Assistant requires URLs for three topics

The failure was procedural, not editorial. The supplied material for the briefing contained no retrievable reporting on Ukraine-Russia peace talks, China property stimulus and the economy, or U.S. election campaign polls. Instead, the upstream note said every fetch for those topics had returned the same search error — “Exa API returned 402” — and the assistant said it could not draft a briefing because it had been instructed to rely on provided source material and cite every key claim. That left the system with three empty folders where the reporting should have been. The note said it would not “responsibly infer or add news from elsewhere” because the request required it to use only the pre-fetched content and to provide citations in source-title-and-URL form. The recommended fix in the same note was operational: resend the request with working pre-fetched articles or URLs, or restore the Exa pipeline, and then generate the briefing. ### Why did the assistant stop instead of filling in the gaps? The upstream briefing said the assistant was bound by a sourcing rule: every key claim had to be traceable to supplied material and cited. Without retrievable articles, it said, any substantive summary would break that requirement. OpenAI’s API documentation separately describes error-code handling as part of normal API behavior, and Exa’s own documentation says HTTP 402 is tied to its x402 payment flow, which uses “Payment Required” as the trigger for a paid retry. ### What exactly was missing from the package? The three missing topic areas were named explicitly in the source note: Ukraine-Russia peace talks, China property stimulus and the economy, and U.S. election campaign polls. The note did not include article text, publication names, working URLs, or quoted reporting for any of them. Because those source packets were absent, the assistant said it could not produce the requested world briefing in the required 600-800 word format. (developers.openai.com) It instead returned a short explanation that no usable articles had been supplied and asked for working pre-fetched articles or URLs. ### What does “Exa API returned 402” point to? Exa’s documentation says x402 is an open payment standard built on HTTP 402 Payment Required and that clients can “pay and retry” requests on supported endpoints. A public GitHub issue posted in February reported Exa-backed web search failing with HTTP 402 in another toolchain, describing the result as a consistent Payment Required error rather than a normal search response. ### Why couldn’t the assistant just use other reporting? The source note said the request’s rules barred that. It said the assistant had to rely solely on pre-fetched content and cite every key claim, which meant it could not substitute fresh outside reporting once the fetches failed. (exa.ai) OpenAI’s help documentation also says its older Assistants API is being superseded by newer building blocks under the Responses and Agents platform, but that does not change the basic requirement that a system can only cite material it actually has. (github.com) ### What would fix the problem fastest? The source note named two remedies: provide live URLs for the missing articles or fix the Exa API path that produced the 402 errors. Either step would give the assistant source material it could quote, summarize, and cite. The same note said that once working articles were supplied, the assistant could turn them into a concise, fully cited briefing organized by theme. The next concrete step, then, is not another rewrite request. (help.openai.com) It is a new fetch with valid URLs or a repaired Exa connection for those three topics.