Worked example with synthetic data
How to check an AI-written marketing report before anyone reads it
This page walks through a small campaign report built from four invented rows. You can see the table, the formulas, what the data cannot tell you, and a review checklist you can reuse with your own approved exports.
Synthetic training data. These numbers were made up for this exercise. They are not a client report and make no claim about real campaign performance.
The input: four rows, one week
All rows cover October 1 to 7, 2026 and use US dollars. A lead here is a counted inquiry, not a sale.
| Campaign | Impressions | Clicks | Spend (USD) | Leads | CTR | CPC | CPL |
|---|---|---|---|---|---|---|---|
| Brand search | 10,000 | 500 | $250 | 25 | 5.00% | $0.50 | $10.00 |
| Nonbrand search | 20,000 | 600 | $900 | 30 | 3.00% | $1.50 | $30.00 |
| Social prospecting | 30,000 | 300 | $600 | 12 | 1.00% | $2.00 | $50.00 |
| Social remarketing | 8,000 | 240 | $360 | 18 | 3.00% | $1.50 | $20.00 |
| Total | 68,000 | 1,640 | $2,110 | 85 | 2.41% | $1.29 | $24.82 |
The formulas
- CTR = clicks / impressions x 100
- CPC = spend / clicks
- CPL = spend / leads
Combined rates use the totals: 1,640 / 68,000 is 2.41%, $2,110 / 1,640 is $1.29, and $2,110 / 85 is $24.82. Keep full precision while calculating and round only what you display.
A sample report an AI tool might return
Period: October 1 to 7, 2026, USD, four campaigns.
Totals: 68,000 impressions, 1,640 clicks, $2,110 spend, 85 leads.
Observation 1: Brand search had the lowest cost per lead ($10.00) and social prospecting the highest ($50.00).
Observation 2: Click-through rate ranged from 1.00% (social prospecting) to 5.00% (brand search).
Observation 3: Nonbrand search took the largest share of spend ($900 of $2,110).
Hypothesis, not a finding: Brand search may be capturing demand that already existed.
Question for the owner: How were leads counted, and which were qualified?
What this data cannot tell you
- Revenue or profit. There is no sales data.
- Lead quality. A cheap lead may not be a good one.
- Cause. The file cannot show why one campaign differs from another.
- Trend. There is one week and no comparison period.
A report that recommends moving budget from this table alone has gone past the evidence.
Review checklist
- Confirm the reporting period (October 1 to 7, 2026), the currency (USD), and that there are exactly four unique campaign rows.
- Add the rows yourself: 68,000 impressions, 1,640 clicks, $2,110 spend, 85 leads.
- Recalculate combined CTR, CPC, and CPL from the totals, not from the row rates.
- Open the actual exported document and inspect its tables, formulas, and links, not just the chat answer.
- Test a copy of the data with zero leads. Cost per lead should show as unavailable, not as zero.
- Check that observations are separated from hypotheses and that no sentence claims revenue, lead quality, or cause.
- Record who reviewed it, what was corrected, what is still uncertain, and who may see the report.
A reusable assignment
Give the assignment and the CSV to the AI tool you are practicing with, then compare what it returns with the checked output. Swap in your own approved export when you are ready. The files are plain text so you can open them anywhere.
Questions about this example
Is this real campaign data?
No. The four rows are invented for training. They do not describe a client, a real account, or a result anyone achieved.
Why is the combined CTR 2.41% and not 3.00%?
Because combined rates use combined totals: 1,640 clicks divided by 68,000 impressions. Averaging the four row rates (5, 3, 1, and 3 percent) gives 3.00%, which gives the small campaigns too much weight.
Does the lowest cost per lead mean the best campaign?
Not from this data. The file has no revenue, lead quality, or attribution information, so it cannot show which leads were valuable or what caused them.
Practice this with your own team
Evan can run this kind of exercise live, using approved exports from your team. Explore training for marketing managers, training for marketing agencies, or ChatGPT Work training. For a group, request a custom corporate quote. Read the guide to reviewing AI-generated work for the wider habit.