Judge a B2B ChatGPT Ads test by whether inquiries become relevant sales conversations and commercially worthwhile opportunities. Keep form submissions, qualified leads and won business as separate measures. A low cost per submission can coexist with expensive sales work and little usable pipeline.
Reviewed October 15, 2026. GPTAd's scorecard below is an independent business-evaluation method. The numbers are hypothetical, not client results or platform benchmarks. Internal CRM stages do not automatically become supported optimization events.
Define the business stages before scoring
Start with a short agreement between marketing and sales. An inquiry is a completed request for contact. A valid inquiry has a usable contact route and is not an identified duplicate or spam record. A qualified inquiry meets your written commercial criteria. An opportunity has passed your separate sales acceptance process.
Choose criteria tied to the offer: a relevant business problem, a serviceable organization, an appropriate purchasing process and a plausible implementation timetable. Avoid personal characteristics that do not belong in the decision. Record concise reason codes in the authorized CRM rather than copying personal conversation details into shared marketing sheets.
Appoint a sales reviewer and a backup. A scorecard left unfilled is missing evidence, not proof of poor-quality traffic. Make response timing visible so the team can distinguish an unserviceable inquiry from one nobody contacted.
Map the real action to the documented event
OpenAI's supported-events reference distinguishes lead_created for a submitted lead form or contact request from appointment_scheduled for a booked meeting, demo or consultation. A form submission is not evidence that a meeting took place or that sales accepted an opportunity.
Write an event contract in plain language: the exact completed action, the authoritative system, the occurrence identifier and the intended event type. Apply the conversion-event and deduplication checklist before relying on event totals.
OpenAI's current conversion-optimization guidance requires supported standard events and says custom events are not supported for that objective. It also describes one selected event per campaign. A custom internal score or CRM label therefore must not be advertised as a natively supported optimization target without separate documentation and account verification.
Choose a supported action you can reliably observe. Maintain richer commercial evaluation in your own reporting without claiming undocumented offline-stage imports or automatic optimization toward won revenue.
Use this hypothetical lead-quality scorecard
Imagine a fictional implementation consultancy evaluating a single offer. It decides to classify inquiries using these business-defined fields:
- Cohort: week the inquiry was first accepted, plus campaign identifier where reliably available.
- Validity: valid, duplicate, spam or unresolved; preserve the original record count separately.
- Need fit: confirmed relevant problem, unclear problem or outside scope.
- Service fit: the organization can be served under the consultancy's published scope, or it cannot.
- Buying context: appropriate contact or referral path, unknown, or no current purchase process.
- Timing: near-term project, later project or unknown; the exact periods are defined before review.
- Follow-up: first attempted contact date, first successful contact date and unresolved attempts.
- Sales decision: accepted opportunity, nurture, disqualified or awaiting review, with a concise reason.
- Outcome: open, won, lost or withdrawn, with the date of the latest update.
Rather than inventing a universal numerical score, this consultancy requires confirmed need fit, service fit and an accepted next step before labeling an inquiry qualified. Another business can choose different rules, but it should apply them consistently across the channels it compares.
Reconcile a cohort with honest denominators
Suppose a hypothetical cohort contains 40 submitted inquiries, of which eight are duplicates or spam. Sales has reviewed 24 of the remaining 32 valid inquiries and classified nine as qualified. Eight valid inquiries are still awaiting review.
The observed qualified share among reviewed valid inquiries is 9 divided by 24, or 37.5%. Across all valid inquiries, nine are confirmed qualified so far: 28.1%. Neither figure estimates the final qualification rate without assumptions about the unresolved group.
If the associated media spend were USD 1,600, cost per submission would be USD 40 and cost per confirmed qualified inquiry to date would be approximately USD 177.78. Those are arithmetic examples only. They exclude service fees and sales costs, and they do not establish that advertising caused each inquiry.
Show the unresolved count alongside the cost. Comparing this immature cohort with an older cohort that has completed review would conceal a material difference.
Preserve time and reporting definitions
Review cohorts at comparable ages, chosen around the business's actual sales process. Save a snapshot at each review so later wins do not silently rewrite what the team knew when it made a budget decision.
OpenAI's reporting documentation supports delivery, cost and attributed conversion reporting with explicit reporting periods. Save the account, dates, time zone and report definitions with each export. Use the attribution reconciliation workflow when business records and platform totals disagree.
Do not force a match by deleting awkward records. Record why the sets differ: duplicate handling, unmatched source, different date boundaries, unresolved sales review or a different conversion definition.
Decide what the next investment would learn
Continue only when the measurement is understandable, sales follow-up is adequate and the observed opportunities justify further learning under your approved economics. If relevance is weak, identify one offer or destination change. If review is incomplete, fix the handoff before claiming the channel failed.
A useful conclusion may be “the cohort is still immature.” Name the missing evidence and the next review date. The goal is a defensible commercial decision, not a flattering conversion total.
References
- Supported Events — OpenAI; reviewed 2026-10-15.
- Conversion-optimized Campaigns — OpenAI; reviewed 2026-10-15.
- Reporting — OpenAI; reviewed 2026-10-15.
