Exa API returns 402 errors
- On September 2, 2026, three requested news searches failed after the Exa API returned 402 errors, leaving the assistant without articles, summaries or URLs. - The most specific available detail was the repeated error string, “Exa API returned 402,” across searches on the G20, Fed signals and Philippines tensions. - Next, the workflow can only proceed with a retry, a placeholder note, or user-supplied article text and URLs.
Three requested news searches failed on September 2, 2026, after the Exa API returned 402 errors instead of article results. The failed queries covered the 2026 G20 agenda, signals on Federal Reserve rate cuts, and South China Sea tensions involving the Philippines. No article titles, summaries or source links were returned. Without URLs, the assistant could not produce a sourced world-news briefing under the stated citation rules. ### Which searches failed, and what did they have in common? The three failed searches targeted distinct news topics: the G20 summit agenda for 2026, Federal Reserve rate-cut signals, and tensions between China and the Philippines in the South China Sea. Each search produced the same result — “Exa API returned 402” — rather than a list of articles. That common failure mattered because the workflow depended on pre-fetched news material. The assistant had been instructed to use supplied sources and cite key claims with exact URLs. With no returned links, there was nothing to verify or cite. ### What does a 402 error change in this workflow? The 402 response blocked retrieval at the search stage, not the writing stage. Because no article metadata came back, the assistant had no titles to inspect, no summaries to compare, and no URLs to open. The sourcing requirement then became the limiting factor. The assistant said it could not assemble a substantive briefing “without inventing material,” because any factual write-up on those topics would have required unsupported claims. ### Why couldn’t the assistant just write a brief from general knowledge? The governing instruction required supplied sources and exact citations. In that setup, general background knowledge was not enough to support a publishable briefing. No source bundle was available for any of the three topics. That meant the assistant could describe the failure itself, but not report out the G20 agenda, the Federal Reserve outlook or South China Sea developments as news items tied to cited reporting. ### What options were left after the searches failed? The assistant offered three next steps. The first was a placeholder “World in Brief” note stating that source retrieval had failed. The second was a blank briefing template that could be filled once sources were available. The third was a retry using user-supplied article text, titles and URLs. Those options reflected the narrow set of actions still possible after the failed searches. Without source material, the system could document the outage or wait for inputs, but it could not complete the requested briefing in the required format. ### What was the minimal reportable output? The shortest compliant version was a note saying source retrieval failed for all requested topics. That note named the three affected subjects and repeated the returned error message. The same placeholder also recorded the key operational fact: no article URLs were supplied. In this workflow, that single missing element prevented citation, verification and final assembly of the world-news brief. ### What happens next if the workflow is retried? A successful retry would need article text, titles or URLs for the same three topics before any sourced briefing could be written. If the retrieval step works, the next stage would be straightforward: review the returned reporting, verify the claims against the supplied links, and assemble the brief. Until then, September 2, 2026 remains the only firm date in the record, the 402 error string remains the only direct output from the failed searches, and user-supplied source material remains the named next input.