Google SERP + AI Overviews API
Pricing
from $4.00 / 1,000 search completeds
SerpAPI-style Google search JSON: AI Overview text + citations + organic results in one clean response per query. Batch up to 1000 queries per run.
AI Overview
ai_overview
Optional
AI Overview answer: text_blocks in document order — paragraph and heading (snippet), list (list: [{title, truncated, truncation_reason, inline_link_sources?}]), table (headers + rows of cell text) — plus references (citation title, link, snippet). resolved_by marks how a deferred overview was resolved (folsrch/refetch/poll). List items and paragraphs carrying truncated=true lost words in Google's response: truncation_reason is inline_link_cut (the sentence stops at an inline entity link; Google streams the tail client-side and it is NOT in the response — inline_link_sources holds the sources Google attached to that link: [{url, title, source}]), or a text-level cut (unclosed_bracket, ends_on_comma, ends_on_open, ends_on_joining_word). A block with truncation_detected has any item cut; a bare missing full stop is not a cut. null when Google shows no AI Overview.
Organic results
organic_results
Optional
Organic search results: position, title, url, displayed_link, snippet.
People also ask
people_also_ask
Optional
People Also Ask questions shown by Google for this query.
Search metadata
search_metadata
Optional
status (Success/Partial/Empty/Error), created_at, total_time_taken_ms, google_url, warning, drift_evidence (KV key of saved raw HTML when status is Empty/Partial), partial_recovered (true when organics were recovered by re-fetch), citations_missing (true when Google served the AI Overview text without its citations and they could not be recovered — such rows are status Partial and are NOT charged), parse_suspect (true when Google's AI Overview markers were in the response but no answer text could be extracted — probable Google markup change; such rows are status Partial and are NOT charged), ai_mode_detected (true only when the answer was served from the AI Mode surface, i.e. not inside an AI Overview block; classic AI Overview answers never set it), ai_mode_excluded (true when excludeAiMode dropped an AI Mode answer: ai_overview is null, status Success, charged), thin_recovered (true when the AI Overview was thin — ended in a dangling heading or under 600 characters — and one fresh SERP fetch returned a richer answer, which replaced it), thin_recovery_attempted (true when the AI Overview was thin, one fresh fetch was made, and the original answer was kept because the fresh one was not richer or was AI Mode), modules_present (which modules actually parsed: organic_results, ai_overview, people_also_ask, knowledge_graph, top_stories, local_results, discussions_and_forums). truncation_detected (true when any AI Overview list item or paragraph lost words — cut at an inline entity link, on an unclosed bracket, a trailing comma or a joining word; a missing full stop alone does NOT set it), truncation_evidence ([{tail: last 80 chars of the cut text, reason: truncation_reason}, max 10]), folsrch_evidence (KV key of the saved answer HTML when captureFlaggedFolsrch is on and the row ended truncated — the raw folsrch fragment when resolved_by is folsrch, else the shell), drift_evidence_folsrch (KV key of saved folsrch-fragment HTML on citation-missing rows when the fragment is a different surface than the shell).
Search parameters
search_parameters
Optional
Echo of the query: q, gl, hl, location.
Knowledge graph
knowledge_graph
Optional
Entity panel: title, description, facts, source (Wikipedia link). null when Google shows no knowledge panel.
Top stories
top_stories
Optional
Top news stories for the query: title, url, source. null when Google shows none.
Local results
local_results
Optional
Local pack results: title, rating, reviews. null when Google shows none.
Discussions & forums
discussions_and_forums
Optional
Reddit/Quora discussion blocks: title, url, source. null when Google shows none.
Selector strategy
selector_strategy
Optional
Which parser strategy fired per module: primary (current Google classes), fallback (stable anchors), or structural (class-free heuristics). Drift telemetry — primary everywhere = healthy.
