How ChatGPT Ads work: a perspective for Vietnamese marketers

ChatGPT Ads have entered the Vietnamese market

On September 23, 2026, OpenAI announced the start of ChatGPT Ads rollout in Vietnam and six other Asian markets. The Ads Manager list currently marks Vietnam as having self-service access. This milestone makes conversational advertising something Vietnamese marketers can monitor and test rather than merely a distant forecast.

OpenAI says ads may appear for eligible Free and Go users. Plus, Pro, Business, Enterprise and Edu plans do not display ads. Whether an individual sees an ad also depends on region, rollout stage, account eligibility and the ad's relevance to the conversation.

What matters is not simply another media placement. When users ask AI to make shortlists, compare pros and cons or prepare a purchase, brands gain a touchpoint potentially connected to an explicitly expressed problem.

Where do ads appear, and do they affect answers?

According to OpenAI, sponsored placements may appear below answers, labeled and visually separated. An ad unit may contain one or more products or advertisers. It is a paid placement, not a sign that OpenAI recommends the brand.

Ads operate on a separate system. Advertisers cannot pay to edit answers, force ChatGPT to rank their products higher in responses or turn an organic answer into an advertising message. Marketers must communicate this boundary accurately in plans and reports.

The article's illustration depicts only the principle of a sponsored label below the conversation, not an official interface layout or advertising format.

Which signals determine ad selection?

OpenAI says the system considers relevance and expected results. Signals include context and intent in the current conversation, landing pages, headlines, ad copy, contextual hints and advertisers' targeting choices. When users enable ad personalization, some signals from their broader ChatGPT experience may also be used according to their settings.

Advertisers can suggest suitable conversations, topics or keywords. These are not exact-match commands ensuring an ad appears whenever a particular question occurs. When several ads qualify, both relevance and bids play a role in delivery order.

Hypothetical example: a user asks how to choose accounting software for a newly opened shop with multiple sales shifts and invoicing needs. A message clearly identifying supported shops, relevant features, implementation conditions and a transparent explanatory landing page may fit better than a generic advertisement. This example does not claim the system will select a particular brand.

The strategic difference lies in the question the user is trying to solve

In Google Search, marketers often begin with the query users type. On social platforms, content and ads meet users in a discovery feed. ChatGPT adds another context: users may explain goals, budgets, constraints and comparison criteria, then ask follow-ups across several turns. This is a directional perspective, since Google and Meta also offer many formats and delivery signals beyond these familiar examples.

Someone asking 'Which accounting software is good?' has provided only a topic. Someone asking 'A two-branch shop selling across channels needs inventory sync and invoices, with a budget under X' adds constraints and selection criteria. Marketers should move from short keyword lists to maps of customer situations: are they researching, comparing, eliminating options or ready to contact a supplier?

An ad creates only an opportunity to appear at a touchpoint. To progress, a brand needs a landing page addressing that situation, credible evidence and a clear action. Keep paid-ad strategy separate from building useful content users may discover during independent research.

What do advertisers know about users?

OpenAI says it does not share conversations, chat history, memories or personal information with advertisers and does not sell them user data. Advertisers receive aggregate performance reports, such as impressions and clicks. If users proactively message an advertiser directly through an ad, it sees only the messages sent to it.

Users can disable ad personalization, hide or report ads and view why an ad appeared. Disabling personalization does not eliminate ads; current-chat context may still be used to select them. OpenAI also says ads are ineligible alongside certain sensitive topics such as personal health, mental health and politics.

For marketers, this sets a clear limit: do not plan as though you can read private prompts or build audience lists from personal conversations.

How can ChatGPT Ads be purchased and measured?

OpenAI has opened impression- and click-based buying, along with Ads Manager beta and advertising partners. Current documentation describes goals, budgets, bids and creatives including advertiser names, headlines, descriptions, images and landing pages. In Vietnam, businesses should check account eligibility and permitted paying legal entities against Ads Manager's status when registering.

Beta reports may currently include impressions, clicks, spend, CTR, CPC, CPM and conversions. OpenAI also describes conversion-measurement tools and UTM parameters on landing-page URLs. Consider these metrics alongside lead quality, actual revenue received and contribution margin, rather than stopping at clicks.

A reasonable experiment selects a specific problem group, dedicated landing page, capped budget and measurement goal before launch. There is not yet a basis to assert that this channel has lower CPC, higher ROAS or can replace Google and Meta in every industry.

A checklist of 5 things marketers should do now

First, record actual customer questions while researching, comparing and preparing to buy. Second, group questions by goals and constraints such as budget, scale and implementation time. Third, revise the landing page to address one specific situation instead of listing every feature.

Fourth, prepare verifiable evidence: applicable conditions, product images, processes, policies, use cases and contact details. Fifth, once the account has advertising access, measure from impressions through qualified leads to business results, while comparing with existing channels.

The strategic question is not 'How do I make AI say my name?' A more useful question is: when a customer is solving a specific problem, is the brand relevant and trustworthy enough for them to choose to investigate further?

Suggested actions