Autocomplete & Keyword Suggest Scraper - seven engines in one run

For SEO keyword research, content planning, PPC negative-keyword lists and product-demand research: autocomplete suggestions from seven services in a single run — Google, YouTube, Bing, DuckDuckGo, Amazon, eBay and Wikipedia — one row per suggestion, with the engine that produced it, the exact query that was sent, and the rank it held in that engine's list.

Autocomplete is the only free, first-party record of what people actually type. Asking one engine tells you what searchers type; asking seven tells you something you cannot get any other way: which engine surfaces a term and which does not. A phrase Google suggests but Amazon does not is informational; one Amazon suggests and Google does not is commercial; one YouTube suggests is a video someone should make.

It expands a seed the way a keyword tool does, and it tells you the cost before it runs away with your bill. A–Z (26 queries), 0–9 (10), question words (12: how, what, why, where, when, which, who, can, is, are, does, should) and prepositions (8: for, with, without, near, like, to, from, vs). All four on turns one seed into 57 queries per engine — three engines is 171 requests and three seeds is 513 — so there is a request limit as well as a row limit. Measured on 2026-09-08, responses came back in 206–370 ms each with no proxy needed.

Every seed gets its fair share of the row limit. Measured on 2026-09-08 before this was fixed, a plain cut left the third of three seeds with 6 rows out of 307 and said nothing about it.

No API key. Export as JSON, CSV or Excel.

What one run gives you

Column	What it is
engine, engineTitle	which service produced the suggestion
seed	the term you asked for
query	the exact string that was sent - with expansion on, this is seed a, how seed and so on
expansion	which expansion produced that query: seed, alphabet, digits, questions, prepositions
suggestion	the suggested phrase
rank	its position in that engine's list (1 = first)
isSeedItself	engines usually repeat the seed as the first suggestion; this marks it
wordCount, characterCount	for sorting long-tail phrases
requestedCountry, requestedLanguage	what was asked for - see the honest note below
engines, engineCount, bestRank	with deduplication on: every engine that produced the phrase, and its best rank

Expansion: one seed into many queries

Autocomplete only answers what you type, so the standard way to go deep is to type more:

Switch	Queries added per seed, per engine
A-Z expansion	26 (seed a … seed z)
0-9 expansion	10 (seed 0 … seed 9)
Question expansion	12 (how seed, what seed, why, where, when, which, who, can, is, are, does, should)
Preposition expansion	8 (seed for, with, without, near, like, to, from, vs)

All four on turns one seed into 57 queries per engine. Three engines means 171 requests, and three seeds means 513. That is why there is a request limit as well as a row limit: the row limit cannot stop requests, because filtering happens after the answer arrives. When the request limit is reached the run stops asking and says so in a free row. Within one seed the same query is never sent twice; two seeds that expand onto the same query will each send it.

What is measured, and what is not

• The seven engines above are the ones that answered when this was measured from Apify on 2026-09-08 (206-370 ms each, no proxy needed). The old Play Store suggest endpoint answered 404 and is not offered rather than being offered and failing.
• 🔴 Country targeting is honest here. Google and YouTube decide the region mostly from the IP the request leaves from, not from the parameter. Measured on the same day: de returned German suggestions, jp returned the same US English list. So the column is called requestedCountry - what was asked for - and this README does not claim it always changes the answer. Bing, DuckDuckGo, Amazon, eBay and Wikipedia ignore the parameter entirely.
• An empty list is an answer, not a failure. All seven services return an empty list for a term with no suggestions. Those queries produce no charged rows, and a free row at the end says how many of the run's requests came back empty - so you can tell "nothing is suggested here" apart from "this was never asked".
• The row limit never drops a whole seed. When more suggestions were collected than you asked for, the rows are taken evenly from every seed and a free row lists how many each seed ended up with. Measured on 2026-09-08 before this was fixed, a plain cut left the third of three seeds with 6 rows out of 307 and said nothing.
• The same suggestion from the same engine is charged once. Expansion makes an engine repeat itself across queries; those repeats are dropped by default, and keepRepeatsPerEngine brings them back if you want to see every query that produced a phrase.
• An answer in a shape this Actor has not measured is reported, not read as "no suggestions". That is the difference between "Amazon has nothing for this" and "Amazon changed its format".
• An engine name that is not one of the seven is rejected before the run starts (Apify validates the input against the list), so a typo costs you nothing instead of producing a half-empty result.

Rows that are never charged

status	What happened
no-results	every request came back with an empty suggestion list - that is the engines' answer
blocked	a service answered with a bot check instead of suggestions
unreadable	a service could not be read, or answered in a shape this Actor has not measured
no-match	suggestions were read and none matched your filters
budget-reached	the run collected more than your row limit, or hit the charge limit you set; the row says how many were left out

Filters

filterKeywords / excludeKeywords (searched in the suggestion, the query and the seed), minWords, maxWords, excludeSeedItself, and deduplicate to collapse the same phrase across engines into one row that lists them all. Filters run before the cap, so maxSuggestions counts the rows you actually receive.

Input

Field	What it does
keywords	the seed terms
engines	which of the seven services to ask
expandAlphabet, expandDigits, expandQuestions, expandPrepositions	how far to expand each seed
country, language	sent to Google and YouTube; language also picks the Wikipedia edition
amazonMarketplaceId	which Amazon storefront to ask (default is Amazon US)
deduplicate, excludeSeedItself, minWords, maxWords, filterKeywords, excludeKeywords	shaping the result
maxSuggestions	hard cap on returned rows, taken evenly from every seed
maxRequests	hard cap on requests sent to the engines - the limit that protects them
keepRepeatsPerEngine	keep every occurrence when one engine repeats a phrase across queries
useProxy	retry through a proxy only if a service answers with a bot check