ChatGPT Ads: new formats, lead measurement and availability in Italy

The novelty of ChatGPT Ads concerns a new visual experiment and tools to measure more events along the commercial path. In the announcement of 5 October 2026, OpenAI describes a format in image generation for Free and Go users, scheduled for testing in the United States later in the month. It is not an announcement of availability of the new format in Italy: the perimeter is indicated in the Official Communication.

For an Italian SME it is above all the evolution of measurement that deserves attention. A request-bearing campaign must also be evaluated after completing the form. Does contact become an opportunity? Do you answer to the salesman? Does it have a requirement compatible with the service? The number of leads, alone, does not tell the quality of the demand generated.

What already applies to Italy and what remains a US test?

The expansion of advertising self-service in Italy was announced earlier, in August, in the OpenAI European Communication. The October update should therefore not be told as the first arrival of advertising in the Italian market. At the same time, the existence of self-service does not prove that any later format can already be purchased in your account.

The starting point, from who sees the ads to how they are chosen, is in our guide to ChatGPT Ads. The steps to open an account and launch your first campaign are in the one on ChatGPT Ads Manager.

The new visual experiment has a specific geographic and product perimeter. Before preparing creativity or promising a plan to the customer, the manager must check what their account really allows to set up. This verification is distinct from the strategic analysis: an interesting format can remain outside the operational plan until there is an availability applicable to the Italian campaign.

It is advisable to keep in the brief a legible distinction between available, announced and to be verified. It prevents a commercial presentation from turning a roadmap into a buyable feature. For day-to-day work, it matters to be able to link a campaign that is actually active to events and results, while maintaining control over the data sent and the significance attributed to each event.

What events help to follow a request up to CRM?

The measurement innovations described by OpenAI include Pixels and CAPI for events such as lead, purchase, installation and registration, along with dedicated integrations. The possibilities must be verified in the official page on measurement. The presence of a technical event does not automatically define what it must mean for a company: a registration to the site and a request for a quote have different value.

The qualification can remain in the CRM if there is no appropriate event or if the project does not plan to transfer it to the advertising platform. The important thing is to be able to analyze it internally. The theme relates to the use of AI for CRM and Sales, where the quality of the information matters as much as the number of contacts collected.

How to define a qualified lead without changing rule every week?

The definition should precede the launch. For a B2B consultation it can include sector, problem size, availability to a comparison and compatibility with the area served. For a local service can count the area along with the type of request. It is not necessary to collect every possible data: we need those that allow the commercial to decide the next step in a coherent way.

Exclusions should be made explicit. Job requests, suppliers, internal testing and messages without a usable delivery should not enter the same indicator as potential customers. An unqualified request is not necessarily useless, but belongs to a different category. Without this separation, a campaign may look better simply because the team loosened the criteria during the observation period.

The rule must also be understandable to those who update the CRM. If the state depends on personal impressions, comparing two campaigns becomes difficult. A short tab with criteria and border cases allows you to read the outcomes in a stable way. Automatic classification can help, but ambiguous cases must maintain a human destination.

Where are consistent deduplication and identifiers needed?

The same request can cross browsers, servers, module and CRM. If each step creates an independent event without recognizing the original action, the report can count multiple conversions for one person. The project must define an internal event identifier and apply the deduplication rules provided by the tools used. It is not enough to see an event appear in the panel to consider the measurement correct.

Subsequent mailings also require attention. An update to the thank you page should not create a new lead, just as correcting a phone number does not represent a second acquisition. Trade states must be recorded in their order, without rewriting the past in an opaque way. The person responsible must be able to distinguish a new request from an update of the same practice.

One CRM Automation Well designed links these steps and makes errors visible. The system needs to know what to do when the CRM is unavailable: keep the request, try again without duplicating it and report the status. The capture process cannot depend solely on whether the pixel has received a notification.

How can a B2B service company evaluate requests?

Hypothetical scenario. A small industrial maintenance company promotes a scheduled verification service. The module collects type of plant, area served and delivery. The commercial distinguishes compatible requests, generic information and out of perimeter cases. The campaign is read with respect to actual opportunities, not just the compilations received.

In the pilot the team checks that the module questions do not create an unnecessary barrier, but neither do they eliminate essential information. If many contacts ask for an intervention that the company does not offer, the problem can be in the advertising message or on the landing page. Increasing the budget before correcting this inconsistency is likely to amplify an unsuitable demand.

Initial metrics are hypotheses to be measured: valid contacts, opportunities, appointments and reasons for exclusion. No returns from other advertisers constitute a forecast for this SME. A small but well-connected CRM campaign can provide a better decision than a larger volume observed only through cost per build.

Can an e-commerce read the value beyond the assigned order?

Second hypothetical scenario. An Italian e-commerce promotes products that require an informed choice. Some users buy immediately, others ask for clarification on compatibility or size. The team wants to distinguish orders, pre-purchase requests and contacts that become customers after a response. The CRM or assistance system preserves the path, without assuming that each interaction has the same value.

The report must also consider duplicate cancellations and requests. An order awarded is not necessarily an order confirmed according to the administrative criteria of the company. Definitions must be aligned between marketing, assistance and accounting. If each department uses a different version of the result, the yield becomes a discussion of the data rather than the business process.

The broader context on AI-assisted purchases is deepened in Shopify reports on AI purchases in Italy. For the individual campaign, however, the decision requires verified events in its own store. Market trends and advertising account results are distinct levels of evidence.

Do attribution and causal increments answer the same question?

Attribution assigns an outcome to an interaction according to defined rules and windows. Instead, the causal increase tries to understand what would have happened without advertising. An attributed conversion may include a person who would have purchased anyway; an unattributed conversion may have been influenced by the campaign. The report must state which question it is facing and with what limits.

Conversions after a view also require careful reading. They do not prove a causal relationship on their own. For an SME it can be useful to start from a coherent operational comparison, maintaining stable definitions and business process. If the volume allows, you can design a more robust verification, but without presenting a simple comparison as a definitive proof of increment.

The KPI of an AI project help choose indicators related to a decision. In the advertising case, knowing which contact has become an opportunity is often more useful than adding new metrics to an already difficult-to-read panel. Each indicator should have a possible owner and action when it changes.

Which data to send and which to keep in the business system?

The technical link must follow an explicit choice of the necessary data. Reconciling events does not require indiscriminately transferring business notes, attachments, or conversation content. The project must verify platform requirements, information and conditions applicable to the treatment, with those who follow these aspects for the company. The measurement should be designed together with the data flow.

Technical logs can use internal references instead of repeating personal data. A request for assistance with the platform should not include entire commercial archives when event identifiers and problem description are sufficient. Access permissions should be distinguished between those who manage campaigns and those who see complete customer information. This keeps the diagnosis possible without unnecessarily enlarging the circulation of data.

Pilot preparation includes a map: module, server, CRM and advertising tools. For each step, it indicates what is transmitted and what is preserved. A simple map also allows you to understand where a request can be lost or duplicated, linking quality of measurement and quality of customer experience.

How to prevent AI from turning labels into commercial results?

A classifier may suggest that a request is commercial, but may not automatically know if the contact has accepted a proposal. The CRM must record actual events and statuses updated by people or authorized systems. The OpenAI Decisions API it concerns structured choices; the result of the classification should not be confused with a verified opportunity.

Collaboration tools can also contribute to the transition between marketing and sales. Rovo, OpenAI, Jira and Confluence is about the working context of the team. In the advertising project, decisions, definitions and changes to the campaign must be traceable, so a new trader understands why a contact has been excluded or qualified.

The AI ROI Calculation Finally, it requires comprehensive costs and consistent results. No Italian CPC or CPM universal tariff has been verified here in the sources of measurement consulted. The plan must start from the actual conditions and a spending cap decided by the company, without importing benchmarks of partners as a promise of return.

The check before reading the first campaign report

Send an identifiable proof request and follow it from the form to the CRM. Check that it appears only once, that the tests are excluded from the results and that the trade status maintains the agreed meaning. Then check an incomplete request and an update of the same practice. The report is useful when telling recognizable business events, not just technical notifications received.

Before interpreting a change in performance, the manager should check whether the team's tracking, qualification criteria, or response times have changed. A campaign may look worse because the CRM no longer receives a status, or better because duplicates are back in the count. The change register helps to distinguish a change in demand from a change in measure.

We also need a responsibility for unfollowed contacts. If a valid request remains unanswered, the channel report cannot explain for itself why it has not become an opportunity. The process must show assignment, first contact, and closing pattern. This visibility makes it possible to improve the business path before asking the advertising platform to compensate for a loss that occurred within the company.

What decision to make after the first observation period?

The choice should not be reduced to continuing or stopping the channel. The team can discover a message to correct, a too-generic form, a slow business response, or an integration that loses updates. Each cause requires a different intervention. A seemingly good cost per lead can live with few opportunities, while a higher cost can accompany much more relevant requests.

The novelty of ChatGPT Ads becomes useful when the measurement follows the commercial work to a defined outcome. To link campaigns, forms and status of opportunities, an intervention on the Business Software can start from the verification of the events already available and the steps that remain disconnected today.