Our newsletterAcademicarXiv, 5 Aug 2026
ChatGPT accounts in lower-income US ZIP codes were more likely to be shown ads
Audience targeting
- Reader Note:
- Based on 91 synthetic U.S. ChatGPT accounts assigned to ZIP-code proxies for race/ethnicity signal and income tercile. The outcome is whether an account ever received an ad during the full study window. Income is median household income of the assigned ZIP code, not each account’s actual income. The authors say the study uses synthetic accounts, one country, and an early rollout window, so this is an early baseline rather than a general rule for mature delivery.
- Evidence:
- Observed behaviour
- Vendor interest:
- None
- Sample:
- 91 synthetic ChatGPT accounts in the US, stratified in a 3×3 design by race/ethnicity signal (Black, Hispanic, White) and income tercile (low, medium, high), with at least 9 accounts per cell; 335 prompts; 127,801 conversations total.
Reader Note:
Evidence:
Vendor interest:
Sample:Good for
Comparing yourself against a similar segment, and choosing what to test next.
Not for
- Saying that one thing caused another, or predicting what you would get.
- The two groups were not split at random — they picked their own side. The difference may have been there before anything was measured.
Related cards
Audience targetingTrend
TrendPrompts that name a brand are much more likely to show an ad
Idea
Go deeper on this topic
Our Substack turns the data behind these cards into full breakdowns: what changed, why, and what to do about it.
Audience targetingIdea
Category-adjacent ads became the norm at 58%