Troubleshooting

Web Search Returned Nothing?

If your agent says it searched but came back empty, don’t guess. Run a quick capability + query isolation flow and recover in minutes.

Recurring support pattern: users ask for recent news, competitors, or links and get “no results,” vague summaries, or stale answers. Most of the time this is not a model problem — it’s a query, scope, or tool-availability mismatch.

Reality check: “No results” can mean very different things: blocked tool access, over-constrained filters, too-broad prompts, or an unavailable source.

Fast diagnosis (2 minutes)

  1. Run a tiny canary query: "latest OpenAI news".
  2. If canary fails too, treat this as a tool/session capability issue.
  3. If canary works, your original query is likely too broad, too narrow, or mismatched to available sources.
  4. Retry with explicit constraints: timeframe + region + output format.

Known-good recovery flow

  1. Start with one narrow query (single topic, single intent).
  2. Add one filter at a time (date range, country, language).
  3. Ask for 5 links max with title + one-sentence relevance.
  4. Then expand to deeper research only after the first pass works.

Copy/paste rescue prompt

Search task (strict): 1) First run a canary search for: "latest OpenAI news" 2) Report whether search capability is working (yes/no) with 3 links. 3) Then run my query with only these constraints: - timeframe: last 30 days - language: English - output: 5 links max, each with title + 1-line why it matters 4) If no results, tell me exactly which constraint likely caused it and retry once with relaxed constraints.

Avoid this trap: asking for “everything about X” with strict date/geo filters in one step often returns weak or empty output.

High-friction edge cases

Better query pattern

Research topic: [exact topic] Goal: [decision you need to make] Constraints: - Time range: [e.g., last 14 days] - Region: [US/global] - Source type: [news/docs/blogs] Output format: - 5 links max - For each: title, date, URL, 1-sentence relevance - End with: 3-bullet takeaway summary

When to escalate to #help

Support packet

Web search issue report Surface where it fails (dashboard/DM/server/thread): Canary query used: Canary result (worked/failed): Original query: Filters applied (date/region/language): Expected output: Actual output: Did relaxing constraints help (yes/no): Any exact error text: Message link/screenshot:

Bottom line: prove search capability first, then tune query constraints one variable at a time. That avoids 80% of “search is broken” loops.