Skip to content
BlogRSS ↗

Notes on ChatGPT ads.

What the documentation says, what we’ve tested while building Serge, and what advertisers are finding. I’m keeping the evidence and the open questions in one place.

September 2026 · Articles in English
Four luminous paths converge: conversation, landing page, ad copy and context hints all contribute to matching.
Matching8 min read

Context hints aren’t keywords.

What guides matching, what your export can tell you, and how I’d write the hints.

Read the notes
Three glass checkpoints connected by a light path: ad click, landing request and analytics session.
Measurement8 min read

A click isn’t a session.

Use the request log to find the gap between the ad click and the analytics session.

Read the notes
A prismatic budget reservoir with a clear upper boundary illustrates a total campaign spending cap.
First campaign9 min read

Set the limit before launch.

Budget types, review gates and the stop rule I’m working from for our first campaign.

Read the notes
Two separate glass layers represent the response and the sponsored placement below it.
Creative8 min read

Write the next answer.

The real character limits, the placement, and what happened when our first ad went through review.

Read the notes
An illuminated audience aperture separates the ad-supported Free and Go plans from the ad-free paid plans.
Audience7 min read

Who actually sees the ads?

The eligible plans, the exclusions, and why a subscription tier isn’t a buyer profile.

Read the notes
New to the channel?

Start here.

Five notes, in the order I’d read them before setting up a first campaign.

  1. 01Who actually sees the ads?
  2. 02Context hints aren’t keywords.
  3. 03Write the next answer.
  4. 04Set the limit before launch.
  5. 05A click isn’t a session.