Keyword Ideas & Questions — Google, YouTube, Bing, Amazon

Give it a topic, get back hundreds of long-tail keywords and questions that real people type into Google, YouTube, Bing and Amazon. Each keyword is ranked by a cross-engine demand score, tagged with its search intent and grouped into topic clusters, so the list arrives sorted and ready for content planning.

$1 per 1,000 keywords. No API keys, no subscription, no accounts to connect.

What you get

One seed such as yoga mat with the default settings returns about 1,500 unique suggestions in under 5 seconds. The best 500 come out ranked:

demandScore	keyword	intent	cluster	engines
100	yoga mat towel	informational	towel	Amazon, Bing, Google, YouTube
99	yoga mat with strap	informational	strap	all 4
97	yoga mat cleaner	informational	cleaner	all 4
93	yoga mat for kids	informational	kids	all 4
87	yoga mat near me	local	near me	Amazon, Bing, Google
83	yoga mat price	transactional	price	Bing, Google, YouTube
83	yoga mat alternative	comparison	alternative	all 4

Questions are tagged too: which yoga mat is best, how thick should a yoga mat be, are yoga mats waterproof, can yoga mats be recycled. They work well as blog topics, FAQ entries and "People also ask" targets.

One dataset row

{
  "keyword": "yoga mat towel",
  "seed": "yoga mat",
  "demandScore": 100,
  "intent": "informational",
  "cluster": "towel",
  "sources": ["amazon", "bing", "google", "youtube"],
  "sourceCount": 4,
  "timesSuggested": 12,
  "bestPosition": 1,
  "googleRelevance": 1251,
  "wordCount": 3,
  "country": "us",
  "language": "en"
}

Cluster summary

The run's key-value store gets an OUTPUT record with every cluster: its size, its top keywords and its intent mix.

{"cluster": "best", "keywords": 31, "avgDemandScore": 54,
 "topKeywords": ["which yoga mat is best", "best yoga mat for hot yoga", "best yoga mat brands"],
 "intents": {"commercial": 22, "question": 7, "local": 1, "transactional": 1}}

How the demand score works (and what it is not)

Autocomplete does not publish search volume, and this actor does not invent one. demandScore is a relative rank inside your run. It is built only from signals the engines actually return:

• How many engines suggested the keyword (40%). Google, YouTube, Bing and Amazon each rank completions by what people type, so a phrase that all four suggest is in demand everywhere.
• How many different probes surfaced it (30%). A keyword that comes back for yoga mat, yoga mat t, best yoga mat and yoga mat for is popular.
• Its best position in any suggestion list (20%).
• Google's own suggest relevance score, when Google returns one (10%).

If you need exact monthly volumes and CPC, take the shortlist from this actor to a Keyword Planner–based tool. Checking 200 good keywords costs far less than checking 5,000 raw ones.

How it finds keywords

For every seed and every engine you pick, the actor sends autocomplete probes:

• the seed itself, then the seed followed by a space;
• question words in front: how, what, why, which, can, is… They are built in for English, French, German, Spanish, Italian, Portuguese and Dutch, so a French run asks comment, pourquoi, quel…;
• modifiers: best, cheap, vs, for, with, without, near me, price, review, alternative, free… (in French: pas cher, avis, comparatif…);
• the alphabet: seed a … seed z, and in deep mode seed 0 … seed 9 plus the question words with a trailing space.

The results are lowercased and de-duplicated across engines, scored, labelled and sorted. Your seed itself is not returned or charged.

Intent labels

question, commercial (best, review, top), transactional (price, buy, cheap, coupon), comparison (vs, alternative), local (near me), navigational (login, app, download, official) and informational (everything else). The rules cover the same seven languages.

Clusters

Each keyword joins the topic word it shares with the most other keywords: bag, kids, price, near me… Plurals and accents are folded, so epais and épais land together, and two-word phrases such as near me and pas cher stay whole. A keyword whose extra words appear nowhere else goes to (other).

Input

Field	Default	Notes
keywords	["keyword research"]	Seed topics, up to 50.
sources	google, youtube, bing	Add amazon for product research. Amazon covers US, CA, UK, DE, FR, ES and IT, and falls back to US elsewhere.
country	us	Two-letter code: gb, de, fr, es, it, ca, au, in…
language	en	Two-letter code.
depth	standard	quick ≈ 30 probes per seed and engine, standard ≈ 55, deep ≈ 80.
maxKeywords	500	The best-ranked keywords come first, and you pay per delivered row.
minDemandScore	0	Drops weaker keywords.
intents	all	For example ["question"] or ["commercial", "transactional"].
questionsOnly	false	A shortcut for FAQ and blog-topic research.

Example: French product research on Amazon and Google.

{
  "keywords": ["tapis de yoga"],
  "sources": ["google", "amazon"],
  "country": "fr",
  "language": "fr",
  "maxKeywords": 300
}

Pricing

Pay per event:

• $0.001 per keyword row delivered, which is $1 per 1,000 keywords;
• Apify's standard run-start event: $0.00005.

There are no other fees. Set Max total charge on the run and the actor stops cleanly at your limit. A run with the defaults (500 keywords) costs about $0.50.

Use it from an AI agent

The actor works as a tool in Apify's MCP server: https://mcp.apify.com?tools=suitable_sunspot/keyword-ideas-autocomplete. An agent can ask "find question keywords about home insurance in France" and get back ranked rows with intent and cluster.

Use cases

• Content plans: pick a cluster, write one article per cluster, and answer the questions inside it.
• Product and marketplace listings: with Amazon enabled, the keyword list shows what shoppers type: materials, sizes, audiences.
• YouTube topics: YouTube suggestions show what people want to watch.
• Local and multilingual SEO: run the same seed per country and language.
• PPC seed lists: filter intents to commercial and transactional.

Limits and notes

• The data is autocomplete suggestions only. There are no search volumes, CPC or difficulty scores.
• The engines change their suggestions over time and by location, so two runs a week apart can differ slightly.
• A probe that fails is retried twice and then skipped. It never fails the run, and the log reports how many failed. If many fail, turn on the proxy setting.
• Unofficial. Not affiliated with Google, YouTube, Microsoft or Amazon; the names only describe where the suggestions come from.