The targeting of ChatGPT Ads requires a change of perspective. Those coming from Google Ads tend to start with the keywords. Those coming from social curtains starting from interests and public segments. In ChatGPT the central question is different:
In what conversation can our offer be relevant, useful and credible while a person is making a decision?
OpenAI explains that the selection of an ad can consider signals such as the context and intent of the current conversation, the target page, title, text, creativity and targeting settings available.
This does not mean that the system automatically understands any company. The advertiser must turn its market into quite precise hypotheses to be tested.
Context Targeting does not mean buying the answer
Ads are labeled and separated by ChatGPT responses. Advertising should not be described as a way to change what the assistant responds or to get an organic recommendation.
The work of the advertiser is to make consistent:
- express need in conversation;
- proposed offer;
- title and text of the announcement;
- image;
- linked page;
- geographical area;
- final conversion.
When these elements tell different situations, targeting weakens. An ad can receive clicks for curiosity without generating useful requests.
From keywords to intent maps
The keywords remain useful as research, but they are not enough to represent a conversation. A map of intent adds depth.
Problem
What situation is he trying to solve?
Example: “We waste time copying data between sheets and management”.
Objective
What do you want to get?
Example: “Automate the update of orders without replacing all software”.
Criteria of choice
What conditions make a suitable solution?
Example: integration with the existing system, permission management, Italian assistance and gradual start.
Constraints
What can’t change?
Example: budget, time, mandatory tools, area served or industry regulations.
Decision stage
Is the person understanding the problem, comparing approaches or looking for a supplier already? Each phase requires a different message.
A practical example of contextual indication
A weak indication could be:
software companies
It is too wide: it does not describe need, recipient or criterion.
A more useful hypothesis can become:
Small Italian companies that are comparing solutions to reduce the manual insertion of data in orders, want to integrate existing tools and look for a gradual project.
It is not a guarantee of publication or a universal formula. It is a strategic hypothesis easier to connect to an ad, page and conversion.
Geographic targeting
Advertiser API documentation indicates targeting for country, region and DMA. The locations must be selected through the current identifiers provided by the platform. If a geography is not set, the campaign can reach all available locations.
For an Italian company you have to distinguish at least three cases.
National service
The campaign can target Italy, but the page and sales process must handle enquiries from across the country. Artificially adding city names does not create a genuine local service.
Regional or local service
The message must indicate the area served and the page must show consistent information: location, mode, times and territories covered.
Ecommerce
Targeting must follow the areas where the company sells and delivers. Time, shipping costs, currencies and availability must correspond.
Effective options may differ between interface, country and account. Before the launch, the locality catalog should be checked, not taken on the basis of a previous guide.
Personalized advertising and European limits
OpenAI documents custom audiences built from customer or prospect lists, but states an important limitation: custom audiences are not supported for campaigns targeting the European Economic Area or Switzerland, where personalised ads are not currently available.
For an Italian campaign, you should not build the plan by counting on functions that the documentation excludes for the area. This makes the quality of context, geography, message and page even more important.
Even in the markets where the public are available, only first-party data should be used for which adequate rights, information, consent or legal basis exist. OpenAI expressly prohibits the use of data purchased by intermediaries in the dedicated documentation.
How to separate ad groups
A group should be a legible hypothesis. You can separate for:
- main problem;
- phase of decision;
- product category;
- recipient;
- geographical area;
- decisive criterion;
- offer or conversion.
Avoid creating a group for each linguistic nuance: you would fragmente budget and data. It also prevents a single group that contains all the possible needs of the company.
A useful rule is to ask: if this group produces weak results, will we know what to change? If the answer is no, the structure is too confused.
The ad and page are part of targeting
Targeting does not end in campaign settings. The system and the person also read what promises the announcement and what shows the destination.
Check continuity:
| Element | Control application |
|---|---|
| Background | What decision are we intercepting? |
| Title | Does the person recognize himself without excessive promises? |
| Description | Does it add recipient, criterion or useful condition? |
| Image | Does it really represent product or service? |
| Page | Does the same reasoning continue? |
| Conversion | Measure a real result? |
If the announcement speaks of gradual integration and the page presents only a generic “digital innovation”, the perceived relevance collapses.
How to measure targeting quality
Impression, CTR and CPC describe only the beginning. To understand if you’re reaching the right conversations, measure:
- actual visits;
- depth and actions on the page;
- confirmed conversions;
- cost per conversion;
- percentage of contacts in the desired profile;
- reasons of discard;
- commercial opportunities;
- generated value.
Ask the sales department to classify contacts with few stable motivations: out of area, incompatible budget, wrong service, timing, unsuitable company. This data helps to understand whether to intervene on context, announcement or page qualification.
A four-step optimization process
- Formulates a hypothesis: problem, recipient, criterion and result.
- Mainten conversion stable: don’t change definition while comparing groups.
- Read the full path: from publication to commercial quality.
- Change a variable: context, message, creativity, geography or page.
With few data, widening and shrinking constantly prevents you from understanding what is happening. Break and change rules should be defined before launch.
Frequent errors in ChatGPT Ads targeting
- Treat the contextual directions as exact match guaranteed.
- Use such large descriptions to include almost every person.
- Enter a long list of unrelated services.
- Count on customized publics not available in the EEA.
- Leave implicit geography.
- Use a generic page for many different purposes.
- Optimize only by clicking.
- Ignore the motives for lead discard.
- Confuse ads and organic presence in responses.
Checklist before launch
- One main decision per group.
- Problem and recipient described with concrete words.
- Geography verified in the account.
- Available functions controlled for the Italian market.
- Announcement consistent with the page.
- Tested conversion.
- UTM distinct for country and variant.
- Method to classify contact quality.
- Break thresholds and written optimization.
E-ROE builds ChatGPT Ads strategies and campaigns in Italy around real customer decisions. For a channel comparison, read ChatGPT Ads vs Google Ads; for the full process, see how to create a ChatGPT Ads campaign.
Official sources
- Official Overview ChatGPT Ads
- Country geographical targeting
- Personalized advertising and limitations for EEA and Switzerland
Frequently asked questions
Does ChatGPT Ads use keywords?
The keywords can help you describe a theme, but the channel should not be designed as a simple exact match. OpenAI considers more relevant signals, including context and intent of conversation, ad, page and settings available.
Can I choose cities where to show ads?
Advertiser API documentation supports targeting for country, region and DMA. Selectable locations depend on the catalog and features available in the account.
Are custom audiences available in Italy?
OpenAI’s documentation states that custom audiences are not supported for campaigns targeting the European Economic Area or Switzerland, where personalised ads are not currently available.
How do you write good contextual indications?
Describe problem, goal, choice criteria, situation and decision-making. Avoid generic formulas that could fit almost every conversation.
How do I understand if targeting is too wide?
See not only clicks and CTRs, but consistency of visits, conversions, qualified requests and commercial feedback. Contacts out of profile are a signal to be traced back to context, message and page.