# Serper — no-answer behaviour on Q48–Q52 (unanswerable-by-design)

Source: raw responses at `serper/input-3/outputs/raw/search/Q48.json` – `Q52.json`, run `2026-08-16-r2`,
all `http_status: 200`, 0 errors. `answer_text`/`citations` are `null`/`[]` in all five — Serper's
`/search` endpoint never synthesizes an answer, so this measures whether its *ranked snippets* would
mislead a downstream synthesis step.

**Verdict: worse than SerpAPI on this axis.** Serper stays generic on 2 of 5 queries but returns
specific, wrong, or fabricated-sounding claims on the other 3.

- **Q48** (Serper's own SOC 2 date) and **Q51** (Olostep index size) — generic SOC2-explainer content
  (`secureframe.com`, `sprinto.com`) and generic Olostep docs/marketing, no specific number. Low risk.
- **Q49** (Jina ARR) — `getlatka.com`: *"Jina AI Revenue 2025: $6.3M Est. ARR, $37.4M Raised"* directly
  beside `growjo.com`: *"Jina AI's estimated annual revenue is currently $17.5M per year."* Same
  contradiction SerpAPI surfaced, confirming it's an index-wide artifact, not a SerpAPI-specific one.
- **Q50** (Exa Enterprise price) — `nimbleway.com/blog/exa-alternatives`: *"Four usage-based tiers are
  available: Free with 1,000 API credits/month; Pay As You Go at $0.008 credit; Project at $30/4,000
  API credits/month"* — a tier structure that does not match Exa's real usage-based billing at all,
  stated as plain fact with no hedge.
- **Q52** (Linkup index size) — `linkedin.com/company/linkup-platform`: *"crawl the web, clean it,
  index it... pulls billions of pages a day."* Same vague-but-specific-sounding marketing claim
  SerpAPI returned for this query.

**Disposition:** No-answer behaviour is now scorable for Serper. Confirmed: it never abstains or
returns nothing (it's a pure search API), and on 3/5 probes its snippets carry unverified/incorrect
specifics that read as confident facts. Cite this file + the five raw JSON files above.