The dropdown under the search box is the shortest honest summary of what people actually search for. This service turns it into a table - one row per suggestion, with the rank it held, for as many seed queries as you care to submit.
The shape of this export is unusually predictable, and two of its properties decide how you should model it.
Every seed query comes back as exactly ten suggestions. That held on every block in our runs without exception, so you can size a job before you run it: rows equals seeds times ten. It also means position runs 1 to 10 and then restarts for the next query - it is the rank inside one query's dropdown, not a row number in the file. Submit three seeds and you get thirty rows numbered 1–10 three times over. Group by query before you sort by anything.
The second property is that the first suggestion is your own query, handed back to you. On every block we measured, suggestion 1 was the seed string itself. That is Google offering the literal search as the first option, and it is not a defect - but it does mean roughly a tenth of your export is data you already had. If you are counting suggestions, decide up front whether to keep the echo row or drop it, and do it consistently.
Everything after that is a genuine completion. Every suggestion in our data began with the seed string, so the dropdown extends what you typed rather than reinterpreting it. Search for a brand and you get the questions people attach to that brand; search for a topic and you get the shapes that topic takes in real queries.
There is no URL to build and no page to load. You submit words.
1000 queries per request.Ten rows come back per seed, so a hundred seeds is a thousand rows. Billing follows rows.
The whole export. Three columns is not a limitation - a suggestion is a string and a rank, and there is nothing else to carry.
query group.The interesting column is position, and it is easy to misread. It is a rank within one seed's dropdown, not a row index - a fifty-seed export contains the number 1 fifty times. Read it as ordering, not as identity: nothing in these three columns is a stable key, because a suggestion can move rank or disappear between runs. If you need a key, make one from query plus suggestion.
Six runs, all six populated, eight query-blocks, four distinct seeds. Small, but every block behaved the same way, and one finding is worth designing around.
All eight query-blocks in our data returned exactly ten suggestions - no short blocks, no long ones. That makes this one of the few exports whose size you can predict to the row before you spend anything: seeds × 10. It also means a block that comes back short is worth investigating rather than shrugging at.
On 8 of 8 blocks the top suggestion was the seed itself. Ten rows per query therefore means nine new strings and one echo. Nobody's analysis is wrong because of it, but a "suggestions per seed" metric that quietly counts the echo is inflated by about eleven percent, and that is the sort of error nobody finds later.
We ran the same seeds more than once. One seed returned a byte-identical top ten both times; another returned the same ten strings in a different order; a third shared eight of ten. So the list moves, but slowly - which is exactly the property that makes it worth sampling on a schedule. Store the run date with every pull, and treat rank changes as signal rather than noise only once you have several samples.
The reference documents a language parameter defaulting to en and a region parameter it recommends setting for a better search experience. Autocomplete is one of the most locale-sensitive surfaces Google has - what people half-type differs by country even within one language - so if you are researching a specific market, set both and record what you set. Nothing in the three output columns tells you afterwards which locale a row came from.
Where the dropdown is more useful as a spreadsheet than as a dropdown.
Autocomplete is drawn from what people actually type, which makes it the cheapest honest source of long-tail phrasing there is. Seed it with your category and every product noun you sell, run it, and you have a few thousand real query shapes - including the ones no keyword planner suggests because their volume is too low to report.
Seed your own brand and your competitors' and read what follows. The completions after a brand name are the questions and doubts people bring to it - pricing, alternatives, complaints, login problems. It is a blunt instrument and a very fast one, and the rank order tells you which of those associations is loudest.
One pull is a snapshot of how a subject is being asked about. The same seeds on a schedule, matched on query plus suggestion, show new phrasings arriving and old ones dropping out. Because the list moves slowly, a suggestion appearing in the top ten and staying there is a real shift rather than a blip.
No subscription, no minimum, no per-seat licence. Your first 500 rows are on us - after that it is pay-as-you-go.
For every new account, one time. No credit card. All scrapers unlocked. At ten rows per seed that is fifty seed queries explored before you spend anything, which is a real piece of keyword research rather than a toy.
Roughly $2 per 1,000 suggestions - about $0.02 per seed query, since each returns ten rows. This is the rare export whose cost you can calculate exactly before running it, so there is no need to guess: multiply your seed list by ten.
Volume pricing, SLAs, dedicated workers and tailored onboarding for teams tracking large seed lists across languages and regions on a repeating schedule. Tell us your numbers and we will quote.
LIVESCRAPER10 at checkout.They do not tell you who is answering it or how often it is asked. These two finish the picture.
Take the suggestions you just collected and run them as real searches - who ranks, with what title and link, for every phrase people actually type.
Autocomplete ranks a suggestion but never quantifies it. Trends puts relative interest over time behind the phrases worth committing to.
Short answer: yes. A suggestion is a public feature of a public search box, shown to everyone who starts typing.
Autocomplete is public. Anyone can see it without an account, on any device, simply by typing. Collecting publicly visible information for research is long-settled practice, and as long as the data is publicly available and the process does not disrupt the service, there are no federal laws against it.
The subject matter is about as harmless as data gets. A suggestion is an aggregate - a phrase common enough across many people's searching that Google offers it to everyone. It is not one person's search history, and nothing in these three columns identifies anybody. There is no author, no session, no user.
Google's terms restrict automated access, so this remains a question of terms. We touch nothing behind a login, run no third-party trackers on the data layer, and your exports self-delete after 30 days.
The questions we hear most. Something else? Talk to us - humans write the answers, not bots.
query, suggestion and position. One row per suggestion. That header was identical on all six of our runs, and there is no fourth column - a suggestion is a string and a rank.position is the rank inside one query's dropdown, so it runs 1 to 10 and then restarts for the next seed. A file covering fifty seeds contains the number 1 fifty times. Always group by query first.query as an array and supports batching up to 1000 queries in a single request, so a thousand-seed list is one job rather than a thousand.language parameter that defaults to en, and a region parameter it recommends setting for a better search experience. Autocomplete varies a lot by locale, so set both deliberately - and note the setting yourself, because none of the three output columns records which locale produced a row.Hand over a list of seed words and get the dropdown back as rows - the suggestion and the rank it held, ten per seed. Your first 500 rows are free.
Autocomplete is the list of completions Google offers under the search box as you type, and it is one of the most useful public signals on the web. It is drawn from what people actually search for, which makes it a direct read on demand phrasing rather than a modelled estimate of it - and unlike a keyword tool it surfaces the long tail that is too small to report a volume for. The catch is that it exists for a fraction of a second inside a dropdown. This service turns it into a table: one row per suggestion, three columns, exported as CSV, JSON or Excel, for as many seed queries as you want to submit.
The export has an unusually predictable shape, and that is its best feature. Every seed query returns exactly ten suggestions - that held on every block we measured, with no short blocks and no long ones - so the size of a job is simply your seed count times ten, known before you spend anything. The published reference accepts the query parameter as an array and batches up to a thousand queries per request, so a large seed list is a single job. Two optional parameters do the rest of the work: language, which defaults to English, and region, which the reference recommends setting for a better search experience. Autocomplete is strongly locale-dependent, so both are worth setting deliberately, and worth recording separately - none of the three output columns tells you afterwards which locale a row came from.
Two properties of the data itself are worth planning around. The first is that position is a rank within one query's dropdown rather than a row number: it runs 1 to 10 and then restarts for the next seed, so a fifty-seed file contains the number 1 fifty times, and grouping by query has to come before any sorting. The second is that the top suggestion is your own seed handed back - that happened on every block we measured. It is Google offering the literal string as the first option, and it means ten rows per query is really nine new phrasings plus one echo. Neither is a flaw, but both quietly distort a naive count, and a suggestions-per-seed figure that includes the echo is overstated by about a tenth.
One honest note about volatility, and it cuts both ways. We ran several seeds more than once. One returned a byte-identical top ten; another returned the same ten strings reordered; a third shared eight of ten with its earlier run. So the list is neither fixed nor random - it drifts. That is what makes it worth sampling repeatedly rather than once, and it is also why nothing in these three columns should be treated as a stable identifier: build a key from the query and the suggestion together, store the date of every run, and let a rank movement earn its interpretation across several samples rather than one. Start free: your first 500 rows cost nothing and need no credit card.