A keyword research API expands one seed term into the queries people actually type. This collector drives Google's own autocomplete: alphabet soup (seed + a…z), question words, prepositions and comparison modifiers, then returns every distinct suggestion Google served, each tagged with the probe that surfaced it and Google's own relevance score.
The distinction from a volume tool is the important one. This is not estimated monthly searches — it is the literal list of completions Google offers real users, which is demand as expressed rather than demand as modelled. Autocomplete is a plain JSON endpoint, so a run never touches a browser, and duplicate suggestions appearing across several probes are delivered and billed once.
$0.0002 per delivered keyword, up to 500 keywords per run. Nothing delivered means nothing charged.
You send a seed and choose modes. Each mode is a family of probes: alphabet appends a through z, question modes prefix how/what/why/when, preposition modes insert for/with/near/without, comparison modes add versus and alternatives. The collector fires all of them against Google's autocomplete endpoint from an exit in your country, merges the results, de-duplicates them and returns up to 500 distinct keywords.
Two fields make the output more useful than a flat list. probe records exactly which query produced the suggestion, so you can see the shape of the demand rather than just its content — a term that only ever appears behind "how to" is an informational cluster, not a commercial one. relevance is Google's own score for the suggestion in that context, which is a better ordering signal than the alphabetical accident of the probe.
seed is required and should be a real head term, not a long phrase — the expansions come from what follows it. modes selects the probe families; more modes mean broader coverage and more delivered keywords, so it is also the main cost lever. country and lang matter a great deal: autocomplete is heavily localised, and the German completions of a term are not translations of the English ones.
Each keyword is one row. keyword is the suggestion, seed echoes what you asked for, and probe is the exact string that surfaced it — the field that turns a list into a map, because grouping by probe separates informational from commercial from comparative intent without any classification model. mode is the probe family, relevance is Google's score, and type distinguishes plain queries from entity suggestions.
$0.0002 per delivered keyword ($0.20 per 1,000). Nothing delivered means nothing charged, and the $2 monthly free credit covers roughly 10,000 keywords here. Volume tiers take up to 30% off.
$0.0002 per delivered keyword — 20 cents per 1,000, the cheapest unit in the catalogue. A full expansion of one seed across all modes might deliver 300 keywords, so six cents. Researching a thousand seeds is roughly $60 for something that would take a person weeks.
De-duplication happens before billing: a suggestion that surfaced from six different probes is delivered once and charged once.
Keyword Planner is Google's own tool and it has one thing this cannot give you: search volume estimates, straight from ad auction data. If your work depends on volume figures, use it. Two caveats decide whether you can. It requires a Google Ads account, and for accounts without meaningful spend it returns bucketed ranges rather than numbers — the familiar "1K–10K" bands that make precise prioritisation impossible. Its API is the Google Ads API, which means OAuth, a developer token and an approval process before the first call.
The two datasets are complements, not substitutes. Autocomplete gives you the real phrasing, including long-tail questions that never appear in a volume tool because their individual volume rounds to zero — and long-tail questions are exactly what wins featured snippets and AI citations. Volume tools tell you how big a term is; autocomplete tells you what people actually say.
Hitting Google's autocomplete endpoint is genuinely easy, and that is why so many scripts do it. What they run into:
The collector is that script with the exit pool, the probe families and the merge already done — and priced so that running it is cheaper than maintaining it.
Group by probe to separate question intent from comparison intent, then write a page per cluster instead of a page per guessed keyword.
Question-mode probes surface the exact phrasings people use, which is what FAQ blocks and featured-snippet targeting need.
Run the same seed per country and language. The differences are usually larger than a translation would suggest, which is precisely the point.
Expand a seed here at $0.0002 per keyword, then push the shortlist through the search results collector to see who ranks.
Up to 500 distinct keywords per run, which is a full multi-mode expansion of one seed. More seeds are more runs, parallel up to your plan's rate limit — 60 requests/minute on pay-as-you-go, up to 1,200 on the top tier. Note that one run fires many autocomplete probes internally; that is the collector's concern, not a charge against your rate limit per probe.
Legally: autocomplete suggestions are publicly served data and collecting public data is generally lawful in most jurisdictions, though Google's terms restrict automated access. Suggestions occasionally contain personal names because people search for them — apply your own filtering before storing anything that identifies an individual. Not legal advice.
Google's Keyword Planner is free with a Google Ads account but gives bucketed volume ranges to low-spend accounts and requires the Google Ads API for programmatic access. Here there is a free allowance of $2 a month with no card, which at $0.0002 per keyword is about 10,000 delivered keywords — several dozen full seed expansions.
No, and it is important to be clear about that. It returns the actual suggestions Google serves, with Google's own relevance score for ordering. Volume estimates come from ad auction data and only Keyword Planner and tools licensing similar data have them. The two are complementary: volume tells you size, autocomplete tells you phrasing.
$0.0002 per delivered keyword — 20 cents per 1,000 — the cheapest unit in the catalogue. Duplicates across probes are merged before billing, so you pay once per distinct keyword. Volume tiers take up to 30% off.
Probe families. Alphabet appends a through z to the seed; question modes prefix how, what, why and when; preposition modes insert for, with, near and without; comparison modes add versus and alternative phrasings. More modes mean broader coverage and more delivered keywords.
Yes, and you should — autocomplete is heavily localised. Set country and lang together. The German or Italian completions of a term are genuinely different demand, not translations of the English list, which is why running the seed per market beats translating a keyword list.
It depends on the seed and the modes, up to the 500-per-run cap. A broad commercial term across all modes usually lands in the low hundreds; a narrow technical term produces far fewer, and you pay only for what is delivered.
No. Autocomplete is a plain JSON endpoint, so a run is a set of requests and a merge. That is why it is the cheapest collector available.
Skip the marketing. QuantumProxies.io publishes a machine-readable site map, Markdown for every page, a free MCP server, and APIs billed from the same balance as your proxies.
Connect in one command: npx -y quantumproxies-mcp