Google Autocomplete Scraper: Suggestions, YouTube, Amazon

This Actor collects search autocomplete suggestions for keyword research. Give it seed keywords, pick where to ask (Google web search, YouTube, Google Shopping, Amazon, Bing), and it returns one row per unique suggestion with the query that found it, its position, and Google's relevance score where Google publishes one. It reads the public suggest endpoints these sites use for their own search boxes, so there is no login, no API key and no browser. A typical run of a few hundred suggestions finishes in seconds.

• Five sources: Google web, YouTube, Google Shopping, Amazon and Bing. Each has its own suggestion list, so the same seed gives different keywords on each.
• Expansion modes: the seed alone, the seed plus a to z, the seed plus 0 to 9, a letter before the seed, question words before it (how, what, why, can, is, does, where, which, when, who) and comparison words after it (vs, versus, for, near, with, without, and, or, to, like).
• Second round: with depth 2, the best first-round suggestions are sent as queries of their own, which finds longer, more specific phrases.
• Country and language: set the Google country and language, the Amazon marketplace (19 countries) and the Bing market.
• Deduplicated: every row is a unique suggestion for its seed and source, and timesSeen tells you how many different queries returned it.

What data you get

One item per unique suggestion for each seed keyword and source:

Field	What it holds
suggestion	The suggested search phrase
seed	The seed keyword you entered
source	google, youtube, shopping, amazon or bing
query	The exact text sent when this suggestion first appeared, for example the seed plus a letter
expansion	seed, alphabet, number, prefix letter, question, preposition or second round
position	Position in the list returned for that query, starting at 1
relevance	Google's relevance score for the suggestion (Google, YouTube and Shopping only; higher is more prominent)
type	QUERY or ENTITY for Google sources, KEYWORD for Amazon, empty for Bing
description	Short label Google attaches to entity suggestions, empty for most rows
country, language	Codes used for the request
depth	1 for the first round, 2 for a suggestion found by expanding a first-round suggestion
timesSeen	How many queries in the run returned the same suggestion
scrapedAt	When the data was collected (UTC)

How to use it

1. Enter one or more seed keywords, one per line.
2. Choose the sources. Google web search is selected by default.
3. Set the country and language (for example de and de for German results).
4. Pick expansion modes. "Seed alone" returns about 10 to 15 suggestions per source. Adding a to z or question words returns hundreds.
5. Optionally set depth 2 for a second round, and set "Max suggestions per seed and source" to cap the cost.
6. Run it, then download the dataset as JSON, CSV or Excel, or read it through the API.

Example input:

{
    "keywords": ["web scraping"],
    "sources": ["google"],
    "country": "us",
    "language": "en",
    "expansions": ["seed", "alphabet"],
    "depth": 1,
    "maxSuggestions": 100
}

Example output (from a real run, shortened):

[
    {
        "suggestion": "web scraping python",
        "seed": "web scraping",
        "source": "google",
        "query": "web scraping",
        "expansion": "seed",
        "position": 1,
        "relevance": 1250,
        "type": "QUERY",
        "description": null,
        "country": "us",
        "language": "en",
        "depth": 1,
        "timesSeen": 1,
        "scrapedAt": "2026-11-02T07:46:15.745Z"
    },
    {
        "suggestion": "web scraping app",
        "seed": "web scraping",
        "source": "google",
        "query": "web scraping a",
        "expansion": "alphabet",
        "position": 3,
        "relevance": 650,
        "type": "QUERY",
        "description": null,
        "country": "us",
        "language": "en",
        "depth": 1,
        "timesSeen": 1,
        "scrapedAt": "2026-11-02T07:46:15.745Z"
    }
]

A run also writes a SUMMARY record to the key-value store: suggestions and queries per seed and source, request counts, and any queries that failed with the reason.

How much does it cost?

The Actor uses pay-per-event pricing:

• $0.0002 per suggestion (one unique row in the dataset)
• $0.0005 per run start (the standard Actor start fee)

Queries that return nothing, repeats of a suggestion already found, and failed requests are not charged. Example: 1,000 suggestions cost $0.20 plus the $0.0005 start fee, so $0.2005. The default run of 100 suggestions costs about $0.0205. You can also set a maximum cost for a run in the run options, and the Actor stops when it is reached.

Use cases

• SEO and content planning: build a long-tail keyword list for a topic, then group it by the question words and comparison words people type.
• YouTube topic research: see what people type into YouTube's search box for a subject before choosing a video title.
• Amazon and shopping research: find product phrases shoppers type, by country, for listings and ads.
• Paid search: feed the suggestions into a keyword planner as candidate terms.
• Market and language checks: compare the same seed across countries and languages to see how search habits differ.

FAQ

Does it return search volume? No. Autocomplete endpoints publish no volume or CPC. You get the phrases people actually type, ordered the way the source ranks them, plus Google's relevance score. Use it to find candidates, then check volume in a planner.

Why is there no Google News source? Google's news suggest list was identical to the web list in testing, so it would only return duplicate rows. It is left out.

How many suggestions can I expect? Google web, YouTube and Shopping return up to about 15 per query, Amazon 10 and Bing between roughly 10 and 25. Turning on every expansion mode sends 83 queries per seed and source. In one test run, a seed with every mode plus a depth 2 round (103 requests, 5 seconds) returned 1,217 unique Google suggestions.

Are the question words translated? No. The question words and comparison words are English. For other languages, use the a to z modes, or write your own seeds that already contain the question word (for example "wie funktioniert").

Which Amazon countries work? us, ca, mx, br, gb, de, fr, it, es, nl, se, pl, tr, ae, sa, eg, jp, au and sg. If you pick another country, Amazon is skipped and the other sources still run. Amazon India's suggest endpoint returned errors in testing and is not included.

Do the letter modes work for every script? The letters are a to z and digits 0 to 9, which work well for Latin scripts. For Japanese, Arabic or Cyrillic seeds, the seed-alone and question or comparison modes work, but the a to z modes mostly return suggestions with a Latin letter appended.

What if a source blocks requests? The Actor retries with backoff on rate limits. If eight queries in a row fail for a source, it stops using that source for the rest of the run, keeps the other sources, and lists the errors in SUMMARY. If nothing could be loaded at all, the run fails with the error. Failed queries are never added to the dataset or billed.

Is the data accurate? Autocomplete is personalized and changes over time. This Actor sends no cookies or sign-in, so you get the generic list for the chosen country and language, which can still differ slightly between runs.

Other Actors by this developer

• Google Trends scraper for interest over time and related queries
• Google Trending Now for what is rising in searches today
• YouTube channel and search scraper for the videos and channels behind a keyword
• Google Ads Transparency scraper to see who advertises on your keywords
• Website contact extractor for emails and phone numbers from company sites