Associated Press Provides OpenAI With US Election Results

The Associated Press announced that it will provide its U.S. election vote-count results to OpenAI, supplying vote count data for national, state and local races in major American cities from today through the 2028 general election, per an AP press release. The release says the feed will be available to users of ChatGPT and OpenAI's services. The press release additionally states that in the 2024 general election AP counted the vote and declared winners in nearly 7,000 races with a 99.9% accuracy rate. "We are pleased to work with OpenAI to make available AP's factual, accurate and nonpartisan elections data available to users of ChatGPT and OpenAI's services," said David Scott, vice president of AP Elections, in the AP announcement.

Industry-pattern observations: Integrating a licensed, canonical tally feed into a large language model ecosystem typically reduces the need for the model to infer election outcomes from heterogeneous web content, and it improves the ability to cite a single authoritative source. For practitioners, common technical challenges with that pattern include handling update frequency and latency, preserving provenance metadata through downstream retrieval and response generation, and ensuring cached or offline model components do not serve stale tallies.

Editorial analysis: The announcement follows a period of public scrutiny over AI chatbots producing incorrect or misleading election information. News and platform operators have increasingly pursued direct licensing or partnerships with established data providers to reduce error rates and simplify attribution. The AP-OpenAI arrangement fits that broader pattern of tech platforms licensing specialized, high-integrity data to improve factuality.

Observers should track how the AP feed is surfaced inside ChatGPT responses: whether vote counts are presented with explicit attribution to AP, how timestamps and update windows are displayed, and how local and down-ballot races are handled. Also monitor operational details such as feed latency during high-throughput events on election night and whether OpenAI integrates the feed into retrieval pipelines or uses it as a single-source-of-truth layer for final answer generation.

Editorial analysis: Teams building or integrating LLM-based interfaces that surface time-sensitive facts can expect similar tradeoffs when adopting canonical feeds: improved factual grounding in exchange for engineering work on ingestion, provenance, freshness, and UX for attributed answers. Organizations evaluating such integrations should explicitly design for update handling and provenance propagation in model outputs.