Two ChatGPT advertising dashboards show the same 24 outcomes. One reports a conversion rate of 3%, the other 4%. Both formulas may be arithmetically correct. The difference can be that one divides by 800 clicks and the other by 600 website sessions. Until the denominator is named, the percentages cannot be meaningfully compared.
A denominator defines the population or opportunity behind a rate. For an advertiser, this is a decision about what the number means, not merely a spreadsheet detail. The wrong denominator can make a campaign look more efficient without changing a single customer outcome.
Write the fraction in words
Replace a dashboard label with a sentence: “Of these specific opportunities, this share produced these specific outcomes.” For click-through rate, the opportunities are impressions and the outcomes are clicks under the reporting definition. For a website session conversion rate, the opportunities are sessions in the selected analytics scope.
Those populations are not interchangeable. A click and a session are different observations. Even if their counts happen to match in one period, that does not establish that they represent the same set. Avoid naming both measures simply conversion rate in the same review.
Use a display label that includes the denominator when ambiguity is likely: outcomes per reported click, completed forms per measured session, or accepted leads per submitted lead. Longer labels are acceptable when they prevent an incorrect budget comparison.
Check membership, not just arithmetic
Return to the hypothetical 24 outcomes. If they belong to a mature click cohort while the 800 clicks include a new campaign with no time to convert, the ratio blends different observation opportunities. If the 600 sessions exclude visitors without analytics consent while the outcomes include a broader set, that ratio also needs qualification.
The point is not that one denominator is always superior. It is that numerator and denominator should answer the same question over a coherent scope. Record account, campaign set, time basis, outcome definition and eligibility rules for both sides before calculating.
If the scopes cannot be aligned, use separate descriptive figures instead of forcing a ratio. Showing reported ad clicks alongside observed website outcomes can still be useful. A precise percentage with incoherent membership is less informative than two honest counts.
Same outcomes, different ratios
- Per click
24 / 800 clicks = 3%.
- Per session
24 / 600 sessions = 4%.
- Name the question
Verify outcomes belong to each denominator before comparing.
Do not use a subset as if it were the whole
A dashboard filter may remove low-volume campaigns from the denominator while retaining outcomes from the account total. That inflates the rate. The reverse mismatch can depress it. Whenever filters change, verify that both sides of each derived ratio inherit the intended scope.
Watch for joins that multiply only one operand. If a campaign’s spend repeats across several descriptive rows but its outcomes are deduplicated, cost per outcome becomes inflated. A denominator investigation may therefore reveal a data-model problem rather than an advertising problem.
Use the metric dictionary to specify scope and the weighted CTR method for combining rates. Never average row percentages merely because the dashboard component accepts a numeric field. Aggregation should follow the measure’s meaning.
Interpret zero and small denominators explicitly
A zero denominator does not support a conventional finite ratio. Mark the result unavailable or not applicable with an explanation appropriate to the metric. Returning zero conceals that no eligible opportunities were counted, while an enormous replacement number creates a false performance signal.
Small denominators are calculable but unstable. One outcome from two clicks gives 50%; one from two hundred gives 0.5%. Neither percentage alone tells the reader how much evidence is available. Display counts beside rates in low-volume campaign reviews.
Do not impose an invented minimum sample threshold as a universal rule for ChatGPT Ads. The evidence needed depends on the decision, variability and acceptable uncertainty. A descriptive monitoring table and a budget-changing experiment have different requirements.
Apply the same discipline to money
Cost per outcome requires a defined cost numerator as well as an outcome denominator. Media spend alone differs from media plus agency work or internal sales handling. If one channel includes these costs and another excludes them, their acquisition ratios do not support a fair ranking.
Similarly, revenue divided by spend describes a different relationship from contribution divided by total acquisition cost. Keep financial ratios named according to their ingredients. A familiar acronym should not disguise a change in the economic question.
For a ChatGPT review, trace each important ratio to its two source measures before publication. Ask a second reviewer to explain the fraction without seeing the formula. If the explanation differs from the intended definition, revise the label or calculation.
OpenAI’s reporting documentation defines the platform’s returned metrics. Derived dashboard measures remain your responsibility. Click-cohort comparisons address time eligibility specifically. The result of this review should be a smaller set of interpretable rates, each tied to a population that makes sense for the advertising decision.
Sources and scope
Choose coherent eligible populations for derived campaign rates and prevent misleading ratios assembled from unrelated reports.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.