A BETTER WAY TO SHARE THE WORK
Turn a spreadsheet into an HTML report that explains the numbers
A spreadsheet can contain the answer without making it obvious. Give readers the finding, the comparison that supports it, and enough context to trust it.
New to creating with an agent? The request will guide it through the task. Bring your notes; you don’t need to write HTML.
How do I turn spreadsheet data into a useful HTML report?
Supply a checked data extract with column definitions, units, and the reporting period. Ask an agent to verify calculations before explaining the findings. Build a static HTML report that puts conclusions beside supporting numbers and makes denominators, missing data, and limitations visible.
01 / SEE THE DIFFERENCE
More signups, but a lower conversion rate
An illustrative two-period dataset uses the same definition of eligible visits and completed signups in both periods. The figures below are invented to demonstrate the calculation.
Period A: 1,000 eligible visits and 100 signups. Period B: 2,000 eligible visits and 160 signups. A headline says 'Signups increased 60%.'
- Volume
- Signups rose from 100 to 160: an increase of 60 signups, or 60%.
- Rate
- Signup conversion fell from 100 / 1,000 = 10% to 160 / 2,000 = 8%.
- Difference
- The rate decreased by 2 percentage points, which is a 20% relative decline from the original 10% rate.
- Interpretation
- The larger visit count coincided with more total signups, while a smaller share of eligible visits converted.
- Open question
- The totals alone do not establish why the rate changed. Comparable segment data would help investigate it.
02 / MAKE IT YOURSELF
Start with the evidence. Shape the page around the reader.
Name the question before making charts
Tell the agent what readers need to understand: whether conversion changed, where work is accumulating, or which category explains a cost increase. Avoid requesting a dashboard of everything. A focused question helps determine which comparisons are relevant and which attractive charts would only add noise.
Check the meaning of the numbers
Have the agent show calculations and identify missing definitions before writing conclusions. Confirm whether a blank means zero or unknown. Check denominators, date coverage, and duplicate rows. A mathematically correct percentage can still answer the wrong question if the underlying populations differ.
For campaigns, explain what the measurement counts
Keep platform-attributed conversions separate from verified customers or revenue unless your sources establish the link. Compare matching definitions and attribution settings; disclose differences you cannot reconcile. Recent results may be provisional because conversions and processing can arrive later. Fictional example: an export contains $1,200 spend and 40 attributed trial signups, with no paid-customer or revenue data and possible later conversions. The calculation is $1,200 / 40 = $30 per attributed signup—not $30 customer acquisition cost or proven return. Do not add overlapping platform totals and call them unique customers without a verified deduplication method.
Pair each finding with its evidence
Use a plain-language heading, the supporting values, and a short explanation of the comparison. Distinguish percentage-point change from relative percentage change. Do not label correlation as causation. If several explanations are possible, state what the data shows and what would require further investigation.
Create the reading view and verify it
Ask for static HTML with labeled charts or tables, units, the source date, and a method note. Use your saved style if available, with color highlighting the finding rather than decorating every category. Recalculate key figures independently and inspect mobile readability before approving publication.
Adapt this to your work
Keep the same approach. Change the inputs and review checks for the job you need.
A website analytics report
Compare complete periods with matching definitions. Identify the export time, timezone, filters and processing status. A key-event count is not automatically a count of people, new customers or revenue.
Fictional source notes: Fictional export: the last complete week has 1,000 sessions and 40 key events. The current week has 300 sessions and 8 key events, but includes only two days and is still processing. The timezone and event definition have not been supplied; no customer reconciliation exists.
A clearer result: Report the completed week's counts with the missing definitions flagged. Show the current period as a partial provisional snapshot, not a decline. Do not describe 40 events as 40 customers or compare the two unequal periods as a trend.
The matching instructions are included in the creation request below; your agent should use them only when relevant.
A budget variance explanation
Explain approved arithmetic and supplied drivers, not accounting treatment or financial advice. Define whether a positive number means actual minus budget and keep expense overruns distinct from favorable revenue movements.
Fictional source notes: Fictional approved September expense figures in USD on the same stated basis: budget $10,000, actual $11,200. A supplied supplier-cost explanation accounts for $800 of the difference. The remaining difference has not been reconciled. No cause or accounting adjustment is supplied for it.
A clearer result: Actual minus budget is +$1,200, or 12% above the expense budget. The supplied driver explains $800; $400 remains unresolved. Label it an expense overrun under this sign convention. Do not invent inflation, timing effects or an adjustment to make the explanation balance.
The matching instructions are included in the creation request below; your agent should use them only when relevant.
03 / TRY THIS PROMPT
Copy this into your agent. Add your source material.
Use the conversation where you already did the work, or start one with your notes. Replace the placeholders and supply any missing context; only attach information you are allowed to share with your agent.
Get the ready-to-copy creation request for guided questions, this task, and a draft review before anything is published.
Using my supplied spreadsheet data and definitions, create a report answering [audience question] for [period]. First check units, denominators, missing values, duplicate records, and comparability with the baseline. Show the key calculations for my review. Do not invent missing values or treat blanks as zero without evidence. For campaign data, preserve the exact conversion event, currency, attribution model and window, date basis, timezone, and export time. Ask for missing settings that affect the comparison. Separate platform-attributed results from deduplicated customers, booked revenue, and measured incremental impact. Do not sum overlapping platform counts as unique people. Mark provisional periods and do not invent lag adjustments, return, or causal lift. Lead with a small number of findings, each paired with supporting values and an appropriate labeled chart or table. Distinguish percentage points from relative changes and observations from causal explanations. Include the source date, exclusions, and a concise method note. Produce static HTML in my saved artifact style if available, without scripts, forms, or implied live refresh. Show me the draft and ask me to verify the figures, sensitive data, and recipient access. Do not publish or change access without separate explicit approval. For website analytics, preserve timezone, metric and key-event definitions, complete comparison periods, filters and export time. Label processing delays and incomplete periods. Do not infer customers from events or claim a trend from unequal or unreconciled periods. For budget variance communication, use approved figures on matching period, currency and accounting basis. State the sign convention and verify actual-minus-budget and its denominator. Keep supplied drivers separate from an unreconciled remainder; do not invent causes, recommend accounting treatment or provide financial advice.
04 / BEFORE YOU SEND
A polished page still needs your judgment.
- Have key totals, rates, and differences been independently recalculated from the supplied data?
- Do charts and tables display the correct units, periods, denominators, and meaningful axis labels?
- Are partial periods, excluded rows, missing values, and differences in coverage disclosed?
- Does the published view omit personal data and avoid promising live updates or unsupported causal conclusions?
- For campaigns, are attribution settings and data maturity visible, with attributed events separate from unique customers and causal impact?
When HTML fits
Use an HTML report when people need to understand findings without navigating cells, tabs, and formulas. Keep the workbook or approved extract as the calculation source and make the report a dated reading view.
When another format fits better
Keep a spreadsheet for editable models, input scenarios, and recalculation. This static workflow does not supply interactive filters, live data connections, or a replacement for the underlying analysis.
FROM DRAFT TO A LINK
Make it clear. Make it yours. Then share it.
Keep creating and revising in the agent you already use. With a compatible connection, share/artifacts publishes the reviewed HTML so you do not have to rebuild it in another editor. Ask your agent to use your saved artifact style for headings, color, and tone; review the result before sending it.
Choose who should be able to open the page and verify the returned link. For corrections to the same deliverable, update the existing artifact without changing its link. Publish a separate artifact when the old edition needs to remain available.
These are static pages, not live apps: no scripts, forms, or automatic data refresh. Source collection and scheduling depend on your agent and its available tools, not on share/artifacts.
Make this with my agentQuestions before you start
Can I use this for a marketing campaign report?
Yes. Supply authorized campaign exports and their measurement definitions. Ask your agent to explain spend, outcomes, comparisons, and limitations for the intended reader. Keep analytics tools as the measurement source and use the HTML page as a dated explanation—not a live dashboard or campaign-control tool.
Should I publish the whole spreadsheet as a table?
Only if readers need every row and the data is safe to share. A focused report usually benefits from selected evidence and an explanation. Keep detailed data separately when it is large, sensitive, or intended for further analysis.
What if the agent finds a surprising trend?
Check the calculation and data coverage first. Then test whether a changed denominator, segment mix, or incomplete period could explain it. Present a verified observation as an observation; do not turn an interesting pattern into an unsupported explanation.
Will the HTML report update when my spreadsheet changes?
Not automatically in this static workflow. Give the agent the new approved data, review the revised report, and update the published page. Display the data's effective date so recipients know which snapshot they are reading.