Study: One-Third of In-Assistant ChatGPT Ads Land in Irrelevant Conversations

An empirical analysis of more than 11,000 advertisements served inside ChatGPT conversations found that 33% had no discernible relevance to the surrounding discussion. Conducted by AI visibility platform Searchable between July 4 and August 4, 2026, the study paired live ad impressions with conversation logs to evaluate contextual alignment, commercial intent, and sector-specific accuracy across OpenAI's conversational ad inventory. OpenAI introduced advertising into ChatGPT in February 2026, r

2 min
Study: One-Third of In-Assistant ChatGPT Ads Land in Irrelevant Conversations

An empirical analysis of more than 11,000 advertisements served inside ChatGPT conversations found that 33% had no discernible relevance to the surrounding discussion. Conducted by AI visibility platform Searchable between July 4 and August 4, 2026, the study paired live ad impressions with conversation logs to evaluate contextual alignment, commercial intent, and sector-specific accuracy across OpenAI's conversational ad inventory.

OpenAI introduced advertising into ChatGPT in February 2026, relying on semantic "context hints" rather than traditional keyword targeting. Advertisers define broad themes, user scenarios, and conversational triggers, which OpenAI's matching engine maps against real-time dialogue.

Relevancy Tiers and Purchase Intent

Searchable categorized each served ad into three primary relevancy tiers:

  • Unrelated (33%): Advertisements bearing no thematic or semantic connection to the user's prompt or ongoing dialogue.
  • Contextual Match (40%): Placements tied to tangential topics raised earlier in a session history, but disconnected from the active query in view.
  • Direct Match (27%): Targeted promotions directly corresponding to the specific product, service, or brand requested by the user.

Beyond topical alignment, the study found a substantial mismatch in user commercial intent. Approximately 68% of served ads appeared in threads where users exhibited no commercial intent: no indicators of searching for products, comparing pricing, hiring services, or booking transactions. Across all ad-carrying sessions, 40% included at least one completely unrelated ad, while 28% consisted entirely of irrelevant placements.

Infographic diagram illustrating conversational ad relevancy distribution, purchase intent gaps, and vertical performance

Vertical Performance Disparities

The accuracy of OpenAI's ad matching engine varied widely across commercial sectors:

  • Marketing and B2B Services: Recorded the highest irrelevance rate, with 47% of placements classified as unrelated to the conversation.
  • Software and SaaS: Followed closely with 45% unrelated ad placements, frequently triggering on technical troubleshooting queries without purchasing intent.
  • Data Brokers and Background Checks: Achieved the highest direct match rate among measured sectors at 50%.
  • Travel and Automotive: Registered direct match rates of 43% and 42% respectively.
  • Insurance: Exhibited a low unrelated rate (23%) but also a low direct match rate (11%), with 66% of impressions falling into the contextual match band based on life events referenced earlier in conversations.

No commercial vertical achieved a direct relevancy match rate exceeding 50%.

Algorithmic Signals vs. Conversational Context

The findings highlight the technical friction inherent in monetizing conversational generative AI. Unlike search engines where explicit query strings provide deterministic purchase signals, conversational assistants are utilized for exploratory tasks, software debugging, education, and writing.

In June 2026, former OpenAI Chief Revenue Officer Denise Dresser noted that user ad dismissal rates had dropped by half since rollout, citing dismissals as a proxy for improving relevance. However, Searchable's longitudinal dataset indicates that the underlying proportion of completely unrelated ad placements held steady at approximately one-third throughout the evaluation window, suggesting that dismissal reductions may reflect user acclimatization rather than improved matching precision.

Sources

Written by

More to read

  • Vision-Language Model Serving in Production: Visual Token Pruning, Encoder Caching, Dynamic Resolution, and Inference Economics

    Deploying Vision-Language Models (VLMs) into high-concurrency production environments introduces a distinct set of systems bottlenecks that text-only large language models do not exhibit. While text models ingest prompts with compact token densities, visual inputs require processing high-dimensional pixel arrays through vision encoders, expanding a single image into hundreds or thousands of visual tokens before autoregressive generation begins. In production architectures running models such as

    1 min
  • Duke Study Finds Creative Output Across Frontier LLMs Is Converging Toward Semantic Monoculture

    A longitudinal study by researchers at Duke University has documented a measurable decline in output diversity across major large language model families. Tracking 69 foundation model checkpoints released between March 2023 and July 2026, the researchers found that generative responses to open-ended creative and ideation prompts are increasingly converging toward semantic uniformity across competing providers. The paper, titled Are LLMs becoming similarly creative? Evidence from three years of

    1 min
  • Study Exposes Benchmark Overfitting and Acoustic Leakage in Speech Recognition Models

    A joint empirical study by researchers from Hugging Face and Hume AI has uncovered evidence that top-ranking automatic speech recognition (ASR) models frequently exhibit benchmark optimization, commonly referred to as "benchmaxxing." The findings demonstrate that models often achieve low word error rates on public leaderboards by memorizing test-set reference anomalies and responding to subtle acoustic channel signatures rather than generalizing audio transcription. Evaluating 11 open-source sp

    1 min