Google Autocomplete Scraper - Keywords & Scores
Pricing
Pay per event
Google Autocomplete Scraper - Keywords & Scores
Pull Google's own search suggestions for any keyword, in any language and country, with the relevance score Google publishes for each one. Expand a seed by alphabet, question words, prepositions or digits.
Pull Google's own search suggestions for any keyword, in any language and country, and get the relevance score Google publishes for each one.
No API key, no browser, no login.
The scores are the point
Google does not just return a list of completions. It returns a number saying how strongly each one belongs to what you typed:
| suggestion | relevance |
|---|---|
| best laptops | 1250 |
| best laptops for college students | 1200 |
| best laptop 2026 | 750 |
| best laptop for home use | 601 |
That is Google's own ranking, not a position we counted. Most tools that read this endpoint throw it away and hand you an ordered list, which looks the same until you try to compare suggestions from two different seeds: rank 3 of one seed and rank 3 of another are not comparable, and 750 against 601 is.
Where a score is missing it is empty, never filled in from the position. A number derived from the rank would look exactly like a real score.
Choose the client deliberately
Google answers three different clients with three different things, and the run summary reports which you used:
| client | suggestions | scores |
|---|---|---|
| chrome | 15 | yes |
| firefox | 10 | no |
| psy-ab | 10 | no, but marks the matched part |
chrome is the default and the reason to use this Actor. A fourth client,
toolbar, is deliberately not offered: it answers HTTP 200 with an empty
list, which is the worst kind of option because nothing about it looks wrong.
Expansion: one seed, hundreds of keywords
A seed on its own returns about fifteen suggestions. That is rarely what anyone wants, so the seed can be fanned out:
| expansion | what it sends | queries |
|---|---|---|
| None | the seed | 1 |
| Alphabet | laptop a … laptop z | 27 |
| Question words | how laptop, what laptop, … | 13 |
| Prepositions | laptop for, laptop vs, … | 9 |
| Digits | laptop 0 … laptop 9 | 11 |
laptop with the alphabet expansion returned 80 distinct suggestions in nine
seconds. web scraping with question words returned 60.
Each expansion multiplies the number of requests, which is exactly why it is a
visible choice and not a hidden default. Every row records the query that
produced it alongside the seed, so a fanned-out result can always be traced
back to what asked for it.
Language and country are separate, and both matter
Suggestions are regional even at a fixed language. Language sets what the suggestions are written in; Country sets what Google thinks the market is and where the request comes from.
laptop in Dutch from the Netherlands returns laptop hoes, laptop kopen,
laptoptas dames. The same seed in English from the US returns nothing like it.
Both are on every row, so two runs are comparable rather than accidentally
different.
What you get per suggestion
suggestionandrelevanceseedandquery— what you asked for, and what was actually sentrankwithin that querylanguage,market,position
Duplicates are removed per seed, so an alphabet run does not hand you the same completion twenty times.
Run summary
Seeds, queries sent, suggestions returned, how many carried a score, and
queriesThatReturnedNothing.
That last one is worth watching. A query returning nothing is a normal thing:
why web scraping may simply have no completions. A run where most queries
return nothing means the client or the market is wrong, not that Google is out
of ideas.
What people use this for
Keyword research. The alphabet expansion on a head term is the classic long-tail sweep, and the scores let you sort the output by something better than alphabetical.
Content planning. Question-word expansion returns the questions people actually type, which is the raw material for an FAQ or a comparison page.
Market entry. The same seed across several language and country pairs shows what a market asks that yours does not.
Brand monitoring. What Google suggests after your brand name is what people search next, and it changes. Run it on a schedule and keep the rows.
PPC. Suggestions with a high relevance score against a head term are the phrases with real volume behind them, before you pay anyone for volume data.
Notes
Suggestions are personalised in a browser and not here: what you get is the unpersonalised set for that language and country, which is the right basis for research and the reason two people running this get the same answer.
They also change, sometimes daily. A run is a snapshot, which is why nothing is cached between runs.
This returns suggestions, not search volume. For volume, CPC and difficulty, that is a different question and a different source.
