ChatGPT ads need a measurement plan

A new advertising channel does not create a new growth strategy.

It creates another place to test whether a clear offer, a useful website and reliable measurement can turn attention into qualified business.

That is the practical way Australian small and medium businesses should assess ChatGPT ads. The platform is now accessible enough to test. It is still new enough that a business should resist treating early clicks as proof of a durable acquisition channel.

The useful first move is a bounded campaign with one offer, one landing page, one conversion goal and a budget the business can afford to learn from.

What changed on 16 September 2026

OpenAI announced new ChatGPT advertising capabilities on 16 September.

The release included AI assistance for campaign creation and analysis, suggested copy and imagery based on a landing page, optional text customisation, and integrations with HubSpot and Shopify. OpenAI also introduced Sponsored Agents, which let a person continue from an advertisement into a clearly labelled conversation with a business sponsored agent.

These capabilities do not all have the same maturity or availability.

OpenAI describes Ads Manager as a beta product. Its current availability page lists self service access as available in Australia. Sponsored Agents are being tested only with selected advertisers in the United States. The Shopify app is available to United States merchants now, with an international launch planned for 23 September in markets where ChatGPT Ads is available.

That distinction matters. An Australian business can assess and test the advertising channel now. It should not plan around Sponsored Agents or assume every integration and campaign feature is generally available to its account.

Paid placement and AI visibility are different decisions

OpenAI says ChatGPT ads appear below responses, are labelled as advertisements and remain separate from ChatGPT's answer. It also says ads are not shown to Plus, Pro or Business plan users, or to accounts identified as belonging to people under 18.

That means paying for an ad does not make the business part of ChatGPT's independent answer.

This is the most important strategic distinction.

AI Search Optimisation improves how clearly a business can be understood through its content, structure, evidence and entity signals. ChatGPT ads buy access to eligible advertising inventory. One does not replace the other.

The same separation applies to crawler policy. A deliberate approach to AI crawler access can support discoverability and content preferences. It does not determine paid placement, and an advertising campaign does not repair weak organic content.

Businesses should measure the channel on its own terms: cost, qualified visits, useful enquiries, sales and contribution margin.

Which businesses should pay attention

The first candidates are businesses with an offer that is already understandable and measurable.

That may include an ecommerce store with reliable product information, a local service with a clear enquiry path, a travel or experience business with current availability, or a digital product with an established conversion journey.

OpenAI's current ad policy says the initial programme focuses mainly on consumer goods, local services, travel and experiences, digital products and education. Other categories may be restricted, prohibited or reviewed individually. Financial, health and legal advertising is particularly limited outside the United States.

An available account does not guarantee an eligible offer.

The business should check category eligibility before preparing a campaign. It should also own the advertiser account. OpenAI's setup guidance says an agency cannot create the account on a client's behalf, although the client can invite the agency after creating it.

A business should wait when the offer is still changing, the landing page is unclear, conversion tracking is unreliable or nobody can explain what a qualified result looks like.

Ads make an unclear path more expensive. They do not make it clearer.

The landing page is part of the ad system

ChatGPT Ads does not use an exact keyword model.

OpenAI says delivery can consider the current conversation, the landing page, the ad title and copy, advertiser supplied context hints and, when advertising personalisation is enabled, selected signals from a user's wider ChatGPT experience. Context hints describe relevant conversations or needs, but do not guarantee that an ad appears in a particular conversation.

The landing page therefore has two jobs.

It needs to help the advertising system understand the offer. It also needs to help a person decide whether the offer fits after they click.

A useful page should make the following information obvious:

• what the business provides;
• who the offer is for;
• which problem it solves;
• where the business operates;
• what the offer includes and excludes;
• what proof supports its claims;
• what the visitor should do next.

This is normal Website Development work. The page still needs clear copy, accessible structure, fast performance, accurate metadata and a useful conversion path. A new advertising platform does not remove those fundamentals.

Start with one bounded acquisition test

Do not begin by moving an existing advertising budget into ChatGPT because the channel is new.

Start with one offer whose current performance is already understood.

Before launching, record:

• current qualified visits and enquiries;
• conversion rate for the landing page;
• cost per qualified enquiry or sale where paid traffic already exists;
• average order value or expected client value;
• common reasons leads are unsuitable;
• the time spent reviewing and following up each enquiry.

Then create a test with one primary conversion. For an ecommerce business, that may be a completed order. For a service business, it may be a qualified enquiry rather than any form submission.

Keep the initial campaign narrow:

1. Use one stable offer and one relevant landing page.
2. Write specific context hints around customer needs, not broad topic labels.
3. Use a budget and test period that limit the cost of weak early delivery.
4. Keep campaign claims, generated creative and translations under human review.
5. Track the source in the business's existing analytics with clear campaign parameters.
6. Review lead quality, not only impressions and clicks.
7. Change one meaningful variable at a time before expanding spend.

The aim is not to prove that ChatGPT can send traffic.

The aim is to learn whether that traffic creates commercially useful actions at a cost the business can sustain.

Treat the published pricing as a starting signal

OpenAI's advertiser guidance supports impression billing and valid click billing. For click campaigns, it recommends starting with a maximum bid of US$3 to US$5 per click.

That is platform guidance, not an expected Australian cost or a forecast of campaign performance.

Actual delivery depends on relevance, competition, targeting, inventory and the chosen campaign settings. OpenAI's API documentation also says daily minimums depend on the account currency, rather than publishing one universal campaign minimum.

The dashboard currently reports impressions, clicks, spend, click through rate, average click and impression costs, and conversions. OpenAI warns that spend reporting can lag behind impressions and clicks.

For a smaller business, that makes an external budget record useful. Set the approved campaign budget before launch. Check platform spend and payment settings regularly. Compare the advertising data with enquiries, orders and CRM outcomes rather than treating the advertising dashboard as the whole commercial record.

Review privacy before adding the measurement pixel

OpenAI supports campaign parameters, a browser measurement pixel and a server side conversions API.

The simplest test may begin with campaign parameters and the analytics the business already operates. More advanced measurement should be introduced only when the data is necessary and the privacy implications have been reviewed.

OpenAI's measurement pixel documentation says pixel consent defaults to true unless the implementation sets it to false or a stored denial is found. It also supports automatic advanced matching, which can detect supported customer information and send hashed values with conversion events.

Hashing does not remove the need for privacy judgement.

The Office of the Australian Information Commissioner says organisations using third party tracking pixels should conduct due diligence, minimise data, explain the collection clearly, provide an appropriate opt out for direct marketing and avoid sending sensitive information without the required consent.

Before adding the pixel, document:

• which pages load it;
• which events and fields are sent;
• whether customer information matching is enabled;
• when measurement consent is granted or denied;
• how the privacy notice explains the technology;
• who reviews the setup after the campaign ends.

Privacy Act coverage and consent requirements depend on the business and the information involved. Businesses handling personal or sensitive information should obtain appropriate privacy or legal advice where the position is unclear.

AI assistance still needs a responsible reviewer

OpenAI can suggest copy and imagery from a landing page. Its optional text customisation can adapt headlines and descriptions to the context of a conversation and translate the copy.

Those features may reduce production effort. They can also change a qualification, product detail, price, service area or claim that the business needs to state precisely.

Every published variation should remain consistent with the landing page and the real offer.

The Australian Competition and Consumer Commission says businesses must be able to support advertising claims and that qualifications must not contradict the overall message. OpenAI's ad policies similarly require accurate claims and consistent landing pages.

Human review should remain responsible for:

• factual product and service details;
• prices, exclusions and availability;
• comparative or performance claims;
• translated copy;
• generated images;
• category and audience suitability;
• campaign budget and approval.

AI can prepare a variation. The business still owns the promise.

Keep the test portable

ChatGPT Ads Manager is in beta. Features, eligible categories, prices, inventory and integration behaviour will change.

Keep the campaign brief, approved claims, source assets, landing page copy, analytics definitions and results in systems the business controls. Record why the campaign was changed or stopped. Preserve the baseline so performance can be compared with other channels later.

For Shopify merchants, the new app can reduce setup effort once it is available internationally. Its current App Store listing says it can sync the product catalogue, install measurement and manage campaigns, but it also requests access to customer, product, marketing and analytics data. Convenience should not replace an access and privacy review.

For HubSpot users, the integration may reduce handoffs between advertising and lead follow up. It should still preserve clear campaign ownership, lead qualification and access controls.

The business should be able to leave the platform without losing the strategy, creative history or understanding of what converted.

The next decision

ChatGPT ads are a meaningful new acquisition option because Australian businesses can now access the beta directly and measure traffic and conversions.

They are not a shortcut to organic AI visibility, a reason to rebuild a website or proof that conversational advertising will outperform existing channels.

Start with one offer, one landing page, one conversion baseline and one controlled budget. Review the claims, privacy settings and lead quality. Expand only when the complete path from conversation to commercial outcome is working.

For a business with an established website and a measurable offer, Website Growth & Care can frame the test, improve the landing path and keep measurement connected to ongoing website decisions. When the choice affects search strategy, analytics, customer data and several marketing systems at once, a Fractional Technical Partner can help decide what to test, what to protect and what not to scale yet.