Google Autocomplete by country: how much suggestions differ, and how stable they are

Autocomplete is a map of what people type, and what people type depends on where they are. If you do keyword research for more than one market, the suggestions for the same seed will not be the same, and if you collect them more than once, they will not be the same either.

We measured both. We ran 10 seeds in five English-language countries, and ran the US ones again to find the noise floor. This guide has the numbers, the code to measure your own seeds, and what to do about the instability. We used the Google Autocomplete API on 16 October 2026. Suggestions change, so yours will differ.

How we ran it

We used 10 seeds: best phone, how to apply for, tax, football, cheap flights, bank account, school holidays, minimum wage, best credit card and student loan. Each was requested in us, gb, au, ca and in, which is 50 requests, and every request returned 15 suggestions. We then ran the 10 US seeds a second time to see how much two identical requests differ.

It is a small sample of seeds we chose, so the figures describe it and nothing wider.

How different the countries are

For each country we counted how many of the US suggestions it also returned:

A UK searcher saw about three of the 15 suggestions an American saw, and an Indian searcher about one. Some seeds shared more: best phone shared 8 with the UK and 9 with Australia and Canada. Others shared nothing, such as minimum wage with the UK, Australia and India.

The seeds show why. For minimum wage, the first suggestions were:

us  minimum wage in georgia, minimum wage atlanta, minimum wage in georgia 2026, minimum wage in california, minimum wage in new york
gb  minimum wage leeds, minimum wage bradford, minimum wage 2026, minimum wage uk, minimum wage 2027
au  minimum wage australia, minimum wage melbourne, minimum wage vic, minimum wage victoria, minimum wage australia 2026
ca  minimum wage alberta, minimum wage calgary, minimum wage edmonton, minimum wage alberta 2026, minimum wage canada
in  minimum wages, minimum wages in telangana, minimum wage in india, minimum wages act, minimum wages hyderabad

And for football, the US wanted games, the UK wanted scores and fixtures, Australia wanted Football Queensland, boots and Football Australia, and India wanted match, jersey and live. The seed tax showed how a short seed can mean different things: in the UK it led with tax my car and tax free childcare, in Australia with tax return and tax calculator, and in India the first two suggestions were taxim o and taxim o 200 tablet, which are a medicine.

How much of that is noise?

Different countries gave different suggestions, but so did the same country asked twice. To separate the two, we ran the 10 US seeds again and compared with the first US run:

Two requests for the same US seed shared 10 of 15 on average, from 4 to 15. Compare that with 1.2 to 4.3 between countries: the country differences are far larger than the run-to-run noise, so they are real. But the noise is not small, and some seeds are very unstable.

Local intent is the unstable part

Seeds that invite a place, such as things to do, pizza near or cheap flights, change the most, because Google adds towns to the suggestions. We requested things to do three times with country set to us and nothing else:

run 1  things to do amsterdam ny, things to do albany ny, things to do near me, things to do in lake george
run 2  things to do wichita ks, things to do salina ks, things to do in wichita, things to do near me
run 3  things to do wichita ks, things to do salina ks, things to do in wichita, things to do near me

Only 2 of the 15 suggestions appeared in all three runs. pizza near named Miami, Wichita, and New York across three runs, and even tax produced taxi luray va, taxi fort worth and taxi houston in different runs.

The location parameter did not fix it

The location parameter works on the search and local endpoints, as covered in Google results by country. The autocomplete endpoint accepts it too, but in our test it did not steer the suggestions to the city. With location set to Austin,Texas,United States, the things to do suggestions named:

Edinburg TX and McAllen          (first run)
Zelienople PA and Pittsburgh     (repeat run 1)
Kansas City                      (repeat run 2)
St Marys OH and Lima OH          (repeat run 3)

One of the four was in Texas. With Brooklyn,New York,United States, things to do named Tupelo and Columbus, Mississippi, and pizza near named Phoenix and Ahwatukee. The parameter changed the suggestions, but not in a way that tracked the city we asked for. Do not use autocomplete for city-level work. For city-level results, use the search endpoint with location.

The code

Put your key in an environment variable:

export CRAWLERDATA_API_KEY="YOUR_API_KEY"

import os
import time
from collections import Counter

import requests

API_URL = "https://api.crawlerdata.com/v1/jobs"
HEADERS = {"Authorization": f"Bearer {os.environ['CRAWLERDATA_API_KEY']}"}


def suggest(query, country="us", retries=4):
    body = {"endpoint": "google-autocomplete", "params": {"query": query, "country": country}, "queue": "live"}
    for attempt in range(retries):
        resp = requests.post(API_URL, json=body, headers=HEADERS, timeout=120)
        if resp.status_code == 429:
            time.sleep(int(resp.headers.get("retry-after", 10)))
            continue
        resp.raise_for_status()
        job = resp.json()["job"]
        if job["status"] == "completed":
            return [s["value"] for s in resp.json()["result"]["suggestions"]]
        if "retry" in (job["error"] or "").lower():  # "service is busy" and "temporarily blocked" both say retry
            time.sleep(5 * (attempt + 1))
            continue
        raise RuntimeError(job["error"])
    raise RuntimeError("Autocomplete failed after retries")


def stable_suggestions(query, country="us", runs=3, min_runs=2):
    """Suggestions that appear in at least `min_runs` of `runs` requests. Costs `runs` credits."""
    seen = Counter()
    for _ in range(runs):
        seen.update(set(suggest(query, country)))
    return [s for s, n in seen.most_common() if n >= min_runs]


def compare_countries(seed, countries=("us", "gb", "au", "ca", "in"), baseline="us"):
    """How many of the baseline country's suggestions each other country shares."""
    results = {c: suggest(seed, c) for c in countries}
    base = set(results[baseline])
    return {c: {"shared_with_baseline": len(base & set(r)), "first": r[0]} for c, r in results.items()}

For tax, stable_suggestions kept 10 suggestions that appeared in at least two of three runs, led by taxslayer, taxact, tax refund and tax brackets 2026. And for minimum wage, compare_countries printed:

us {'shared_with_baseline': 15, 'first': 'minimum wage california'}
gb {'shared_with_baseline': 0, 'first': 'minimum wage london'}
au {'shared_with_baseline': 0, 'first': 'minimum wage australia'}
ca {'shared_with_baseline': 2, 'first': 'minimum wage ontario'}
in {'shared_with_baseline': 0, 'first': 'minimum wages in delhi'}

Note the US first suggestion, minimum wage california, differs from the one in our first run (minimum wage in georgia). That is the noise this guide measures.

What to do about it

• Build separate lists per market. One list for all countries hides most of what each audience types.
• Request every important seed more than once. Keep suggestions that show up in at least two of three runs, as stable_suggestions does. It costs 3 credits per seed.
• Be careful with local-intent seeds. Their suggestions change by run and name random towns. Avoid them, or aggregate many runs.
• Check a seed before you trust it. Short seeds can mean several things in different countries, like tax in India. See Google Autocomplete keyword research for the sweep and the relevance filter.
• Use the volume tool and the results for the verdict. Suggestions are candidates, not proof of demand.

What it costs

Every request is 1 credit, on every queue. Starter is $1.00 per 1,000 requests and the largest pack is $0.30 per 1,000, see pricing.

Failed requests are not charged. The whole set of tests behind this guide used about 90 credits.

Next steps

You now have numbers on how far Google Autocomplete moves between five markets and between two identical requests, plus a function that keeps only the suggestions that survive repeated runs. The Google Autocomplete API page lists every parameter with a full example response, and Google results by country does the same comparison for search results.