ChatGPT ads campaign management: writing hints and fixing delivery

In search advertising the user types a query and you bid on that query. In ChatGPT ads the user is holding a conversation; the system understands the context and shows your ad if it fits. What you bid on is not a word but a description of a need.

That is why a keyword list carried over from Google Ads is useless here. What you need are a few sentences of context hints explaining in which conversations your product is a natural answer.

First, describe the situation, not the product. Instead of “accounting software”, write “a small business owner is looking for a way to issue invoices and track VAT”. Second, state who it is not for; the system uses exclusions as context too. Third, give each hint a single need; packing several scenarios into one paragraph blurs the match.

A good hint for an online language school: “the user wants to practise spoken English quickly for a job interview or a move abroad; we offer live tutors with flexible hours”. A bad one: “English course, online learning, best price”.

Split the structure by need rather than by product. Let each campaign represent a user situation: just starting, comparing, ready to buy. Under each campaign keep two or three hints specific to that situation and a matching landing page.

Concentrate the budget in one campaign at first. Small scattered budgets get stuck in learning before collecting enough impressions. When you update hints the old versions do not remain, so note the change date yourself in order to measure its effect.

Each campaign's landing page should answer the situation described in the hint directly. Sending a comparing user to a page with a comparison table rather than to the home page gets more conversions out of the same click.

Ad approval status first; a rejected ad waits silently. Then budget and bid: a daily budget below the platform's suggested minimum may receive no impressions. Third, hint breadth: a hint that is too narrow fits no conversation, while one that is too broad enters irrelevant chats and is quickly filtered out.

The fourth check is the targeting intersection: see whether country, language and device limits together leave an almost empty audience. Most “no delivery” problems are solved at one of these four steps.

The most common mistake on a new channel is launching the ad and setting up conversions a week later. The GA4 event and campaign parameters must be on the landing page before launch; otherwise the first budget spent teaches you nothing.

In the Tecrube connector, GA4 and Tag Manager sit in the same chat: “add a form submission event to the landing page for the ChatGPT campaign and mark it as a conversion in GA4” arrives as a preview and is applied on approval. After that you compare the channel's results with Google, Meta and the other connected platforms in the same panel, under the same conversion definition. You do not have to launch the ad the moment it is built; verify measurement first, then switch it on.

Judged on its own in the first weeks, a new ad channel almost always misleads. Impressions are low, learning is still running and the conversion count is not statistically meaningful. A healthier approach also looks at the channel's effect on the others: after the week ChatGPT ads went live, did brand searches rise, did direct traffic change?

Because GA4 and Google Ads share the Tecrube connector, that question is answered in one chat: “compare brand search impressions and direct traffic over the last four weeks, before and after the ChatGPT campaign's start date”. It is a read and spends no credits.

Make the budget decision on at least four weeks of data. Switching the channel off before that, or doubling its budget, rests on impatience rather than data. A software company that sees zero conversions in the first two weeks and shuts the campaign down never sees the lift in brand searches that might have started in week three.

At the end of four weeks a single question is enough for the decision: is the cost of conversions from this channel reasonable compared with similar campaigns on Google and Meta?