You have launched a ChatGPT Ads campaign. The dashboard shows impressions, clicks and spend. The useful question is not only “is it running?” but who reaches the site, what happens after the click and how many people become real opportunities?
A report may be full of numbers yet lead to the wrong decision. A high click-through rate does not show that enquiries are a good fit; cheap clicks do not make a campaign profitable if they do not become customers. This guide follows the path from ad to commercial outcome. If you are still assessing the channel itself, start with our introduction to ChatGPT Ads in Italy.
In short: four levels of reporting
Keep these four levels apart when evaluating a campaign:
| Level | What to check | Question |
|---|---|---|
| Delivery | Impressions, spend, campaign and period | Is the ad being shown in the intended context? |
| Interest | Clicks, CTR and CPC | Does the message bring people to the page? |
| Action | Observed visits, completed forms, confirmed purchases | Can visitors complete the intended action? |
| Quality | Qualified leads, opportunities, valid orders and value | Is the outcome worth its cost? |
Do not jump from impressions to estimated revenue. Each step has its own tools and definitions. OpenAI’s Insights documentation describes delivery and cost metrics, attributed conversions, CPA and, for correctly tracked purchases, attributed sales value and ROAS. Check the metrics available to your account and their definitions in the current report and official documentation.
Define what counts before comparing numbers
For a B2B campaign, an enquiry may be the primary goal, but “lead” needs an operational definition. Did the server accept the form? Did the contact meet minimum requirements? Did a sales conversation begin? These are three different events.
For an online store, starting a basket suggests interest; a confirmed order represents a different outcome. When a dashboard combines different actions under “conversions”, an increase can hide a decline in the actions that matter most.
Before the test, write down:
- the primary conversion;
- secondary events used to diagnose the journey;
- the source that confirms the result, such as the site, CRM or order system;
- the maximum sustainable cost for a valid enquiry or order;
- a reporting period appropriate for the sales cycle.
For event implementation, E-ROE’s guide to the OpenAI Pixel and Conversions API explains when to send an event and how to avoid false or duplicate counts. Here the focus is interpreting the outcome.
Read impressions, CTR and CPC without making them the goal
Impressions count ad displays, not necessarily distinct people. CTR is clicks divided by impressions. CPC is average cost per click according to the finalised values in the advertising system.
A higher CTR may mean a more relevant message. It may also mean a promise so broad that it attracts the wrong people. Check what happens after the click: the landing page’s heading, offer, conditions and next step should match what the ad led people to expect. Our ChatGPT Ads landing page guide covers this connection.
Compare equivalent periods and do not judge the current day too early. OpenAI notes that impressions and clicks can appear before costs have been finalised, while attributed conversions can be updated later. Use complete dates, consistent attribution settings and the report’s own cost metrics rather than dividing figures at different processing stages.
From click to enquiry: where does the journey break?
If ads receive clicks but no confirmed enquiries, do not rewrite the ad immediately. Check in order:
- Arrival: does the URL work, load well on mobile and retain useful parameters?
- Continuity: does the top of the page deliver the ad’s promise?
- Understanding: are the offer, limits and next step clear?
- Action: do the form or checkout work through confirmation?
- Measurement: does the event fire only when the action succeeds?
- Quality: do enquiries meet the commercial criteria?
A phone or messaging click suggests interest, but it does not prove that a conversation took place. Track confirmed enquiries separately from preliminary interactions.
Why Ads Manager, analytics and CRM show different totals
The systems answer different questions. Ads Manager attributes outcomes to ads under its rules; site analytics describes observed sessions and events; the CRM records a contact’s commercial status. Identical totals are not an appropriate expectation.
For a meaningful comparison, record alongside the numbers:
- date range and time zone;
- which date an attributed conversion is reported against;
- attribution type and window;
- the event configured as a campaign goal;
- consent status and measurement coverage;
- how recently each report was updated.
In OpenAI’s documentation, general Insights report click-through conversions associated with configured goals. A dedicated conversion report can distinguish click-through and view-through results according to selected windows. If you place those totals side by side without checking the settings, a difference is not automatically a tracking error. Recheck current definitions before changing the campaign.
A weekly report that supports a decision
One page of well-defined figures can be more useful than several dashboards if every line leads to an action. Structure a weekly check like this:
| Question | Figure to record | Possible next step |
|---|---|---|
| Is the campaign delivering? | Impressions, spend, status and change | Check settings or budget if delivery is abnormal |
| Does the ad create interest? | Clicks and CTR by proposition | Review the message or its fit with intent |
| Does the page work? | Visits, errors and completed forms | Repair the page or conversion path |
| Can the data be trusted? | Confirmed events and differences between systems | Fix measurement before optimisation |
| Are contacts useful? | Qualified leads and rejection reasons | Clarify the promise, offer or qualification |
| Is the cost sustainable? | Spend per valid result and actual value | Continue, change or end the test |
No universal CTR or CPA threshold can decide this for every company. Sustainable cost depends on margin, close rate, sales cycle and the ability to handle enquiries. A useful report defines every metric and identifies who will make the next decision.
An illustrative calculation
Suppose a campaign spends €600 and produces 30 confirmed enquiries. Cost per enquiry is €20. If only 6 enquiries meet the commercial criteria, cost per qualified lead is €100. The first figure describes form efficiency; the second is closer to the business decision. Without knowing how many opportunities close and what they are worth, neither proves a return. These are hypothetical numbers, not E-ROE campaign results.
Three common diagnoses
Many impressions, few clicks
Check whether the ad gives people a clear reason to continue and whether the context is relevant. Do not change message, destination and audience at once; you would not know which change mattered.
Many clicks, few confirmed actions
First check the URL, mobile loading, page continuity and form. Then examine event tracking. A successful but untracked enquiry and a failed submission counted as a lead call for opposite fixes.
Reported conversions, but few suitable customers
Compare the ad’s promise with your eligibility criteria, page content and CRM rejection reasons. A qualifying question or clearer scope may reduce total leads while improving useful ones.
When to decide whether to continue
Set the evidence needed for a decision before the test. Too short a period may reflect random fluctuation; waiting indefinitely spends budget without learning. At the end of the chosen period, write a specific conclusion: which hypothesis was tested, which step worked, what remains unknown and which single change comes next.
If measurement is incomplete, the right conclusion may be “we do not yet have reliable data”, rather than “the channel does not work”. If confirmed contacts repeatedly fall outside the target, ending the test and revising the offer may be reasonable.
Frequently asked questions
Which KPIs matter most for ChatGPT Ads?
It depends on the goal. Impressions, clicks and cost describe delivery; confirmed conversions and commercial quality help evaluate outcomes. For ecommerce, add valid orders and value. For B2B, track qualified leads and opportunities.
Why do platform clicks and website visits differ?
The systems observe different events and may use different filters, consent settings, periods and time zones. Check the URL, parameters, loading and analytics before assigning the difference to one cause.
Can I calculate ROAS for a campaign that generates leads?
Not reliably from form counts alone. To calculate a return, connect contacts to actual sales value and use consistent costs. Label an assumed value as an estimate.
How often should I check the report?
You may monitor delivery more frequently, but judge cost and conversions over complete, comparable periods. The sales cycle can take longer than the click data to reveal its outcome.
Turn the report into the next decision
A campaign improves when the data helps you choose whether to refine the message, repair the page, fix an event or change the offer’s scope. If you want a campaign with a coherent goal, landing page and measurement plan, explore E-ROE’s ChatGPT advertising service and tell us about your project.
Technical source
- OpenAI Ads — reporting, metrics, attribution and data freshness, checked 30 September 2026. Platform features and definitions can change.
