# sonar-pro and sonar-reasoning-pro returning empty search\_results across all queries (2026-05-01) **URL:** \ **Category:** Bug Reports\ **Created:** [May 5, 2026, 9:54pm UTC](https://community.perplexity.ai/t/sonar-pro-and-sonar-reasoning-pro-returning-empty-search-results-across-all-queries-2026-05-01/5074 "2026-05-05T21:54:17Z")\ **Posts on this page:** 1\ **Page:** 1
**Author:** ![Steve\_Linn](https://sea1.discourse-cdn.com/flex001/user_avatar/community.perplexity.ai/steve_linn/32/199_2.png) [@Steve\_Linn](https://community.perplexity.ai/u/Steve_Linn)\ **Post date:** [May 5, 2026, 9:54pm UTC](https://community.perplexity.ai/t/sonar-pro-and-sonar-reasoning-pro-returning-empty-search-results-across-all-queries-2026-05-01/5074/1 "2026-05-05T21:54:17Z")
## **Reporter:** [slinn201@gmail.com](mailto:slinn201@gmail.com) **Date filed:** 2026-05-01 (Friday) **Affected models:** `sonar-pro`, `sonar-reasoning-pro` **Severity:** Service-impacting — automated job-discovery pipeline returning 0 results for all categories all day ## Summary ## Sonar’s web-search step is no longer feeding search results to the language model for our API calls today (2026-05-01). The model receives an empty `` block (or “None”) and correctly responds with `[]`. This was visible in `sonar-pro` only as `completion_tokens=1` and an empty content body, but became fully diagnosable when we swapped to `sonar-reasoning-pro`, which exposed `` traces in which the model itself states: _“the search\_results section is empty/None… There are NO search results provided.”_ The issue affects 3 of 4 of our prompt categories with `searchResults=0` and 1 category with `searchResults=1` (returning a single irrelevant URL: `https://mmwlegal.com/author/mmwmmwlegal-com/`). The same prompts and configuration worked normally earlier this morning (6:22 AM CDT) — at least for one of the four categories. We suspect either an account-level search degradation or a broader Sonar search-pipeline incident not currently reflected on [status.perplexity.com](http://status.perplexity.com). ## Reproducer `POST https://api.perplexity.ai/chat/completions` with body: ```json { "model": "sonar-pro", "messages": [ { "role": "system", "content": "You are a thorough job-search assistant. Search multiple job boards (LinkedIn, Indeed, BuiltIn, company career sites) and return real currently-live postings that match the user's criteria. Return ONLY a valid JSON array of job objects — no commentary, no markdown fences, no prose outside the JSON. Only the 5 REQUIRED fields are mandatory; for OPTIONAL fields, return \"\" (empty string) when unknown rather than dropping the posting. Prefer returning 1–3 imperfect-but-real matches over returning []. Only return [] if you genuinely cannot find anything after searching." }, { "role": "user", "content": "Find Asset-Based Lending (ABL) analyst-level positions in the United States across THREE title families... [full prompt available on request — ~1600 prompt tokens of named-employer anchors and title-phrase enumerations]" } ], "temperature": 0.3, "search_recency_filter": "month", "web_search_options": { "search_context_size": "medium" } } ``` * * * ## Observed behavior Across **4 runs in 30 minutes** spanning **3 different configurations** and **2 models** , all 4 prompt categories returned the same broken pattern. Detail below. ### Diagnostic field summary (sonar-pro) | Run | Time (CDT) | Config | Source | finish\_reason | completion\_tokens | citations | search\_results | content | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | 1 | 06:22 | temp=0.3, context=medium, no max\_tokens | ABL | stop | 1 | 0 | 0 | `"[]"` | | 1 | 06:22 | (same) | SF | stop | 1 | 0 | 0 | `"[]"` | | 1 | 06:22 | (same) | FPA | stop | 1 | 0 | 0 | `"[]"` | | 1 | 06:22 | (same) | **IBR** | stop | **1208** | (real) | (real) | **5 real jobs** ✅ | | 2 | 06:43 | temp=0, context=high, no max\_tokens | ABL | stop | 1 | 0 | 0 | `"[]"` | | 2 | 06:43 | (same) | SF | stop | 1 | 8 | 8 | `"[]"` (cited Merriam-Webster for verb “find”) | | 2 | 06:43 | (same) | FPA | stop | 1 | 1 | 1 | `"[]"` ([mmwlegal.com](http://mmwlegal.com)) | | 2 | 06:43 | (same) | IBR | stop | 1 | 0 | 0 | `"[]"` | | 3 | 06:55 | temp=0.2, context=high, max\_tokens=4000 | ABL | stop | 1 | 0 | 0 | `"[]"` | | 3 | 06:55 | (same) | SF | stop | 1 | 0 | 0 | `"[]"` | | 3 | 06:55 | (same) | FPA | stop | 5 | 1 | 1 | ````json\n[]\n```` ([mmwlegal.com](http://mmwlegal.com)) | | 3 | 06:55 | (same) | IBR | stop | 1 | 0 | 0 | `"[]"` | | 4 | 07:07 | temp=0.3, context=medium, no max\_tokens (= run 1 config) | ABL | stop | 1 | 0 | 0 | `"[]"` | | 4 | 07:07 | (same) | SF | stop | 1 | 0 | 0 | `"[]"` | | 4 | 07:07 | (same) | FPA | stop | 1 | 1 | 1 | `"[]"` ([mmwlegal.com](http://mmwlegal.com)) | | 4 | 07:07 | (same) | IBR | stop | 1 | 0 | 0 | `"[]"` | | **Critical:** Run 4 used the _exact same configuration_ as run 1, in which IBR succeeded — but in run 4 IBR returned the same broken `[]` as everything else. The only variable that changed was time of day and Sonar API behavior. | | | | | | | | | ### Diagnostic field summary (sonar-reasoning-pro) | Run | Time (CDT) | Source | finish\_reason | completion\_tokens | citations | search\_results | | --- | --- | --- | --- | --- | --- | --- | | 5 | 07:18 | ABL | stop | 626 | 0 | 0 | | 5 | 07:18 | SF | stop | 469 | 0 | 0 | | 5 | 07:18 | FPA | stop | 607 | 1 | 1 ([mmwlegal.com](http://mmwlegal.com)) | | 5 | 07:18 | IBR | stop | 620 | 0 | 0 | | **The reasoning model’s `` traces are the smoking gun.** Verbatim excerpts from `choices[0].message.content` (first ~500 chars): | | | | | | | | **ABL:** | | | | | | | > ```auto > > The user is asking me to search for ABL (Asset-Based Lending) analyst-level positions in the United States across three title families. However, looking at the search results provided, I see: > > ``` > > \ > Search results: > None > \ > > ```auto > > There are NO search results provided. The search_results section is empty/None. > > ``` **SF:** > ```auto > > ...However, I've been given search results that are empty (no search results provided). The instruction says "None" under search_results. > > ``` **FPA:** > ```auto > > ...the "search results" provided to me are from a law firm website (mmwlegal.com) discussing legal topics like company directors, shareholder disputes, surrogacy, and insolvency. These search results are completely irrelevant to the FP&A job search request. > > ``` **IBR:** > ```auto > > ...The search results provided are "None" - so I need to conduct a search myself. > However, I'm an AI assistant and I don't have the ability to browse the internet in real-time, access LinkedIn, Indeed, Builtin, company career sites, or any other live job boards. > > ``` ## The model is explicitly stating it received no search results. This points to the search-retrieval step of Sonar — not the language-model step — being the failing component. ## Things we ruled out on our side 1. **Prompt content** — IBR’s prompt was unchanged across runs 1–5; it worked in run 1 and broke in runs 2–5. 2. **Temperature** — tested 0, 0.2, 0.3 with same result. 3. **`max_tokens`** — tested with and without an explicit 4000-token budget. 4. **`search_context_size`** — tested both `medium` and `high`. 5. **Model** — tested both `sonar-pro` and `sonar-reasoning-pro`. 6. **`search_recency_filter`** — kept at `"month"` throughout; not changed. 7. **API key / auth** — all responses returned `status=200` with no auth or rate-limit errors. 8. **Account/billing** — no plan changes today. ## Things we suspect are on Perplexity’s side 1. **Search retrieval pipeline returning empty results to the LLM.** The reasoning-model `` traces directly evidence this. 2. **The recurring `mmwlegal.com/author/mmwmmwlegal-com/` citation** appearing in 3 of 5 runs across different prompts is suspicious — looks like a stale cache hit or a bug where one fixture leaks into unrelated queries. 3. The status page ([status.perplexity.com](http://status.perplexity.com)) shows no incidents, but the historical pattern of `sonar-pro` runs returning empty all week (per our cron history) suggests this may be a slow degradation rather than an acute outage. * * * ## Request 1. Investigate whether the search-retrieval step of Sonar is failing to populate `` for our account on `sonar-pro` and `sonar-reasoning-pro`. 2. Confirm whether other accounts are seeing the same pattern. 3. Advise whether there is a workaround we can apply on the request side (different parameter combination, different model, etc.) that would re-engage the search step. Happy to enable additional verbose logging or share full request/response payloads for any of the 16+ failed calls captured today. ## Pre-flight checklist - This is about the Perplexity API, not the Perplexity app, Comet, or web UI. - I have removed API keys, secrets, and private data. - I have included enough information for someone else to reproduce the issue. ## Bug type - Error response or failed request - Incorrect or unexpected model output - Missing, invalid, or stale citations/search results - Streaming issue - Structured output or JSON schema issue - SDK issue - Billing, credits, or rate limit issue - Dashboard, API key, or auth issue - Other API bug ## Affected API area - Agent API - Search API - Sonar API - Embeddings API - SDKs - Dashboard, API keys, or auth - Billing or credits - Not sure ## Summary ## Expected behavior ## Actual behavior ## Minimal reproduction ```bash ``` ## Response or error ```json ``` ## Request details - Endpoint: - Model or preset: - SDK/language/version: - Request ID: - Approximate time and timezone: - Consistent or intermittent: ## Additional context