Keyword Ideas: Google, YouTube & Bing Autocomplete

What does Keyword Ideas: Google, YouTube & Bing Autocomplete do?

Keyword Ideas turns a few seed keywords into hundreds of long-tail keyword ideas. It reads the suggestion boxes of Google, YouTube, Google Shopping and Bing, expands every seed with letters, digits, question words, prepositions and your own modifiers, and merges everything into one clean list. Each idea is saved once per seed, with every engine that suggested it, its position in each engine's list and the query that found it. An idea is charged once per run, even when several seeds produce it.

Two things set it apart from a plain autocomplete scraper:

• Cross-engine consensus. An idea that two or three engines suggest independently is a stronger signal than one that appears in a single box. Ideas are ranked by how many engines suggested them, then by their best position.
• Trend direction. Optionally, the best ideas of each seed get a Google Trends verdict: rising, flat or falling, from 12 months of weekly interest. It uses the same HTTP-only engine as our Google Trends Scraper.

You pay once per unique idea in the whole run, never for duplicates.

Why do consensus and trend direction matter?

Autocomplete tells you what people type, but not which ideas deserve your time. When Google, Bing and YouTube all complete "coffee grinder" with "for espresso", that phrase is worth a page, a video or an ad group. A keyword that only one box shows may be a local quirk. Trend direction adds the missing question: is interest growing or fading? Falling ideas can be skipped before you write for them, and rising ones can be prioritised.

Use cases

• SEO: build topic clusters and long-tail lists, and find the questions people ask (isQuestion).
• Content planning: turn a topic into an editorial calendar of ideas that more than one engine suggests.
• YouTube creators: use the YouTube box to find video titles that people already search for.
• E-commerce: use the Google Shopping box to find product phrases, brands and "vs" comparisons.
• PPC: harvest keyword candidates, and use engineCount and the trend to pick the strongest.
• AI agents and automation: flat, predictable items with a stable schema, callable through the Apify API or MCP.

How to use it

Open the Input tab, add your seeds and choose the engines. The default input expands one seed on Google and Bing with the seed itself, a to z and twelve question words (39 queries per engine). It finishes in about two minutes.

Quick start

{
    "seeds": ["coffee grinder"],
    "engines": ["google", "bing"],
    "country": "US",
    "language": "en",
    "expansions": ["alphabet", "questions"]
}

Expansions

• alphabet: "seed a" to "seed z" (26 queries per engine). On by default.
• digits: "seed 0" to "seed 9" (10 queries).
• questions: "how seed", "what seed", "why", "can", "are", "is", "does", "should", "which", "where", "when", "who" (12 queries). On by default.
• prepositions: "seed for", "seed with", "seed near", "seed vs", "seed without", "seed to", "seed like" (7 queries).
• custom: your own customModifiers, for example best, cheap, 2026, used as a prefix and as a suffix.

The seed itself is always queried. Set maxQueriesPerSeed to cap the first-level queries per seed and engine (0 means no cap).

Depth 2

{ "seeds": ["standing desk"], "expansions": ["questions"], "depth": 2, "recursionTopN": 10 }

Depth 2 takes the best recursionTopN ideas of each seed and queries each one again as plain text. It finds ideas two steps away from the seed. These items carry depth: 2 and the idea they came from in parent.

Trend direction

{ "seeds": ["coffee"], "engines": ["google"], "includeTrend": true, "trendTopN": 5, "trendGeo": "US" }

The top trendTopN ideas of each seed (default 20, at most 100) are looked up on Google Trends. The direction is rising when the average of the last 13 weeks is more than 15% above the 13 weeks before, falling when it is more than 15% below, and flat otherwise. The running week is left out because it is not finished. Each lookup takes about five seconds.

Output

Results are available as JSON, CSV, Excel and more from the Output tab and the dataset API. One item per unique idea and seed:

{
    "keyword": "coffee grinder for espresso",
    "seed": "coffee grinder",
    "sources": [
        { "engine": "google", "position": 1, "relevance": 601, "query": "coffee grinder" },
        { "engine": "youtube", "position": 3, "relevance": 601, "query": "coffee grinder" },
        { "engine": "bing", "position": 11, "relevance": null, "query": "coffee grinder" }
    ],
    "engineCount": 3,
    "firstQuery": "coffee grinder",
    "modifierType": "seed",
    "depth": 1,
    "parent": null,
    "duplicateOf": null,
    "country": "US",
    "language": "en",
    "wordCount": 4,
    "isQuestion": false,
    "status": "ok",
    "scrapedAt": "2026-10-10T12:00:00.000Z",
    "trend": {
        "direction": "falling",
        "averageValue": 35.6,
        "recentAverage": 22.77,
        "previousAverage": 33.46,
        "changePercent": -32,
        "peakMonth": "2026-02"
    },
    "trendStatus": "ok"
}

• Ideas are lower case with single spaces and no trailing punctuation. The seed itself is never an idea.
• duplicateOf is null for the first copy of an idea. If the same idea turns up under a later seed, it is saved again with that seed's own sources and trend, and duplicateOf names the seed where it was first saved. These repeats are free.
• sources has one entry per engine, its best-ranked occurrence. relevance is Google's own score and is null for the other engines.
• modifierType tells how firstQuery was built: seed, alphabet, digit, question, preposition, custom or recursion.
• trend and trendStatus appear only with includeTrend. They are null for ideas that were not looked up. Trend values are Google's relative 0-100 index for that idea, not search counts.
• The Output tab also links the ENGINE_STATUS record: for every engine its HTTP requests, ideas, status and a message.

Pricing

You pay per result, in plain words. There is no separate charge for compute.

Event	Free / Bronze	Silver	Gold	Platinum	Diamond
Run start (one per GB of memory, minimum one)	$0.003	$0.003	$0.003	$0.003	$0.003
idea: one per unique idea saved	$0.001	$0.0009	$0.0008	$0.0008	$0.0008
trend-signal: one per trend direction found	$0.002	$0.0018	$0.0016	$0.0016	$0.0016

Apify charges the run start once per GB of memory, minimum one. The default 256 MB counts as one, and the Actor never needs more. The Pricing tab of this page is authoritative: if it differs from this table, it wins.

What is charged, exactly:

• An idea is charged once per run. If the same idea appears under several seeds, the first copy is charged and the later copies (duplicateOf is set) are free.
• A trend signal is charged once per idea and run, when its Google Trends direction was found. Ideas without Google Trends data, and lookups that were blocked or failed, are free. A trend that is reused for a repeat of an idea is not charged again.
• Engines that were skipped or blocked cost nothing, and nothing is charged for results that were not saved.

Worked examples at the Free/Bronze price:

• 300 ideas: 300 x $0.001 = $0.30 (plus the $0.003 start, about $0.30).
• 300 ideas and 20 trend signals: $0.30 + 20 x $0.002 = $0.34.

If you set a maximum cost per run, the Actor stops cleanly when it is reached and keeps what it saved.

FAQ

Why are Amazon, eBay and DuckDuckGo not included?

Their robots.txt files disallow their suggestion endpoints, and this Actor respects robots.txt. It also checks robots.txt at runtime for every host it calls. If a path ever becomes disallowed, that engine is skipped and reported as robotsDisallowed.

How many ideas will I get, and how long does it take?

Google, YouTube and Google Shopping answer from one host, and Bing from another. Every host gets one request at a time with a pause of 1.6 to 2.5 seconds, so a query takes about two seconds. The default input sends 39 queries per engine, which is about 1.5 to 2 minutes per seed per host. Seeds run one after the other, and the engines of different hosts run in parallel. So the default input (Google and Bing) takes about 1.5 to 2 minutes per seed, and every additional engine on the Google host (YouTube, Shopping) adds about the same again. Trend lookups add about five seconds per idea. Google returns up to 15 suggestions per query and Bing 12 to 24, and the merged list is typically a few hundred ideas per seed. maxIdeasPerSeed (default 500, at most 5000) keeps the best-ranked ideas and stops an engine once enough are collected. Seeds are limited to 50 per run.

What happens if the run is about to hit its time limit?

Fifty seeds on the default input take more than an hour, which is the default run timeout. The Actor reads the platform's time limit and, when less than 90 seconds are left, starts no new seed, query or trend lookup. It saves what it collected, and ends with a status message that says it stopped early. Raise the run timeout, or use fewer seeds or expansions, to collect the rest.

Is it legal?

The Actor reads public, aggregated suggestions that the engines show to anyone typing in a search box. It collects no personal data and uses no logins or cookies from user accounts. You remain responsible for using the data in line with each engine's terms and the laws that apply to you. This Actor is not affiliated with or endorsed by Google, YouTube or Microsoft.

What do the statuses mean?

Every saved idea has status: ok. Problems are reported per engine in the ENGINE_STATUS record and in the final status message:

• ok: the engine answered normally.
• robotsDisallowed: its robots.txt disallows the suggestion endpoint, so it was not called.
• blocked: the engine rate-limited the run (HTTP 403 or 429 three times in a row), or the host returned a web page instead of JSON, or redirected the request, which is how bot checks and consent pages are served. A web page or redirect stops every engine on that host: Google, YouTube and Google Shopping share one. The other engines carry on. The Actor never tries to solve or evade a CAPTCHA or bot check.
• error: three queries in a row failed, or its robots.txt could not be loaded (the engine is skipped rather than guessed at).

An engine that was stopped stays stopped if the run is restarted by the platform.

trendStatus is ok (charged), noData (Google Trends has no data for the idea, free), blocked or error (free). Long-tail ideas often have too little volume for Google Trends, so noData is normal. After three CAPTCHA or consent pages from Google Trends in one run, no further trend lookups are made and the rest are blocked. If no idea could be collected because every engine was blocked, the run fails with an explanation instead of ending empty.

Why are some Bing suggestions missing?

For many seeds Bing appends templated places to its answers, for example "coffee grinder best in london" or "... in madrid". When such a suffix repeats across several different queries and its place is a known city, country or region, it is template noise, not an idea, and it is dropped. "near me" is dropped only when it repeats across three or more alphabet letters for a seed that is not local (it is kept for "dentist"). Suffixes like "in python" or "in excel" are never dropped. A suggestion that Google, YouTube or Google Shopping also produced for the seed is never dropped. Suggestions that only repeat the query you sent are removed too.

Why does Google Trends use a proxy from the first request?

By design. The Trend direction option uses the same HTTP-only client as our Google Trends Scraper, which runs through an Apify datacenter proxy session on the platform from its first request, because Google Trends rate-limits Apify's shared IPs much sooner without one. The suggestion hosts are different: they are called directly, and only after repeated HTTP 429 or 5xx answers does a host move to an Apify datacenter proxy session (never residential). If a proxy session does not connect, the Actor rotates to a new session once and then goes direct, instead of waiting out timeouts.

What happens if the platform migrates or restarts my run?

The Actor saves its progress to the run's key-value store. After a migration it continues where it stopped: queries that were answered are not sent again, trend answers that were already found (ok or no data) are not looked up again, and ideas that were already saved are neither saved nor charged a second time. Trend lookups that were blocked or failed are tried again, except after the bot-check cap. Once a seed's list of ideas to save is fixed, a restarted run saves exactly that list, so it never exceeds maxIdeasPerSeed.

What are the limits?

• Suggestions show what people type, not how often. There are no search volumes.
• Trend direction needs enough Google Trends data. It is relative interest per idea and is not comparable between ideas.
• The engines can change their endpoints at any time. Report problems through the Issues tab.

Changelog

• 1.0 (2026-10-10): first release with Google, YouTube, Google Shopping and Bing autocomplete, alphabet, digit, question, preposition and custom expansions, depth 2, cross-engine consensus ranking, Bing template-noise filter, Google Trends direction, robots.txt checks, migration-safe resume, a run time-limit guard and pay-per-event pricing (one charge per unique idea per run).