OpenAI Introduces Negative Phrases for ChatGPT Ads

On October 5, 2026, OpenAI officially introduced Negative Phrases, a brand safety tool that allows advertisers to exclude ads from appearing in ChatGPT alongside user messages containing certain words and phrases.

The feature is designed for advertisers with additional corporate requirements for ad placements and is available only to accounts that meet OpenAI's eligibility requirements.

How Negative Phrases Work in ChatGPT Ads

Unlike Google Ads, where advertisers use Negative Keywords to exclude search queries, ChatGPT Ads selects ads based on conversational context and user intent.

Negative Phrases provide an additional way to restrict that context.

For example, an airline could exclude ad placements alongside messages about plane crashes, while a financial brand could exclude placements alongside discussions of bankruptcy.

In August 2026, Adthena identified the Negative Phrases setting in a UK advertising account. According to Adthena's observations:

• The setting is located in the ad platform's Settings section.
• Advertisers can add up to 100 negative phrases.
• Each phrase can contain a maximum of 100 characters.
• An exclusion applies when a user message contains the specified phrase.

These technical limits were reported by Adthena but have not been confirmed in OpenAI's public documentation.

How Negative Phrases Differ From OpenAI's Standard Restrictions

OpenAI already applies its own brand safety mechanisms to prevent ads from appearing alongside sensitive, potentially harmful, or unsuitable discussions.

Negative Phrases add another layer of control, allowing advertisers to account for their brand's specific restrictions.

However, OpenAI has not disclosed the eligibility criteria for advertisers, the exact phrase-matching rules, or whether advertisers can exclude individual topic categories.

Negative Phrases are also not a tool for controlling ChatGPT's responses. They affect whether an ad is eligible to appear, but they do not change AI-generated responses or organic recommendations.