Why we tested ads in ChatGPT
In October 2026 we launched our first OpenAI Ads campaign, for a real estate client in the Puebla metro area in Mexico. The question we want to answer: can ChatGPT generate homebuyer leads at a cost comparable to Google Ads?
The reasoning is simple. More and more people ask ChatGPT which neighborhood suits them, how much a bank will lend them for a mortgage, or how to buy their first home. Those are questions from someone in the middle of a decision, and until now there was no way to show up in that conversation with an ad.
Here we walk through how we built the campaign, the decisions we made along the way, and what surprised us during setup. We'll share the results in a second part.
How OpenAI Ads works
The big difference from Google Ads is that there are no keywords. Instead, each ad group carries context hints: phrases that describe who your offer is for and in what situation.
- Context hints. They aren't targeting rules. ChatGPT uses them as guidance, together with the ad copy and the landing page, to decide which conversations your ad is relevant to. They work better as complete sentences built around a single idea than as lists of words.
- Location and exclusions. These are set at the campaign level, not in the hints. Mentioning a city in a hint adds context, but it doesn't limit where the ad appears.
- Objective and bidding. You can optimize for traffic or for conversions, and you pay per click or per thousand impressions. You can also choose between automated bidding and a manual maximum bid.
- Format. An ad is a card with an image, a short headline and a brief description that appears inside the conversation.
The structure: one ad group at MXN 150 a day
The client's approved budget was MXN 150 a day (roughly US$8), which is exactly the minimum OpenAI requires for accounts billed in Mexican pesos. That budget shaped the structure. Ultimately, we see this as a test.
Our first draft had four ad groups: buying a home, mortgages, selling a property, and investment. On that budget, each group would have received very few clicks a month and none would have gathered enough data. So we kept just one.
| Element | Setup |
|---|---|
| Objective | Conversions (lead created) |
| Budget | MXN 150 per day |
| Ad group | Buy a house or apartment in Puebla |
| Context hints | 7 phrases: buying guidance, job relocation, first home, comparing options, transparent service, and mortgage financing |
| Ads | 5 variations: guidance, listings catalog, two specific neighborhoods, and first-time buyers |
Mortgage financing went into the same ad group rather than a separate one. For a homebuyer, the loan is part of the same decision.
What we learned while setting it up
- The copy limits are tighter than you'd expect. The form allows 32 characters for the headline and 54 for the description. Some guides online cite longer limits, so we had to trim several versions. Trust the platform's character counter.
- Each ad should match its landing page. An ad that mentions a specific neighborhood should lead to listings in that neighborhood, not the general catalog, to cut down the clicks a prospect needs to find what they came for.
- Tracking can be set at different levels. The platform accepts UTM parameters and dynamic placeholders such as
{campaign_id}and{ad_id}. We added them at the ad group level so we could identify each variation in Google Analytics and in the CRM. - Automated vs. manual bidding was the hardest call. With a small budget, the algorithm has few conversions to learn from. But we also didn't know what a click costs in Mexico for this industry, and a manual bid set too low could leave us with no impressions. We chose automated bidding, with two conditions: confirm first that the conversion event was firing, and set an exit rule at the two-week mark.
- The advertiser account needs verification. Before the campaign can run normally, the account owner has to complete verification.
How we'll measure it
The campaign is already live: it was approved within minutes of submitting it for review. Our benchmark is the client's cost per lead in Google Ads. We expect this platform to match or beat it, though not necessarily right away. Here's what comes next:
- Days 3–4: we review impressions, clicks and actual cost per click, a figure we couldn't find published anywhere for Mexico.
- Week 2: if there are no leads, or cost per lead is double the Google Ads figure, we switch to manual bidding with a cost-per-click cap, the same way we launch campaigns on other platforms, mainly Google Ads.
- End of the month: we compare cost per lead and lead quality against Google Ads, and decide whether to scale, adjust or pause.
We'll publish the numbers in part two of this article.