Amazon Keyword Discovery Tool

Amazon keyword discovery for listing and PPC research

Discover real Amazon autocomplete phrases from seed keywords, compare 20 marketplaces and departments, filter long tails, and keep only new phrases across repeated runs. Every keyword is a full JSON/CSV dataset row with source URL, collection time, best source rank and observed appearance count. No Amazon login or data-provider API key is required.

Use this Actor for autocomplete research, regional wording comparisons and a bounded keyword feed for agents. It does not return actual search volume, CPC, sales estimates, product rows or reverse-ASIN rankings. Appearance counts and wording labels are observations and heuristics, not demand or conversion estimates.

Quick start

{"keywords":["laptop stand"],"marketplaces":["US"],"expansionMode":"alphabet","maxResults":100}

Only keywords is required. Defaults: US, alphabet suffixes, all departments, 100 rows, 300 maximum lookups, automatic direct/residential routing. Direct HTTP is attempted first; proxy traffic, when needed, adds platform usage.

Pricing and total cost

Pay per event plus platform usage. The live Pricing tab is authoritative. All plans receive complete individual JSON/CSV rows. Billing mode depends on the requested workflow:

• Basic: one marketplace, one department, no recursion or persistent monitoring. Localized templates, buyer/question/comparison/preposition expansion and filters are included.
• Advanced: automatically selected by more than one marketplace, more than one department, expansionDepth above 0, or monitorMode enabled. Advanced replaces both Basic pack fees.

Event	Price	When charged
Actor start	$0.00001	Per GB RAM, minimum one
Basic first pack	$0.00005	First up-to-100 delivered keywords, once per run
Basic additional pack	$0.00003	Each additional up-to-10 delivered keywords
Advanced keyword	$0.00025	Each delivered keyword in an Advanced job ($0.25/1,000)

The first Basic pack covers100 rows; further packs cover10 rows. Basic packs do not cross run boundaries; the final partial pack rounds up once. At default memory, Basic event fees are $0.00006 for 1–100 rows, $0.00009 for 101–110, $0.00036 for 191–200, $0.00276 for 1,000 and $0.02976 for 10,000. Advanced event fees are $0.02501 for 100, $0.25001 for 1,000 and $2.50001 for 10,000.

Add actual compute, storage, transfer and proxy usage to every example. Empty, invalid and failed jobs have no result fee; startup and platform usage still apply. Monitoring with no new phrases still incurs startup and usage. Input descriptions and OUTPUT/RUN_SUMMARY identify the selected billing mode. Maximum event cost limits event fees; it does not cap platform usage.

Inputs and limits

Field	Use
keywords	1–50 seed phrases, up to 100 characters each
marketplaces	Multiple markets in one run: US, CA, MX, BR, UK, DE, FR, IT, ES, NL, SE, PL, TR, AE, SA, JP, AU, SG, IN, IE
expansionMode	seedOnly, alphabet, numbers, buyerModifiers, full, custom
customModifiers	Up to 50 suffixes or templates, e.g. best {seed}; custom mode requires at least one
department / departments	One Amazon alias or 1–10 aliases, e.g. aps, kitchen, electronics
expandPrefixAlphabet	Add A–Z before each seed
expandQuestions	Add question prefixes in the market's default language
expandComparisons / expandPrepositions	Add locale-specific comparison/preposition suffixes
expansionDepth	0–3 recursive levels, sharing the same maximum lookup budget
mustIncludeTerms / excludeTerms	Include any matching substring; exclude any matching substring
mustContainSeed / minWordCount / onlyQuestions	Require seed tokens, minimum words or question wording
resultOrder	strongest, alphabetical, source; grouped by seed and marketplace
maxResults	1–10,000 full rows, default 100
maxResultsPerSeed	1–5,000 rows per root seed, default 1,000
maxLookups	1–2,000 requests, default 300; retries may add HTTP attempts
includeSeedKeyword	Keep suggestions equal to the seed, default true
proxyMode	auto, direct, residential
monitorMode / monitorKey	Emit unseen phrases using bounded caller-scoped history

Localized templates cover German, French, Italian, Spanish, Portuguese, Dutch, Swedish, Polish, Turkish, Arabic and Japanese. US, UK, CA, AU, SG, IN and IE use English templates. These are wording aids, not guaranteed grammatical variants or guaranteed non-empty suggestions. Custom modifiers let you supply another language. India uses Amazon’s UK completion host with India’s marketplace ID; the source URL exposes both, and product search links remain amazon.in. Labels such as buyingIntent remain English wording heuristics; question detection also recognizes the built-in question prefixes. Word-count filters use whitespace, which is less useful for Japanese.

Compare departments and discover long tails

{"keywords":["kaffeemühle"],"marketplaces":["DE"],"departments":["aps","kitchen"],"expansionMode":"buyerModifiers","expandQuestions":true,"expandComparisons":true,"expandPrepositions":true,"expansionDepth":3,"maxLookups":120,"maxResults":100,"excludeTerms":["gebraucht"]}

Recursive candidates retain their root seed. Remaining requests are shared between the requested depth levels so early rounds cannot consume the whole recursive budget. Results are unique by marketplace, department and case-normalized phrase. A phrase can legitimately appear in multiple markets or departments. Request caps can limit coverage; source URLs make the resulting scope inspectable.

Monitor new phrases

{"keywords":["coffee grinder"],"marketplaces":["US"],"expansionMode":"alphabet","monitorMode":true,"monitorKey":"weekly-coffee","maxResults":100}

Run the same input again to emit only unseen phrases. The history belongs to the calling Apify user and this Actor/query contract; another caller cannot select that history by guessing a monitor name. Different seeds, filters, request budgets or expansion settings create another scope. Changing row caps or sort order can continue draining unseen results from the same scope. A new monitorKey starts fresh.

History retains the latest 10,000 identities, so evicted phrases may reappear. A history unused for 30 days resets on its next read. No schedule is created. Run a monitor scope sequentially: its best-effort overlap guard is not a distributed transaction or exactly-once guarantee. Named monitoring storage persists between runs; manage or delete it through your account when no longer needed.

Output contract

Each dataset row includes keyword, seedKeyword, matchedSeedKeywords, marketplace, marketplaceDomain, department, depth, position, bestRank, appearanceCount, lookupsForSeed, appearanceShare, firstSeenPrefix, matchedPrefixes, wordCount, characterCount, containsSeed, isSeedKeyword, isQuestion, buyingIntent, keywordType, amazonSearchUrl, sourceUrl, source, collectedAt and optional source metadata (thumbnailUrl, sourceRefTag, suggestionStrategy, null when unavailable).

OUTPUT and RUN_SUMMARY contain outcomes, lookup counts, saved rows, named event counts and estimated event fees. COMPLETE means requested work completed; PARTIAL identifies failures, truncation or insufficient event budget; VALID_EMPTY means no matching new phrases; INVALID_INPUT identifies input errors; UPSTREAM_FAILED and CONFIG_ERROR fail honestly when no trustworthy completion is available. Inspect these records rather than relying only on run status.

Rows are validated before saving. Basic pack charges follow acknowledged row persistence. Advanced rows use the SDK coupled write-and-charge operation; named-event counters reconcile the accepted charge. An ambiguous write/charge or resurrection after a write attempt blocks automatic replay; inspect KEYWORD_WRITE_STATE, existing rows and charges. This favors avoiding duplicate billing over automatic recovery. It is not exactly-once processing. Interrupted jobs can contain useful partial rows.

API and agents

curl -X POST 'https://api.apify.com/v2/acts/khadinakbar~amazon-keyword-discovery-tool/run-sync-get-dataset-items' \
  -H "Authorization: Bearer $APIFY_TOKEN" -H 'Content-Type: application/json' \
  -d '{"keywords":["laptop stand"],"marketplaces":["US"],"maxResults":100}'

For longer jobs start an asynchronous run, wait for completion, then read its dataset and OUTPUT. Export JSON, CSV, Excel or HTML through the Dataset tab. Limited permissions suffice; no owner token or elevated account access is required.

Agent prompt: “Find 100 Amazon US autocomplete phrases for laptop stand, exclude pink, return keyword, source rank, marketplace and Amazon search URL. Check OUTPUT for partial results and report total event fees plus usage.”

For hosted MCP use Apify MCP, inspect this Actor's explicit identifier and schema, call with complete input, then read the returned run/dataset. Public discovery and model preference require separate measurements; availability of an API is not evidence of agent preference. This Actor remains private until its owner publishes it.

Workflow boundaries

For product listings and prices, use Amazon Search Scraper. For Google search demand rather than Amazon autocomplete, consider Keyword Search Volume API. Neither autocomplete rank nor appearance frequency establishes Amazon monthly search volume.

Amazon controls endpoint availability and suggestions. Some valid seeds return none. Department aliases and source thumbnails vary by market. Automatic residential recovery can cost more than direct mode. Use bounded jobs and inspect warnings before scaling.

Responsible use and support

This Actor reads public autocomplete without logging in. Users must review Amazon's Conditions of Use, applicable law and their own data-use policies. Amazon is a trademark of Amazon.com, Inc.; this independent tool is not affiliated with Amazon. Open an Actor issue with the run ID and a non-secret input example for support.