Search Autocomplete Scraper: Google, Bing, DuckDuckGo and Pinterest keyword suggestions

Get the autocomplete suggestions of Google, Bing, DuckDuckGo and Pinterest in one run, for any keyword, language and country, without an API key, an account or a browser. Pinterest suggestions are what people type into Pinterest search, the keywords that Pinterest SEO (pin titles, descriptions and boards) is built on. Expand each keyword with A-Z, 0-9, question words in the chosen language and your own words before or after it, and see in how many engines each suggestion appears. Schedule it daily and get only the suggestions that are new: the earliest sign of a rising search.

A second mode fixes misspelled names: give it nubanc, disney plas or netflx and it returns nubank, disney plus and netflix, decided by a vote between the three engines. A name that the engines also suggest as written is never "fixed".

At a glance

• Price per 1,000 suggestions, Free plan: US$ 0.50 (down to US$ 0.425 on Gold and above)
• Price per 1,000 names checked (canonicalize mode): US$ 1.00
• Fee per run start: None
• Engines in one run: Google, Bing, DuckDuckGo, Pinterest (pick any)
• Any language and country: Yes (language, country)
• A-Z, 0-9, question and custom word expansion: Yes
• Engine count per suggestion, optional merge across engines: Yes
• Only new suggestions since the last run (onlyNew): Yes, free when nothing is new
• Misspelled name to the right spelling, with the engines that agree: Yes
• Error rows (engine down, bad input, run timeout): Free, with an errorCode; one engine failing never stops the others

One real row

From a test run on 2026-10-06 (empty input, so the example keyword iphone 17, language en, country us):

{
  "type": "suggestion",
  "keyword": "iphone 17",
  "suggestion": "iphone 17 pro max",
  "engine": "google",
  "position": 2,
  "seed": "iphone 17",
  "expansion": "base",
  "engines": ["google", "bing", "duckduckgo"],
  "engineCount": 3,
  "language": "en",
  "country": "us",
  "scrapedAt": "2026-10-06T21:44:41Z"
}

With compareWithPreviousRun or onlyNew every row also gets "isNew": true | false and "firstSeenAt".

A row of canonicalize mode (input nubanc, same test day):

{
  "type": "canonical",
  "input": "nubanc",
  "canonical": "nubank",
  "changed": true,
  "corrections": [{"from": "nubanc", "to": "nubank", "votes": 3, "responsesAgreeing": 3,
                   "engines": ["google", "bing", "duckduckgo"], "votesForOriginal": 0}],
  "sourcesAgreeing": 3,
  "enginesAgreeing": ["google", "bing", "duckduckgo"],
  "language": "en",
  "country": "us",
  "scrapedAt": "2026-10-06T21:44:45Z"
}

What you can do with it

• SEO and content: hundreds of long-tail keywords and real questions per topic in seconds ("what ai agents are free", "how ai agents work"), for article ideas, FAQ sections and keyword clusters.
• Trend spotting: schedule your brand, product or category every morning with onlyNew and get only the suggestions that appeared since yesterday.
• Marketplaces and e-commerce: see what shoppers type after your product name, in each country and language, before you write titles and ads.
• Brand monitoring: watch what people attach to your brand ("... login", "... refund", "... vs ...") across three engines at once.
• Pinterest marketing: the searches people type into Pinterest for your niche, to title pins and boards with the words that get found.
• Data cleaning: a column of brand or product names written five different ways becomes one spelling, with how many engines agree, for CRM, survey and transcription cleanup.

Input

Field	What it does
keywords	Seed keywords, one per line (iphone 17, ai agents). In canonicalize mode, the misspelled names
mode	suggestions (default) or canonicalize
engines	Any of google, bing, duckduckgo, pinterest (when left out: the first three)
language	Two-letter code (en, es, de, fr, it, ja...), or a locale like en-GB
country	Two-letter code (us, gb, de, in, au...)
expandAlphabet	Also query keyword a ... keyword z
expandNumbers	Also query keyword 0 ... keyword 9
expandQuestions	Also query the question words of the language before the keyword
prefixes, suffixes	Your own words before or after each keyword (best, vs, near me, 2026)
dedupeAcrossEngines	One row per suggestion instead of one per engine and suggestion
maxItems	Max suggestion rows per keyword, 0 = no limit
contextWords	Canonicalize mode: topic words that help pick the right name (bank, streaming)
compareWithPreviousRun	Add isNew and firstSeenAt to every row
onlyNew	Deliver only suggestions not delivered before for the same keyword
proxyConfiguration	Default Apify Proxy (datacenter)

Example: every question people ask about a topic, merged across engines:

{
  "keywords": ["ai agents", "world cup 2026"],
  "expandQuestions": true,
  "expandAlphabet": true,
  "dedupeAcrossEngines": true
}

Example: daily new suggestions for a brand in the UK:

{
  "keywords": ["taylor swift"],
  "language": "en",
  "country": "gb",
  "onlyNew": true
}

Example: fix a list of misspelled names:

{
  "mode": "canonicalize",
  "keywords": ["nubanc", "disney plas", "netflx", "spotfy", "amazn prime"]
}

Output fields

Suggestion rows (type: "suggestion"): keyword (your input), suggestion, engine, position (1 = top of that engine's list), seed (the exact query typed, for example how ai agents), expansion (base, alphabet, number, question, prefix, suffix), engines and engineCount (every engine that suggested it for this keyword, through any expansion), isNew and firstSeenAt (with the comparison on), language, country, scrapedAt.

Each engine's suggestion comes once per keyword, at the first query that returned it. When an engine just echoes your own query back, that row is dropped and not charged.

Canonical rows (type: "canonical"): input, canonical, changed, corrections (each word changed, with its votes and the votes the original word got), sourcesAgreeing and enginesAgreeing (engines that voted for the new spelling, or that confirm the original when nothing changed).

Error rows (type: "error", free): errorCode is one of ENGINE_FAILED (that engine did not answer any query of the keyword; the other engines are still delivered), ENGINE_PARTIAL, INVALID_INPUT, TIMEOUT_TOO_SHORT, PARTIAL_TIMEOUT, NOT_REACHED (the run timeout came before that keyword) and UNEXPECTED.

Pinterest keyword suggestions

Add pinterest to engines to get the suggestions Pinterest shows while someone types in its search box: the long-tail keywords for pin titles, pin descriptions, board names and Pinterest ads. Every expansion works the same way (A-Z, 0-9, questions, your own words), and engineCount shows which Pinterest keywords are also searched on Google and Bing.

{
  "keywords": ["home decor", "wedding nails"],
  "engines": ["pinterest"],
  "expandAlphabet": true
}

• About 10 to 12 suggestions per query; with A-Z, about 250 to 300 Pinterest keywords per seed keyword.
• language sets Pinterest's locale: de returns German searches (pizzateig rezept), pt Portuguese (pizza caseira). In our tests Pinterest gave the same list for every country of one language, so country changes little there.
• Only search suggestions are returned. Profiles that Pinterest mixes into its suggestion box are left out.
• When the engines field is left out, the run reads Google, Bing and DuckDuckGo, so existing scheduled runs and API calls keep their rows and cost. The Console form starts with all four checked.

How the name fixer decides

Each word of a name (3+ letters) is typed alone and together with its neighbour in the name and your context words. Every engine answer is one vote: a word from the suggestions that sounds like yours (same letters after folding accents, double letters, ph/f, y/i, c/k) gets a vote, worth double when it came with context. The answers that contain your own word vote for keeping it. The new spelling wins only when it has more votes than the original, so real names, plurals and suffixes (AppFile stays AppFile, not Appfiles) are left alone. Works for names in Latin script.

Daily monitoring

With onlyNew (or compareWithPreviousRun) the Actor keeps, per keyword, language, country and set of engines, the suggestions it delivered, in a named key-value store in your account (search-autocomplete-scraper-state). The first run is the baseline: everything is delivered and isNew stays empty. From the second run on, isNew: true marks what was not there before. Suggestions not seen for 60 days are forgotten. Delete the store to start over.

Autocomplete changes a little between requests, so a handful of new rows on a quiet day is normal; a suggestion that shows up in two or more engines (engineCount) is the stronger signal.

Speed and cost

One keyword without expansion is 1 query per engine, about 25 to 40 rows across Google, Bing and DuckDuckGo, plus about 11 from Pinterest. In a test on 2026-10-20, one keyword with A-Z in all four engines gave 1,252 rows in 5 seconds (Pinterest 272, Bing 527, Google 248, DuckDuckGo 205). With A-Z, 0-9 and questions it is about 50 queries per engine and usually 500 to 1,500 rows (fewer with dedupeAcrossEngines). In our tests 15 keywords with every expansion delivered about 10,000 suggestions in 22 seconds. Use maxItems to cap the rows per keyword.

Notes

• Google's suggestions also depend a bit on where the request comes from; country sets the engine parameters and the residential fallback runs from that country.
• Pinterest is asked at a gentler pace than the search engines, so it adds a few seconds per keyword.
• DuckDuckGo suggestions are often close to Bing's; the engines field shows where each one came from.
• The run stops a few seconds before its timeout and keeps everything already delivered; the keywords it did not reach get a free NOT_REACHED row.
• Only public autocomplete text is collected. No personal data.

Changelog

• 2026-10-20: Pinterest added as a fourth engine (engines: ["pinterest"]), with every expansion, engine count, merge and only-new tracking. Default engines when the field is left out stay Google, Bing and DuckDuckGo.
• 2026-10-06: First version: Google, Bing and DuckDuckGo suggestions, A-Z, 0-9 and question expansion, name fixer.