Google Trends Scraper: Compare 100 Keywords on One Scale

Google Trends compares only 5 keywords at a time. Each comparison is scaled so its own highest point is 100 and rounded to whole numbers. Two separate comparisons can't be put side by side, and small keywords next to big ones show up as "0" or "<1".

This Actor compares up to 100 keywords or topics on one shared 0–100 scale, keeps small keywords measurable, and tells you how precise each number is.

What you get

For every keyword:

• its rank, average interest and peak interest, all on one scale;
• the full timeline (one value per day, week or month, depending on the time range);
• a precision label (high, medium, low), so you know which numbers to trust;
• how it was measured (matchedAs: words, topic or topic ID), the topic used, and a note if a requested topic couldn't be found.

Words with several meanings

A plain word counts every search for it: "jaguar" includes both the car and the animal. To measure one meaning, add it in brackets: Jaguar (car make). The Actor then uses the Google topic of that type, and only if its name matches. If Google has no such topic, the plain word is measured and matchNote says so. It never guesses. You can also give a topic ID directly, with your own label if you like: Python language=/m/05z1_.

You also get a wide table (CSV) with dates in rows and keywords in columns, ready for Excel, Google Sheets or pandas. It's saved in the key-value store record TABLE.

Example: 12 keywords, 5,500x apart (real run, worldwide, last 12 months)

Rank	Keyword	Average interest	Precision
1	weather	77.566	high
2	python	4.811	high
3	bitcoin	4.192	high
4	rust	1.543	high
…	…	…	…
12	ocaml	0.014	medium

On Google Trends itself, "ocaml" next to "weather" is just "<1". Here it has a real value and its own shape over the year.

How it works

1. Size check. Every keyword is compared against an anchor to estimate how big it is. Keywords the anchor drowns out are measured again against a smaller anchor.
2. Ladder. Keywords are sorted by size and grouped with neighbours of similar size. Neighbouring groups share one "link" keyword, and a group ends early when sizes jump, so nobody gets rounded to 0.
3. Stitch. The groups are chained through their links, each keyword is taken from the comparison where it was measured best, and the result is rescaled to a maximum of 100.

Why not just repeat one anchor keyword in every batch?

That simple method breaks as soon as your keywords differ a lot in size: small keywords round to 0, and if the anchor is small, every batch breaks.

We tested both methods on simulated Google data where the true answer is known. That's 40 keywords per test, 6 tests per size range, with the same rounding Google does:

Keyword sizes span	Ladder: keywords lost	Ladder: worst error	One anchor: keywords lost
100x	0	under 1%	0 (worst error up to 17%)
1,000x	0	under 1%	57 of 240
10,000x	0	under 1% in 5 of 6 tests*	117 of 240

*In the hardest test the small keywords were off by about 12%, and the Actor labelled every one of them low. Keywords labelled high were within 1% in every test.

Tested on real Google data, too

We compared this Actor's output with Google's own direct comparisons of every neighbouring pair, for the 12 keywords in the example above (5,500x apart):

• no keyword lost;
• neighbouring keywords within 0.5% (median), 2.2% (worst) of Google's direct ratios;
• "weather" vs "ocaml" within 2% of the ratio chained from Google's pairs;
• the same input run twice differed by at most 1.1%.

Values keep 4 decimals. Tools that round to whole numbers after rescaling turn every keyword below 0.5 into 0. In the same 12-keyword test, that approach lost 6 of the 12 keywords.

The raw data is genuine. Google sends automated tools decoy data in some panels (related queries), but not in the timeline. We checked the timeline Google sends us against the Google Trends website, value by value: 53 of 53 weeks identical (python, United States, past 12 months, 7 Nov 2026). This Actor uses only that timeline, and we repeat the check every month.

Input

Field	Example	Notes
Keywords or topics	python, Jaguar (car make), Python language=/m/05z1_	Up to 100. Duplicates are removed. See "Words with several meanings" below.
Location	US, GB, US-CA	Empty means worldwide.
Time range	today 12-m	Same format as Google Trends: now 7-d, today 3-m, today 5-y, all, or 2024-01-01 2024-12-31.
Search type	Web, YouTube, News, Images, Shopping
Category ID	0	Google Trends category number.
Proxy	Residential (default)	Google blocks data-centre IPs quickly.

{
  "keywords": ["python", "javascript", "java", "rust", "kotlin", "typescript", "golang", "haskell"],
  "geo": "US",
  "timeframe": "today 12-m"
}

Output example (one dataset item, shortened)

{
  "rank": 2,
  "keyword": "python",
  "query": "python",
  "topic": null,
  "matchedAs": "words",
  "matchNote": null,
  "averageInterest": 4.811,
  "peakInterest": 7.545,
  "precision": "high",
  "geo": "worldwide",
  "timeframe": "today 12-m",
  "property": "web",
  "category": 0,
  "timeline": [
    { "date": "2025-11-02", "value": 4.996, "isPartial": false },
    { "date": "2025-11-09", "value": 5.098, "isPartial": false }
  ]
}

FAQ

How do I compare more than 5 keywords in Google Trends? Put them all in the Keywords field (up to 100). The Actor runs the batches of 5, links them, and returns one scale. You don't need to pick an anchor keyword yourself.

Is this a pytrends alternative? Yes. pytrends was archived in April 2025. This Actor returns the same interest-over-time data as JSON or CSV, handles Google's rate limits for you, and can be called from Python with the apify-client package.

Why do my numbers differ from what I see on Google Trends? Google scales every comparison so its own top point is 100. Here the top point of all your keywords is 100, so each keyword is shown relative to the biggest one in your list. Ratios between keywords are what matter, and those are kept.

Can I measure my brand against competitors? Yes. For brand tracking in percent, see our Share of Search Tracker, which uses the same engine and reports each brand's share of all brand searches per month, quarter or year.

Do I pay for failed runs? You're charged per keyword in the results. If a run fails (for example, because of an invalid location code), no results are written and no keyword charges apply.

Tips

• Check matchedAs and matchNote when you use brackets. To find a topic ID yourself, search the term on Google Trends, pick the topic from the dropdown, and copy the /m/... or /g/... part from the address bar.
• Keep the same location and time range when you compare runs. Values are only comparable within one run.
• Very short time ranges (now 1-H, now 4-H) change minute by minute. A run can fail if a new interval starts midway. Just run it again.

Limits

• Values are relative interest, not search volume. That's what Google Trends provides.
• Up to 100 keywords per run. Larger runs take longer, because Google is asked politely, with pauses.
• The last point of a timeline is often isPartial: true (the current week or day isn't over yet).