ChatGPT Ads hub

Techniques that matter at launch

The console is simple. The harder shift is the brief: instead of a keyword list you describe a decision moment. The practices below give you a readable test, not just a neat CTR.

Updated: 1 September 2026

Write hints as conversation descriptions

A context hint should tell the system which decision the offer helps with. A pile of product nouns leaves too little context.

Weaker hintBetter context description
chair, armchair, deskThe user is setting up a home office and comparing ergonomic chairs within a budget
English courseThey want regular work English, speaking practice, and are comparing a course with an app
CRMA small company is choosing its first CRM: leads, simple reports and email integration
mattress, bedThey are comparing a mattress for a small bedroom, have back pain and a budget cap

One group, one problem

Do not copy the shop category tree. Build groups around situations: home office setup, a trip with kids, a first CRM, a course for a job change. Keep hint, creative and landing aligned inside each group.

If one group mixes several problems, the report will not tell you anything. You will not separate a weak offer from weak context. Two or three groups are enough at the start, not twenty.

Creative should finish the conversation

  1. The headline answers a criterion, not a brand line The user just compared options. 'Chair for 8 hours at a desk, ships in 48 hours' says more than 'Ergonomics leader'.
  2. The description removes one doubt Size, lead time, returns, who the offer is for. Do not repeat the homepage slogan.
  3. The image shows the product in use or the outcome A generic lifestyle pack does not carry the detail from the chat. A clear product, kit or product UI works better.
  4. The URL continues the same topic The landing page is an auction signal. A homepage about 'complete solutions' weakens match.

The landing page is part of targeting

After the click, the visitor should see the same problem they described in chat. If they asked for a chair for a small flat, do not greet them with the full furniture catalogue.

The page needs a concrete offer, a price or range, delivery or onboarding, and one clear next step. For leads: a short form and what happens after submit. For a store: product, variant, availability.

If you cannot build new landings yet, at least send people to a category or article that matches the context. Leave the generic homepage until you know which groups get clicks at all.

Measurement before the first bid

  1. UTMs on every URL Stable source, medium and campaign names. Otherwise ChatGPT Ads traffic will blend with organic AI referrals in GA4.
  2. One business event Purchase, begin checkout, a qualified lead, trial or subscription. Do not optimise to 'all conversions'.
  3. Pixel or CAPI, with dedup CAPI is more durable. If both fire, use the same event_id. Do not drop a new script next to your CMP without checking consent.
  4. An order or lead identifier Without it you cannot reconcile Ads Manager with CRM, the store and GA4. Clicks alone will not settle the test.

A first-test plan

Do not start with feeds and APIs. First check whether Polish traffic creates useful sessions.

  1. Check offer and landing policy Category, claims, landing content. A rejection after you built creatives wastes a week.
  2. Pick one outcome For a store, purchase or checkout. For a service, a lead you can qualify. For SaaS, trial or payment.
  3. Two or three problem groups Different context, different landing. Then you can see which conversation converts.
  4. Write the decision rule Minimum clicks or conversions, maximum CPA, lead quality and a review date. Without that, novelty is easy to confuse with a result.

How to judge results without a novelty effect

CTR and CPC help diagnose creative. The channel is judged on CPA, session quality, order or lead value, and the share of new customers. In ecommerce, use margin after ad cost, not ROAS alone. In lead gen, pull the source through to CRM.

If the sales cycle takes weeks, a monthly Ads Manager report is not enough. Look at the cohort from click date. Also be careful when comparing with organic ChatGPT traffic: without UTMs the two sources look the same.

In this section

Sources

Want to test the channel without guessing from CPC?

I can help set up the first test: ad group context, landing pages, UTMs, pixel or CAPI, and a decision rule after two or three weeks.

This material is educational. Ads Manager is in beta, and policies and features can change. Check OpenAI's current rules and your category eligibility before you spend.