# Changelog of Google Trends Scraper (`fetchwright/google-trends-scraper`) Actor

- **URL**: https://apify.com/fetchwright/google-trends-scraper/changelog.md
- **Full Actor documentation**: https://apify.com/fetchwright/google-trends-scraper.md

## Changelog

### 2026-11-16

- README: a short note and Apify's video on running it on a schedule.
- README: related actors updated (same-topic links first, current names).

### 2026-11-15

- README links to a Python comparison example on GitHub.

- Default memory is now 128 MB (was 256 MB): a 10-keyword run peaked at 62 MB, so runs use about half the platform resources. You can still choose more memory.&#x20;

- Text block responses such as "Access denied" now stop the run without retrying, alongside verification pages. The refusal record keeps facts about the answer; unfinished keywords remain uncharged.&#x20;

- Long runs now use varied pauses and renew their sessions at a request limit. A block ends the run.&#x20;

- The price comparison names apify/google-trends-scraper, data_xplorer/google-trends-fast-scraper and agenscrape/google-trends-scraper, with Free and Gold prices from the 2026-11-13 Store snapshot; it is scoped to those three listings.

### 0.1.5 - 2026-11-13

- README and Store description: each price names its plan ($0.35 per 1,000 keywords on Free, $0.32 Starter, $0.28 Scale, $0.25 Business and above) and the $0.0005 per-run start fee; the links to other Fetchwright Actors use their current names and list all six. No change to the Actor, its input or its output.

- Start fee per run raised from $0.00005 to $0.0005, so a run of a few keywords covers its own platform cost; the price per keyword is unchanged.

- A web page that Google sends with a request to slow down (HTTP 429) now stops the run like a refusal, whether or not it is a known verification page; only a plain request to slow down is paused out.

- A web page sent with a server error (for example an "access denied" page with HTTP 503) and a web page too large to read now stop the run like a refusal; neither is retried or charged. Recognising a web page now takes the same time whatever the answer contains.

- "Compare with the previous run" assumes one run of a watch at a time; the README and the field help now say so.

- When a run stops because Google refused it, the run summary (`refusalPage`) and a `REFUSAL_PAGE` record say what kind of answer Google sent: status, content type, length, which rule stopped the run, and fixed yes/no facts (for example whether it was Google's standard error page or mentioned unusual traffic) with the kind of page title. No text from the page is stored.

### 2026-11-11 (listing text)

- The listing texts (title, description, README, field help and images) were brought in line with the current behaviour.

### 0.1.4 - 2026-11-11

- A verification page is recognised whatever its size, so it always ends the run. The README says when the per-region share of compared keywords is not available (more than five keywords on one scale).

### 0.1.3 - 2026-11-11

- If the platform restarts a run, the run sends nothing more to Google: it reports what it saved before the restart (with each result's charge as confirmed, or as uncertain), lists the keywords to run again, and ends. Nothing is charged twice.
- If the run's input cannot be read, the run fails at once and says so.

### 0.1.2 - 2026-11-11

- Rows no longer carry a `charged` field: a row is saved before its charge is confirmed. The `SUMMARY` record lists each keyword as charged, not charged or uncertain, from the charge the platform confirmed.
- Related queries are always required when you ask for related searches; related topics only for single keywords, as Google does not show them for compared keywords. A related-queries answer without both lists makes the keyword `partial`.
- If the final `SUMMARY` record cannot be saved, the run ends as failed and says so.
- Accuracy figures measured with an earlier version were removed; they are not yet measured for this version.

### 0.1.1 - 2026-11-11

- When Google asks for verification, the run stops at once: the keyword Google answered that way fails and is not charged, nothing more is sent to Google in that run (also after a platform restart of the run), the keywords not completed are listed and not charged, and the status message says that Google asked for verification and the run stopped. The same holds with the proxy on or off.
- When Google asks the run to slow down, the run waits as asked a limited number of times; if Google keeps limiting the run, or refuses it, the run stops the same way.
- A keyword is charged only when everything you asked for was read and checked. A keyword with a part that could not be read is saved free with the new status `partial` and a `missingParts` list; an answer from Google of an unexpected form fails the keyword instead of being filled in.
- The average of a range with no complete period is now empty instead of the value of the incomplete one.
- Comparing more than 5 keywords on one scale without interest over time is refused at the start with an explanation, because the scale could not be kept.
- The `SUMMARY` record is saved during the run as well as at the end, and names any result whose save or charge could not be confirmed (`chargeUncertain`) instead of calling it not charged. A run in which no keyword was read completely is no longer reported as successful.
- A related-searches answer with a list missing makes the keyword `partial` instead of showing an empty list; a web page where data was expected stops the run like a refusal.
- Speed, reliability and repeatability figures measured with the previous version were removed from this page; they are not yet measured for this version.

### 0.1.0 - 2026-11-10

- First version. One row per keyword with interest over time, average, peak, latest value and change over the range, interest by region, and top and rising related queries with Breakout flags.
- Compare up to 5 keywords on Google's own shared scale, with each keyword's share per region, and more than 5 on one scale through an anchor keyword, with scale factors and a precision flag on every point.
- Any location (country codes, common names such as UK, region codes such as US-CA), preset or exact date ranges, categories, and web, YouTube, News, Image or Shopping search.
- Compare with the previous run: for scheduled tracking, each row says what changed since the last run of the same watch (latest value, average, new and dropped related queries, new Breakout queries, regions in and out of the top 10). No extra charge.
- Keywords from another run's dataset.
- Keywords without enough search interest and keywords that could not be read are saved free. Clean stop at your maximum charge and before the run timeout, with unfinished keywords listed in the SUMMARY record.
- (Superseded in 0.1.1 and later: the Actor no longer continues on another route; see the newer entries.) When Google limits part of the run, the Actor continued on another route; keywords that still cannot be read, including those Google answers with a verification page, are saved free, and a run that meets verification too often stops early and lists the rest as unfinished.
