Keyfern — keyword ideas & autocomplete scraper for Google, YouTube, Bing and Amazon

Keyword research from real searches: give it a seed keyword, get back hundreds of long-tail keyword ideas that people actually type into Google, YouTube, Bing, Amazon and DuckDuckGo. Every idea is deduped across engines and comes with its search intent, a topic cluster, the engines that suggested it and a 0–100 signal score. Optionally, the best ideas also get a Google Trends verdict: rising, falling, spiking or stable. It costs $0.001 per keyword idea ($1 per 1,000) and $0.003 per trend verdict.

First try: the prefilled input (seed standing desk, best 8 ideas, no trend check) runs in about 15 seconds and costs about $0.008.

Why Keyfern

• Five autocomplete engines in one run. Most autocomplete scrapers read Google only. Keyfern asks Google Search, YouTube, Bing, Amazon and DuckDuckGo (pick any mix) and merges the answers. A phrase suggested by several engines is a stronger idea, and the score shows it.
• The whole long tail, not ten suggestions. Each seed is expanded A–Z (standing desk a … standing desk z), with question words (how, what, why, can, is, does, …) and with prepositions and comparisons (for, with, without, near, vs, …). One seed on 3 engines = 147 lookups in a few seconds. In our test, standing desk on all five engines gave 1,041 unique ideas.
• Sorted and labelled, ready to use. Each idea is tagged question, commercial, transactional, local, comparison or informational, grouped into a cluster (converter, treadmill, ikea…), and ranked. A per-seed SUMMARY lists the top clusters, the top questions (FAQ and blog topics) and the intent mix.
• Trend direction, not invented search volume. Autocomplete doesn't reveal search volume, and Keyfern doesn't make one up. Instead it can check the top N ideas on Google Trends (the same engine as our Swellmeter Actor) and tell you whether interest is growing.
• New-keyword alerts. monitor: true returns only ideas you haven't received before for your seeds, plus ideas whose trend verdict flipped, so a weekly scheduled run costs only for what's new.

Real example (live run, 2026-09-23, seed standing desk, Google + YouTube + Bing + Amazon, US):

196 of 196 autocomplete calls answered in about 6 seconds. Top ideas: standing desk chair · standing desk ikea · standing desk with drawers · standing desk with storage · standing desk with wheels. Biggest clusters: walking, treadmill, gaming, keyboard. Google Trends: "standing desk chair" rising (+74% last quarter of the year vs the first), "standing desk ikea" stable.

What you get (one record per keyword idea)

Field	Meaning
keyword	The suggested search phrase (lowercased, deduped)
seed	The seed it came from
signalScore	0–100 relative ranking signal: how many engines suggest it, how high, how often, and whether it shows up for the bare seed. Not search volume.
intent	question, commercial, transactional, local, comparison or informational (inferred from the wording)
cluster, modifier	The main word the idea adds to the seed, and all the words it adds
sources, sourceCount	Engines that suggested it
bestPosition, hits	Best rank in any suggestion list (1 = top), and how many lookups returned it
suggestedOnSeedQuery	true if the engine suggests it for the bare seed (the strongest ideas)
foundVia	How it was found: seed, alphabet:c, question:how, preposition:vs…
trend	Only for the top N ideas: { status, direction, trendChangePct, recentChangePct, peakDate, exploreUrl }
language, country, fetchedAt	Run settings and time
change, changedFields, previousTrendDirection	Monitor mode only: new or changed (trend verdict flipped)

The key-value store record SUMMARY has, per seed: idea count, intent mix, top 15 clusters, top 20 ideas, top 15 questions, and per-engine success counts.

Input

{ "seeds": ["standing desk", "cold brew"], "sources": ["google", "youtube", "bing", "amazon"], "trendCheckTop": 5 }

Option	Default
seeds		Up to 50 seed keywords
sources	google, youtube, bing	Any of google, youtube, bing, amazon, duckduckgo
language, country	en, US	Language and country of the suggestions (also picks the Amazon marketplace and the Trends region)
expandAlphabet, expandQuestions, expandPrepositions	all on	Which expansions to run
mustContainSeed	true	Drop suggestions that don't contain the seed's words
maxIdeasPerSeed	300	Best N ideas kept per seed, which also caps the cost (300 ideas = $0.30)
minSignalScore	0	Keep only ideas at or above this score
intents	all	Keep only some intents, e.g. ["question"] for FAQ research
trendCheckTop	5	Google Trends verdict for the N best ideas per seed (0 = off, max 25)
trendTimeRange	today 12-m	Trends window: past 30 days, 90 days, 12 months or 5 years
monitor	false	Only ideas not returned by an earlier monitor run, plus trend-verdict changes (see below)
monitorKey		Optional name for the monitor's saved state, to keep several seed sets apart

Monitor mode: only new keyword ideas since the last run

Turn on monitor for scheduled runs. Keyfern remembers, per seed, language and country, every idea it already returned and outputs only:

• ideas it has never returned for that seed: "change": "new";
• ideas whose Google Trends verdict changed (both runs had a verdict): "change": "changed", "changedFields": ["trend.direction"] and "previousTrendDirection".

Shape of a monitor record (other fields as in the normal output below):

{ "seed": "standing desk", "keyword": "…", "change": "changed", "changedFields": ["trend.direction"], "previousTrendDirection": "stable", "trend": { "status": "ok", "direction": "rising", "…": "…" } }

Unchanged ideas are left out and not charged. The first monitor run returns everything ("change": "new"). Ideas that drop out of autocomplete for a while aren't reported again when they come back. maxIdeasPerSeed applies before the comparison, so "new" means "new in your top N". The SUMMARY record gets new/changed/unchanged counts. State lives in your named key-value store keyword-ideas-monitor (one record per monitorKey, or derived from the seeds, sources, language and country); delete the record to start over.

Ready-made inputs (first try in one click)

No account yet? See real results first on the free demo page: keyword ideas for 20 popular topics.

Paste one into the JSON tab of the input form (or send it to the API) and press Start. Each is sized so the first run is quick and cheap; you pay only for results (Apify's compute is included in the price).

Preset	What you get	Cost
8 long-tail ideas for one seed	8 ideas with intent, cluster, engines and signal score	$0.008
Questions people ask	8 question-style searches (how/why/can…)	$0.008
Amazon buyer searches + 1 trend verdict	5 product searches + a rising/falling verdict for the top one	$0.008

8 long-tail ideas for one seed

{"seeds": ["standing desk"], "maxIdeasPerSeed": 8, "trendCheckTop": 0}

Questions people ask

{"seeds": ["solar panels"], "intents": ["question"], "maxIdeasPerSeed": 8, "trendCheckTop": 0}

Amazon buyer searches + 1 trend verdict

{"seeds": ["air fryer"], "sources": ["amazon"], "maxIdeasPerSeed": 5, "trendCheckTop": 1}

Same thing from a script (returns the results directly):

curl -s -X POST "https://api.apify.com/v2/acts/cybermax~keyword-ideas/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H 'content-type: application/json' -d '{"seeds": ["standing desk"], "maxIdeasPerSeed": 8, "trendCheckTop": 0}'

Output (trimmed, real record)

{
  "seed": "standing desk",
  "keyword": "standing desk chair",
  "signalScore": 100,
  "intent": "informational",
  "cluster": "chair",
  "sources": ["amazon", "bing", "google", "youtube"],
  "bestPosition": 1,
  "hits": 11,
  "suggestedOnSeedQuery": true,
  "foundVia": ["alphabet:c", "seed", "question:when"],
  "trend": { "status": "ok", "direction": "rising", "trendChangePct": 74.3, "recentChangePct": -25.9, "peakDate": "2026-04-12" }
}

Pricing

Event	Price	Charged when
keyword-idea	$0.001	One unique idea is returned (capped by maxIdeasPerSeed)
trend-checked	$0.003	One idea gets a Google Trends verdict. No-data and failed checks are free

Typical costs: one seed with 300 ideas and 5 trend verdicts = $0.315. 100 question ideas only (intents: ["question"], maxIdeasPerSeed: 100, trendCheckTop: 0) = $0.10. Set a maximum charge per run in Apify and Keyfern stops cleanly when it's reached.

When NOT to use Keyfern

• You need search volume, CPC or keyword difficulty numbers. Autocomplete doesn't expose them; use a paid-data SEO tool. Keyfern gives you real phrasing and trend direction, never estimated volumes.
• You need the full Trends picture for a few keywords (timeline, regions, related queries): use Swellmeter — Google Trends.
• Very long seeds (6+ words) get few suggestions; use 1–3 word topics.

Notes

Keyfern reads the public autocomplete endpoints that each engine's own search box uses: no login, no API key, no personal data. Suggestions change over time and by country and language. Calls go out directly and fall back to Apify Proxy when an engine throttles.

Run it on a schedule

Weekly new-keyword report: Apify Console → Schedules → Create new → cron 0 8 * * 1 (8:00 every Monday, or 0 8 * * * daily) → Add Actor → cybermax/keyword-ideas with your input (or a saved task) including "monitor": true. Or with the API:

POST https://api.apify.com/v2/schedules?token=YOUR_TOKEN
{
  "name": "weekly-keyword-ideas",
  "cronExpression": "0 8 * * 1",
  "timezone": "America/New_York",
  "isEnabled": true,
  "actions": [{
    "type": "RUN_ACTOR",
    "actorId": "cybermax/keyword-ideas",
    "runInput": { "body": "{\"seeds\": [\"standing desk\", \"cold brew\"], \"maxIdeasPerSeed\": 300, \"trendCheckTop\": 5, \"monitor\": true}", "contentType": "application/json" }
  }]
}

With monitor mode a scheduled run only pays for new ideas and changed trend verdicts. To get notified, add a webhook, Slack or email notification on Run succeeded in the Actor's Integrations tab.

Integrations

Make: Apify app → "Run an Actor" (cybermax/keyword-ideas, input JSON, "Run synchronously" on) → "Get dataset items" → Google Sheets "Add a row" / Airtable.

Zapier: Apify app → "Run Actor" action (or "Finished Actor Run" trigger for your schedule) → "Find dataset items" → Google Sheets / Notion / Slack.

n8n: Apify node ("Run actor and get dataset"), or an HTTP Request node:

POST https://api.apify.com/v2/acts/cybermax~keyword-ideas/run-sync-get-dataset-items?token=YOUR_TOKEN
{"seeds": ["standing desk"], "intents": ["question"], "maxIdeasPerSeed": 50, "trendCheckTop": 0}

Google Sheets: Apify's Google Sheets integration (Actor → Integrations), or paste into a cell to pull the last run as CSV:

=IMPORTDATA("https://api.apify.com/v2/acts/cybermax~keyword-ideas/runs/last/dataset/items?format=csv&fields=keyword,signalScore,intent,cluster,change&token=YOUR_TOKEN")

Python (pip install apify-client):

from apify_client import ApifyClient
client = ApifyClient("YOUR_TOKEN")
run = client.actor("cybermax/keyword-ideas").call(run_input={"seeds": ["cold brew"], "sources": ["google", "youtube", "amazon"], "maxIdeasPerSeed": 100})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item["signalScore"], item["intent"], item["keyword"])

JavaScript (npm i apify-client):

import { ApifyClient } from 'apify-client';
const client = new ApifyClient({ token: 'YOUR_TOKEN' });
const run = await client.actor('cybermax/keyword-ideas').call({ seeds: ['standing desk'], intents: ['question'], trendCheckTop: 0 });
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items.map((i) => i.keyword));

AI agents (MCP): Keyfern is a tool in any MCP client via Apify's hosted MCP server, https://mcp.apify.com/?tools=cybermax/keyword-ideas (header Authorization: Bearer <your Apify token>). The records are compact and flat, so an agent can pick blog topics, FAQ questions, product keywords or video titles directly. Claude Desktop / Cursor config:

{ "mcpServers": { "keyfern": { "url": "https://mcp.apify.com/?tools=cybermax/keyword-ideas", "headers": { "Authorization": "Bearer YOUR_TOKEN" } } } }

Example prompt: "Find 50 question keywords about cold brew that are rising on Google Trends."

Also from CyberMax

• Swellmeter — Google Trends: the full Trends picture (timeline, related and rising queries, regions, trending now) for your best ideas.
• Linkheft — Domain authority: how strong the sites ranking for your keywords are.
• SnapScout — Website screenshots: screenshot and text of the pages ranking for your keywords.
• DomainDNA — Tech stack & domain intel: what the competing sites are built with.
• Gripeline — App reviews: the words real users use in App Store and Google Play complaints.

All tools: Boardroom Buys · HireHeat · HireHeat Jobs · Hollerdex · Printwren · Logolark

Changelog

• 2026-09-23: Monitor mode (monitor, monitorKey: only keyword ideas not seen before for your seeds, plus trend-verdict changes), schedule & integration recipes.

Made by CyberMax.