Google Search Results via Serper API — Organic, PAA & Related

Give it a query. Get back organic results, People Also Ask, and related searches — confirmed live, every time, not a maybe.

How it works: this Actor does not scrape Google. It calls the Serper Google Search API (a paid third-party API) with your query and maps the JSON it returns into the dataset format below. No browser, no proxies, no HTML parsing.

What this actually delivers (and how we know)

Before writing this README we ran 13 distinct queries against two separate Serper accounts, including the textbook trigger queries every SERP tool demos with — "eiffel tower," "how tall is the eiffel tower," "elon musk," "barack obama," commercial buy-intent terms like "buy dyson v15 vacuum" and "nike air max." We inspected the raw, unmapped upstream response for several of them, not just our own output.

Result: organic, people_also_ask, and related_searches populated on every query. ads, shopping, knowledge_graph, answer_box, and ai_overview populated on zero of the 13 — including the ones that should be unmissable, like a direct factual question ("what year was the eiffel tower built") or an unambiguous public figure ("barack obama"). That's not a bug in this actor; it's what the upstream API actually returns for these accounts. So that's not what this actor is sold on.

• Organic results — title, url, description, position, and (when Google includes them) sitelinks, per-result attributes, and a date
• People Also Ask — the expandable question list
• Related searches — the query-refinement suggestions at the bottom of the page

What you get back, per query — real, captured output

This is the shape of an actual run against "elon musk", trimmed to save space (the real run returned 9 organic results). Result titles, URLs and snippets are replaced with placeholders here; a real run returns the actual values:

{
  "query": "elon musk",
  "results": [
    { "title": "<result 1 title>", "url": "<result 1 URL>", "description": "<result 1 snippet>", "score": 1 },
    { "title": "<result 2 title>", "url": "<result 2 URL>", "description": "<result 2 snippet>", "score": 0.889 },
    { "title": "<result 3 title>", "url": "<result 3 URL>", "description": "<result 3 snippet>", "score": 0.778 }
  ],
  "organic": [
    { "title": "<result 1 title>", "url": "<result 1 URL>", "description": "<result 1 snippet>", "position": 1 }
  ],
  "ads": [],
  "people_also_ask": [
    { "question": "What is Musk diagnosed with?" },
    { "question": "Does Musk believe in God?" },
    { "question": "Is Elon Musk a trillionaire?" },
    { "question": "Does Elon have 14 children?" }
  ],
  "related_searches": ["Elon Musk net worth", "Elon Musk children", "Elon Musk book", "Elon Musk wife", "Elon Musk car", "Elon Musk Twitter", "Elon Musk age", "Elon Musk money"],
  "knowledge_graph": null,
  "answer_box": null,
  "shopping": [],
  "ai_overview": null,
  "latency_ms": 1978,
  "checked_at": "2026-09-18T07:48:45.997Z"
}

Note knowledge_graph, answer_box, and shopping are null/[] here — for "elon musk," a query that triggers a knowledge panel when searched on Google directly. That's the honest, representative case, not the exception.

The optional fields — one honest mention

ads, shopping, knowledge_graph, answer_box, and ai_overview are present in every response's schema and populate when the upstream response includes them — they're not stripped out or faked empty. We just don't sell them, because across 13 test queries on two accounts, none of them ever did. If you find a query where one of these reliably populates, that's useful data — the code already handles it correctly, it just hasn't been observed happening.

Input

Field	Type	Required	Notes
queries	array of strings	yes	Up to 1,000 per run. One charge event per query executed.
num	integer	no (default 10)	Organic results requested per query, 1-100.
country	string	no	2-letter country code (Google gl). Confirmed working — verified geo-targeted results (Spain, Japan, France) in live testing.
language	string	no	2+ letter language code (Google hl). Confirmed working — verified Spanish-language results for a Spanish query.
page	integer	no (default 1)	1-based result page. Confirmed working — page 2 returns genuinely different results than page 1.
device	string	no	Forwarded to Serper as-is. Accepted without error but no observed effect in testing — not confirmed to do anything.

Every field is pre-filled in the Console form (queries: ["web scraping tools"], num: 3, country: us, language: en, page: 1, device: desktop), so pressing Start runs a small, fast, cheap search.

Errors never fail the run

The Actor always finishes SUCCEEDED and always writes at least one dataset item. Anything that goes wrong is reported as a record with an error code, a human-readable message, and usually a hint (and the upstream HTTP status when there is one), instead of a failed run:

error	Meaning
missing_api_key	SERPER_API_KEY is not set on the Actor — no search was run.
no_queries	Input had no non-empty queries.
too_many_queries	More than 1,000 queries; the first 1,000 ran.
upstream_auth_error	Serper rejected the key (401/403). The run stops after this one record.
upstream_credits_exhausted	The Serper account is out of credits. The run stops after this one record.
upstream_rate_limited	Serper returned 429 twice (one automatic retry is made).
upstream_unavailable	Serper returned a 5xx twice.
upstream_error	Any other non-2xx from Serper; message carries its reply.
request_timeout / request_failed	The request timed out (30 s) or the network failed.
invalid_upstream_response	Serper's reply could not be read.
charge_limit_reached	The run's maximum charge was reached; remaining queries were skipped.
no_output	Safety net: the run produced nothing else.

Only a genuine crash (a bug, out-of-memory) exits FAILED. Error records are not billed as query-executed events; under pay-per-event, Apify's standard per-dataset-item event (apify-default-dataset-item) still applies to every item pushed, error records included.

Pricing

$0.01 per query — $10.00 per 1,000 queries. Pay-per-event: one query-executed event per query, whatever num is, so at the default num: 10 that works out to $1.00 per 1,000 results. Queries that return nothing aren't charged.

What this is not

Not a multi-engine tool — it calls Google, through Serper, and nothing else. Not a relevance engine — score is rank-derived, same as the rest of the category. Not a source of ads, shopping listings, knowledge-graph data, answer boxes, or AI Overviews — those fields exist in the schema and will populate if the upstream ever returns them for your query, but treat that as a bonus, never an expectation.