Keyword Autocomplete Scraper - Google, YouTube, Amazon, Bing

Type a seed keyword and get what four search boxes suggest as you type it: the completions from google, youtube, amazon and bing, in the country and language you pick, one row per suggestion with its rank and the exact text that was typed.

Start typing "coffee grinder" and the box offers to finish it for you: for espresso, walmart, near me. Those completions are real searches, ranked by the engine that serves them. No volumes, though. A suggestion tells you what people type, not how many of them do, and nothing here will give you a monthly search count or a cost per click.

Input	Seed keywords, up to 1,000 per run
Output	One row per suggestion: the phrase, its rank, the engine and the exact text looked up
Ceiling	50,000 suggestions per run
Speed	1,776 suggestions from one seed in about 14 seconds; 50,000 in under 6 minutes
Account needed	None, and no API key
Price	$0.01 per 1,000 amazon suggestions, $0.047 per 1,000 from google, $0.095 from youtube and $0.35 from bing, flat on every plan. The free plan's $5 a month covers about 500,000 amazon suggestions, or about 106,000 from google

🔍 What Keyword Autocomplete Scraper does

It asks each engine for the list its search box shows while you type, and writes down what comes back in order. Every row names the engine it came from, so a phrase two engines both suggest arrives twice, once each, and you can see which of them thinks it matters.

The variations are where the long phrases live. Switch them on and each seed is asked again with a trailing space, with a to z on the end, with 0 to 9, and with question words in front. "coffee grinder" with all four on came back as 1,776 different suggestions from 204 lookups across the four engines. Set expansionDepth to 2 and every new suggestion is asked again as if you had typed it.

A phrase only comes back once per engine. If "coffee grinder electric" turns up under the seed and again under "coffee grinder e", you get it once, from whichever lookup found it first.

📋 What data you get from each keyword suggestion

What you get	Field
The suggested phrase and its place in that list	suggestion, rank
Which engine suggested it	engine
What you typed, and the exact text that went to the engine	seed, query
Which variation produced it, and the letter, digit or word	expansionType, expansion
How many levels deep it was found, on runs that follow suggestions	depth
The country, and the language or the market used	country, language
The Amazon marketplace and department, on amazon rows	marketplace, department
The Google list asked for, whether it is a search or a website, and Google's score, on google rows	searchType, suggestionType, relevance
When the engine answered	scrapedAt

▶️ How to scrape keyword suggestions

1. Open Keyword Autocomplete Scraper and click Try for free.
2. Put your keywords into Seed keywords, one per line.
3. Tick the engines you want, then set Country and Language.
4. Turn on Seed followed by a to z if you want the long tail, set Maximum suggestions, and click Start.
5. Download the dataset as JSON, CSV or Excel, or read it from the Apify API.

Run it with no seeds and you get one sample row, not charged.

💰 How much does it cost to scrape keyword suggestions?

You pay per suggestion delivered, at the rate for the engine that produced it. Flat on every Apify plan, no volume tiers.

Engine	Price
amazon	$0.01 per 1,000 suggestions
google	$0.047 per 1,000 suggestions
youtube	$0.095 per 1,000 suggestions
bing	$0.35 per 1,000 suggestions

On the free plan, the $5 Apify gives you each month covers about 500,000 amazon suggestions, or about 106,000 from google.

A lookup that comes back empty is free. So is a phrase the same engine already gave you in the run, a lookup an engine answered for different text, and the sample row.

📥 What you give it

{
  "seeds": ["coffee grinder", "french press"],
  "engines": ["google", "youtube", "amazon"],
  "country": "GB",
  "language": "en",
  "expandAlphabet": true,
  "expandQuestions": true,
  "maxSuggestions": 3000
}

Field	Default	What it is
seeds	none, the box starts at coffee grinder	One per line, up to 1,000. Written the way you would start typing. Repeats are dropped.
engines	["google"], the box starts at three of them	Any combination of google, youtube, amazon and bing.
country	US	google and youtube lean toward it. bing reads it together with the language. amazon goes by the marketplace instead.
language	en	The language google and youtube answer in. amazon ignores it. google has no separate suggestions in 31 of the 90 and answers them exactly as English: Afrikaans, Albanian, Armenian, Azerbaijani, Basque, Burmese, Galician, Georgian, Hausa, Irish, Javanese, Kazakh, Khmer, Kurdish, Kyrgyz, Lao, Luxembourgish, Macedonian, Maltese, Maori, Mongolian, Pashto, Sinhala, Somali, Tajik, Urdu, Uzbek, Welsh, Xhosa, Yoruba and Zulu.
amazonMarketplace	auto	auto uses the Amazon of your country, or amazon.com where that country has none. Or name one of the 23 yourself.
amazonDepartment	none	Amazon's own short name from its search links: stripbooks, electronics, grocery. A name Amazon does not know returns nothing at all.
googleSearchType	web	web, shopping, images or news. Shopping leans to products: "coffee" led with coffee table and coffee maker where web led with coffee near me. images and news come out close to web.
addTrailingSpace	off	Also asks "coffee grinder " with the space typed. Engines finish that differently.
expandAlphabet	off	26 more lookups per seed and engine. Most of the long phrases come from here.
expandNumbers	off	10 more lookups per seed and engine. Good for model numbers, sizes and years.
expandQuestions	off	One lookup per question word, in front of the seed: "how coffee grinder".
questionWords	13 English ones	Your own list, up to 30. Add best or cheap, or words in the language you picked.
expansionDepth	1	2 also asks each new suggestion as if you had typed it, and 3 goes one level further. That is where the longest phrases come from. maxSuggestions still stops the run.
maxSuggestionsPerQuery	none, keeps everything	Up to 15 come from google, 14 from youtube, 10 from amazon, 25 from bing.
maxSuggestions	5000, the box starts at 300	Suggestions in one run, all engines together, up to 50,000.

If an engine has nothing for a seed, its space, letter and number variations are skipped, since they only make the same text longer. Question words are still asked.

📤 What you get back

A real row from a run on 1 October 2026 with expansionDepth at 2, so this suggestion came from asking an earlier suggestion again:

{
  "suggestion": "running shoes for men sale",
  "rank": 1,
  "engine": "google",
  "seed": "running shoes",
  "query": "running shoes for men",
  "expansionType": "suggestion",
  "expansion": null,
  "country": "GB",
  "language": "en",
  "marketplace": null,
  "department": null,
  "searchType": "web",
  "suggestionType": "query",
  "relevance": 601,
  "scrapedAt": "2026-10-01T00:22:44.101Z",
  "depth": 2
}

Field	How to read it
rank	Its place in the engine's list for that one lookup, starting at 1.
query	The exact text that went to the engine. Here it is an earlier suggestion, "running shoes for men", asked again.
expansionType	seed, trailing space, letter, number, question, or suggestion when a suggestion was asked again.
language	null on amazon rows, which have no language. On bing rows it is the market actually used.
suggestionType	navigation means Google offered a website address rather than a search. google rows only.
relevance	Google's own score for the suggestion. google rows only.

🧾 Reading the output

Row	How to spot it	Charged
A suggestion	_sample is absent	yes
The sample row	_sample: true	no

No diagnostic rows go into the dataset. The status message leads with the split, like "Returned N suggestions for S seeds", then names any engine that stopped answering, any lookup answered for different text, and an Amazon department that returned nothing. RUN_REPORT in the key-value store has it per engine and per seed: lookups planned, made, answered, empty, skipped, repeats dropped and rows returned.

💡 What people use it for

• Building a long-tail keyword list for a page, with a to z on and the question words switched in.
• Finding the questions people actually ask about a product, straight from the question variations.
• Comparing what amazon completes against what google completes for the same seed, which is roughly buying intent against reading intent.
• Watching a seed week by week on a schedule and diffing on suggestion and rank.

From a seed to the pages that rank for it, in three runs:

1. Run this on a seed with expandQuestions on and keep the suggestion values worth writing about.
2. Put the best few into Google Trends Scraper's searchTerms with compareKeywords on, to see which one is growing.
3. Run the winner through Google Search Results Scraper's query to see who already ranks for it.

🚧 What it does not do

• No search volume, cost per click or difficulty score. Only the suggestions and their order.
• It is a snapshot. Engines reshuffle through the day, so tomorrow's run will differ.
• Nothing personalised. The engines see a visitor with no history and no account.
• A few local suggestions slip in. google and bing mix in places near where the run reads from, whatever country you set. In US runs that shows as Washington-area names like "car insurance maryland". A German run checked against the same lookups matched 252 of 254.
• amazon ignores the language, and goes entirely by the marketplace.
• google has no separate suggestions in some languages. For Welsh, Irish, Urdu and 28 more, listed under language above, it gave exactly the English list in a check in September 2026.
• amazon sometimes corrects your spelling and answers the corrected phrase instead. Asked for "does coffee grinder" it answered for "dose coffee grinder". Those are dropped, not charged, and counted in the status.
• bing only answers language and country pairs it runs a market for. English in Germany is not one. Without a match it uses that country's own market and says so, and where it has no market at all it sits the run out.
• When a limit or a budget stops a run early, the rows you keep are whatever arrived first, not an even share across engines.

🧭 Which search scraper do you need?

If you want	Use
The web results page for a keyword: organic, ads, People Also Ask	Google Search Results Scraper
What a search box suggests as people type	This one
Interest in a keyword over time and by region	Google Trends Scraper
News articles with the publisher's real link	Google News Scraper
Image results and the pages they sit on	Google Images Scraper
The ads a brand runs on Google	Google Ads Transparency Scraper
Yandex results by keyword and region	Yandex Search Engine Scraper
Autocomplete from Taobao, in Chinese	Taobao Keyword Suggestions Scraper

❓ Questions people ask

Do I need an API key for keyword suggestions?

No. No key, no account, nothing to apply for.

Why fewer rows than lookups times 15?

Repeats, mostly. Many variations return phrases an earlier lookup already found, and those are not sent twice. Short lists happen too: on amazon.com, "how coffee" returned two suggestions.

How do I get more phrases from one seed?

Turn on Seed followed by a to z first. It is the variation that adds the most, and question words are the next best. Then set Levels of suggestions to follow to 2.

Does it work in other alphabets?

Yes. Seeds in Japanese, Korean, Russian and Hebrew came back in their own scripts. The a to z variation adds Latin letters, which suits some writing systems more than others.

Can I call it from code or connect it to an AI assistant?

Yes. The API tab has ready-made code for Python, JavaScript and the command line. For Claude, ChatGPT or another MCP client, connect https://mcp.apify.com/?tools=fetch-actor-details,dami_studio/keyword-autocomplete-scraper. Either way the run happens on your Apify account at the same price.

Is collecting search suggestions legal?

Suggestions are public, aggregated phrases with no personal data in them. Apify's write-up on the legality of web scraping is a good starting point, and we are not lawyers.

🆘 If something breaks

Open the Issues tab on the actor page. Send the run ID and the seeds you used. The status message and RUN_REPORT usually say which engine fell short.