YouTube Keyword Suggestions Scraper

Get localized YouTube autocomplete suggestions without an API key. Suggestions are not measured search volumes, competition scores, extracted video tags, or guaranteed rankings. Coverage varies by locale and time and is not exhaustive.

Quick start

{"keywords":["minecraft"],"language":"en","country":"US","alphabetExpansion":false,"maxItems":5}

An empty API object uses these same defaults. Existing API input with a single keyword string remains supported. Up to 50 seeds of 1–200 characters are accepted. Invalid types, blank seeds, unsupported modes and invalid locale syntax fail before requests. Language accepts codes such as en, ja, zh-TW; country accepts two-letter codes such as US, JP. The endpoint determines locale support.

Inputs and limits

• mode: suggest only.
• keywords: case-insensitive repeated seeds are processed once.
• language / country: defaults en / US.
• alphabetExpansion: default false; true also requests seed plus Latin letters a–z, even for non-Latin seeds.
• maxItems: integer 1–5000, default 5; global across all seeds and expansions, including existing rows on resurrection.

Default change: previously omitted expansion/limit meant true/500. Specify those values explicitly to retain that breadth. Processing checks charging and result capacity before each request and write. Each request times out after 15 seconds, without automatic retries. Larger inputs require an appropriate overall run timeout. A small limit can stop before later seeds.

Output

Each row is one unique (keyword, seed) association, case-insensitive and Unicode-normalized. The same keyword under different seeds remains multiple results, preserving the existing contract. First occurrence wins across expansions. Existing fields remain: keyword, seed, language, country, rank, source, scraped_at. Added fields are stable id, exact query, and source_url. Rank is the 1-based position in that query's suggestion response, not search rank or popularity.

Illustrative output, not a current capture:

{"keyword":"minecraft survival","seed":"minecraft","language":"en","country":"US","rank":2,"source":"youtube","scraped_at":"2026-09-05T12:00:00+00:00","query":"minecraft","id":"<SHA-256 of keyword, seed and locale>","source_url":"https://suggestqueries-clients6.youtube.com/complete/search?client=youtube&ds=yt&q=minecraft&hl=en&gl=US"}

Export the dataset as JSON or CSV. Existing dataset rows, including old-format rows, are the delivery journal on resurrection. Already delivered associations are not pushed/charged again in normal resurrection. Keep inputs unchanged on resurrection; use a new run for a fresh snapshot. A crash between platform storage and billing is not an application-level transactional guarantee.

Failures and coverage

A valid empty response succeeds with zero result events. HTTP errors (including 403/429), timeouts and malformed responses fail the run; delivered partial rows remain available. Error records are not billable rows. RUN_SUMMARY in key-value storage records complete, limit_reached, or source_error, row count, and request count for this invocation. Complete means all requested queries were processed, not exhaustive YouTube coverage. Previously empty requests may repeat after resurrection.

No API key, proxy, browser fallback or paid third-party service is required. Public endpoints may change or block requests. Report reproducible input and run ID through the Actor issue channel without private data.

Pricing

Existing prices are unchanged: $0.0005 per delivered keyword-plus-seed result and the listed start event of $0.00005 per GB. The start event has a 1 GB minimum, so at 512 MB the start component is $0.00005. Illustrative event totals: 5 results $0.00255; 500 results $0.25005; 30 separate 50-result runs $0.7515. Actual pricing depends on current Console configuration and tier. These are event calculations, not measured infrastructure costs or margin guarantees. No result charging capacity means no further source work; infrastructure can still consume resources during a timeout.