the point Use Claude when you need to understand something, compare periods, spot an outlier or decide what to do. If the answer to "what will you do with this" is "read it," Claude is the right tool. If it is "store it" or "send it on," export. You can also have it both ways. Ask Claude to analyse the report and then produce a spreadsheet of just the part you care about: > Put the top 50 products into a spreadsheet with revenue, units and return rate, and give me a download. ### Figures worth double-checking Not because they are more likely to be wrong, but because the cost of being wrong is higher: * **Tax collected**, for any filing * **Payout reconciliation**, where you are chasing a difference * **Cost and margin figures**, which depend on cost data being current in your store * **Anything you are about to send to a third party** For these, run the report in Report Pundit as well and confirm the totals match. If they do not, ask Claude what date range and filters it used. That is nearly always where the difference is. ### A quick sanity checklist Before you act on a number from a conversation: * Do I know which report it came from? * Do I know the date range it used? * Do I know what the report filters out? * Does the row count look plausible for that period? * Has Claude shown me how it got the figure? Five questions, all answerable in one prompt: > Which report, what date range, what filters, how many rows, and how did you calculate that? ### Related articles * [How to pull a report into Claude](https://claude.ai/pull-saved-report-into-claude) * [How to run a report for a different date range in Claude](https://claude.ai/change-date-range-in-claude) * [Questions to ask Claude about your Shopify reports](https://claude.ai/claude-analysis-questions) * [Report Pundit and Claude: troubleshooting](https://claude.ai/claude-connector-troubleshooting)