How to set up ChatGPT Ads
The setup itself takes about twenty minutes. Three of those minutes decide things you cannot undo. This is the order I would do it in, and the parts where I would slow down.
Before you open the account
- Decide the test budget and the window, and write down the number that would make it worth continuing. Do this first, on purpose, so nobody can move it later.
- Build the UTM scheme. One source value you will never confuse with anything else, used on every link.
- Add a 'how did you hear about us' field to the enquiry form if you do not have one. With attribution this thin, asking people is not beneath you.
- Check your landing page answers the question the conversation was about. Sending considered research traffic to a homepage wastes the click.
The three permanent settings
At account creation you choose a country, a billing currency and a time zone. More than one source reports that none of them can be changed afterwards.
Australia, AUD, and your own capital city time zone. Get the currency wrong and you pay conversion on every invoice for the life of the account. Get the time zone wrong and your daily budget resets in the middle of your trading day.
Targeting, without keywords
There is no keyword list and no match types. Instead of telling the platform which words to match, you describe the conversations your business belongs in.
That is a different skill. A keyword list is a guess at vocabulary. This is a description of a situation, and it rewards people who know their customer's problem rather than their customer's search habits.
Write it as the moment, not the product. Not 'conveyancing Brisbane' but 'somebody who has just had an offer accepted and is working out what happens next and what it costs'.
Running the first campaign
- Day oneOne campaign, one clear offer, geography set to where you actually service. Do not open five variations at once, you do not have the volume to read them.
- First weekLeave it alone. The temptation to tune something after two days is strong and there is not enough data under it to mean anything.
- Weeks two and threeFirst read. Cost per click against what you expected, and whether the enquiries coming through are the right kind, not just the right number.
- Week sixAgainst the number you wrote down at the start: keep, grow or stop. If it is stop, stop. Being early to a channel is not a reason to keep funding a result that is not there.
What you can and cannot track
Reporting is aggregated impressions and clicks. There is no pixel based attribution, no demographic breakdown and no view through data.
So your own analytics carries the weight. Tagged links into GA4, the form question as a backstop, and a plain comparison against what the same money did elsewhere.
If that level of measurement is not enough for how you make decisions, this is a fair channel to sit out for now. I would rather tell you that than take the budget.
Want me to run it
Fixed monthly fee, agreed before anything starts. Not a percentage of spend. No lock-in.
The account is opened in your name with me added to it, so if we stop working together you keep the account, the history and whatever we learned.
How to set up ChatGPT ads, straight answers
Where do I sign up?
ads.openai.com for self serve. Larger budgets can also go through OpenAI's sales team or the holding groups, though for most Australian businesses self serve is the right door.
Do I need a product feed?
Only if you sell products. Service businesses run without one. If you do have a catalogue, feed quality matters more than clever creative.
How long before I know if it works?
Six weeks, with enough budget to buy a few hundred clicks. Any faster and you are reading noise.
Can you set it up and hand it over?
Yes. A one off setup, the tracking wired properly, and a written explanation of how to read it. Then it is yours. I would rather do that than hold an account hostage.
What if it stops being available in Australia?
Then we stop, move the budget back, and you have learned the format ahead of the people who waited. That is the honest downside of an early channel and it is why the test is ring fenced.
Read next
The service behind it
Want somebody to actually run it?
Send me what you are already spending and what you want it to do. You get a straight read on whether this channel is worth your money yet, whether or not it turns into work.
8am to 5pm AEST, Monday to Friday. Every email lands with me, not an inbox somebody checks on Fridays, and you get a reply the same working day.
- 01Send
Read access to the account, or screenshots of the last ninety days.
- 02I read it
And write back in plain language: what is working, what is wasting money, what I would do first.
- 03You decide
No pitch deck, no follow-up sequence. If the account is fine, I tell you that and we are done.