Before the next ChatGPT campaign review, ask three people what “conversion” means. The media buyer may mean the campaign’s configured goal, the sales manager may mean a qualified opportunity and the finance lead may mean a paid invoice. A single dashboard column cannot safely represent all three without explanation.
A metric dictionary is the small agreement that prevents this ambiguity from becoming a budget argument. It connects each label used in the ChatGPT advertising report to a specific definition, source, scope and decision. It is not a glossary of every advertising acronym. The useful version covers the measures your team actually uses.
Draft one entry together
Begin with the metric most likely to change a spending decision. For a lead-generation campaign, that might be cost per accepted lead. Write the business question first: how much campaign cost is associated with leads that meet the agreed qualification rule?
Then identify the numerator and denominator. The numerator could be campaign media spend for a defined interaction cohort. The denominator could be unique leads from that cohort accepted under a written sales rule. This is an internal business metric, not automatically the same as a platform CPA. Give it a distinct name so readers cannot confuse the two.
An illustrative entry might contain:
| Field | Example entry |
|---|---|
| Display label | Cost per accepted lead |
| Decision | Whether the current acquisition route fits our lead economics |
| Cost scope | Media spend only; agency work shown separately |
| Outcome rule | Unique leads accepted under qualification rule version 2 |
| Time convention | Defined cohort with an agreed observation interval |
| Owner | Named analyst, with sales approval of the qualification rule |
The example describes a proposed internal design. It does not state that ChatGPT reports this business metric directly or that AthillyAds imports qualification decisions from a CRM.
Separate platform fields from business measures
OpenAI’s reporting documentation is the source for the meaning of platform metrics. Your dictionary should preserve those definitions when quoting platform results. Derived measures need their own formulas and caveats, especially when they combine advertising data with finance or sales records.
For each entry, mark whether it is returned by the platform, calculated from returned fields or joined with another business system. This simple distinction helps reviewers find the right owner when a number changes. A platform field needs a source check; a derived metric may need a formula check; a joined measure also needs an identity and coverage check.
Do not silently rename a platform goal count “customers”. A goal can represent an action other than a completed new-customer sale. Likewise, do not label all reported purchase value “profit”. A familiar business word can introduce a stronger claim than the underlying event supports.
For ChatGPT reports, save the endpoint as part of the definition. The conversions field in general Insights is click-through, while the dedicated conversion endpoint can combine click and view goal outcomes under its selected windows. A field name alone is therefore insufficient. Link the entry to the relevant click and view definitions so a replacement export does not silently change the metric.
Three meanings of conversion
- Advertiser goal
The configured campaign action under the platform definition.
- Sales assessment
An accepted opportunity under a versioned rule.
- Finance outcome
Confirmed revenue needs its own linkage and scope.
Include the awkward cases
Every decision metric needs rules for missing values, zero denominators and late updates. For cost per accepted lead, what happens when no lead has yet completed qualification? Displaying zero cost would be misleading. A pending status with the observed media spend is more informative until the denominator is established.
Define whether duplicate leads, reopened opportunities and refunded purchases affect the measure. The rule should follow the business question rather than whichever treatment creates the best ratio. If two teams need different views, give those views different labels and explain their relationship.
Specify the unit and currency as well. A rate stored as a ratio and a rate displayed as a percentage can be numerically identical in meaning but very different in a spreadsheet. A money field needs a currency convention before several accounts can be combined. These details belong in the dictionary because they can change the decision.
Make definition changes visible
Suppose sales tightens its acceptance rule halfway through a ChatGPT campaign test. The accepted-lead rate may fall even when the incoming leads are unchanged. Keep a definition version and effective date so the report does not present this as an unexplained deterioration in advertising quality.
Decide whether earlier periods will be restated under the new rule. Either choice can be defensible if the underlying records support it and the report explains what changed. Mixing old and new rules within one trend without a note is the dangerous option.
Ask the metric owner to approve definition changes before they enter a client-facing report. This is a process recommendation for your team, not a claim about an approval feature in AthillyAds. A short shared document can be enough if it stays synchronized with the actual calculations.
Keep the agreement usable
Limit the first dictionary to the measures that appear in a real campaign review. Link each report label to its entry, then test the entries on one actual or safely anonymized reporting period. If the analyst cannot reproduce a metric from its definition, the entry needs more detail.
Use null handling for unavailable values and campaign identity joins for cross-system linkage. Consult OpenAI reporting for platform field meanings. The result should let two people calculate the same ChatGPT campaign metric independently and explain why it is appropriate for the decision.
Sources and scope
Agreeing the operational meaning, scope and owner of campaign metrics before they enter client reports.
Working methods and examples are editorial suggestions. Check current platform requirements and available features before implementation.