Search trends
Comparing up to five terms over time on Web Search, News, YouTube or Shopping, what the momentum verdict reads, and why a repeat inside seven days is free.
Search trends compares up to five keywords over time, on four Google properties, in the market you choose, and returns interest over time, a momentum verdict, and the queries and topics rising around each term.
A fresh analysis costs 2 credits. A repeat inside the seven-day cache window is free, which makes it the one thing in RankX AI that gets cheaper the more you use it.
What it answers, and what it does not
Trends is the right tool for "is this subject worth building around?" and the wrong tool for "what else should we target?" Keyword discovery is the reverse. If you are choosing between two subjects, run trends; if you are looking for subjects, run keyword research or AI prompt research first and bring the shortlist here.
Interest is relative, not absolute. The series is indexed 0 to 100 against its own peak, so a term with a flat line at 90 is not busier than one peaking at 100 in a different market. Compare shapes, not heights, and never read an index value as a volume.
The four properties
The property changes what "searched" means, and picking the wrong one is the commonest mistake here.
| Property | What it reads |
|---|---|
| Web Search | Ordinary Google searches. The default, and the right one unless you have a reason |
| News | Searches inside Google News. Reads coverage cycles rather than buying interest |
| YouTube | Searches on YouTube. Usually the earliest signal for anything people want shown rather than explained |
| Shopping | Searches inside Google Shopping. Product demand, and seasonal in ways Web Search is not |
Time ranges
Four ranges are offered directly, and an all-time range sits behind them:
- Past 7 days
- Past month
- Past 12 months
- Past 5 years
- All time, which reaches back to 2004 on Web Search and to 2008 on News, YouTube and Shopping. That difference is Google's, not ours
The underlying interface accepts more ranges than the four above, down to the past hour, and the MCP tool exposes the full set. The interface offers a curated four because the shorter ranges answer a newsroom's question rather than a marketing one.
Match the range to the decision. Past 5 years is the range that tells you whether a subject is structurally growing or dying; past 12 months tells you whether it is seasonal; past 7 days tells you almost nothing about a business subject and a great deal about an event.
The momentum verdict
RankX AI reduces the series to one of four words so that a shortlist can be scanned rather than read chart by chart.
| Verdict | How it is derived |
|---|---|
| Growing | The last three months average more than 15% above the three months before them |
| Declining | The same comparison, more than 15% below |
| Flat | Inside that band in either direction |
| Unknown | Fewer than six months of data exist. RankX AI says unknown rather than calling it flat |
The last row is the one worth knowing. A short history is not a stable subject, and reporting it as "flat" would put a term with no evidence beside a term with a year of it.
Related queries and related topics
For a single-keyword request, RankX AI also returns four lists:
- Rising queries and top queries: literal searches people make around your term
- Rising topics and top topics: broader entities Google associates with it, which are not the same as the queries and often more useful for planning a cluster
These come back only for a single keyword. That is a property of the source, not a limit RankX AI adds, and it is worth knowing before you plan your session: if the rising list is what you are after, run the terms one at a time rather than as a five-term comparison.
Comparing markets
The same term can be run per market and read side by side, which is how you find out whether a subject growing in one country has already peaked in another. A term with no data in a market returns nothing rather than a zero line, for the same reason the demand column in AI prompt research keeps "no estimate" apart from a measured zero.
How a run behaves
It is dispatched and polled. A fresh analysis returns a task id and completes in the background. Reading the result costs nothing, and a repeat inside the cache window returns immediately.
Five keywords is the ceiling per analysis. Beyond that, run a second analysis; there is no partial-comparison mode.
Saved lists and history. An analysis can be saved to a trend list and reopened later, and past runs stay in your history.
What it costs
2 credits for a fresh analysis. Free from the seven-day cache.
The free half is not a footnote. It is the only place in RankX AI where repeating the work costs less than doing it, and it changes how the feature should be used: checking a shortlist weekly is affordable in a way that checking it daily on a per-run price would not be. It is a property of how the run works rather than a rounding, because a cache hit returns the stored result before any credits are reserved.
Every rate is published in the credit cost reference.
Driving this from an assistant
get_keyword_trends dispatches an analysis and get_keyword_trends_status polls
it. The tool accepts the full range list rather than the curated four, so an
assistant can ask for windows the interface does not offer.
"Compare these three subjects on Web Search over the past 5 years for the UK, and tell me which is structurally growing rather than seasonal."
Report momentum as measured, and say plainly when interest is flat rather than implying a rise. See the tool reference.
Where to go next
- Keyword research, for finding the terms to compare in the first place.
- AI prompt research, for the same job over questions asked of assistants.
- Topic clusters, for the structure a growing subject should become.
Last updated
Keyword research
Discovering keywords in RankX AI, what each figure means, how the cache makes repeats free, and how discovery feeds rank tracking and topic clusters.
Topic clusters
The single structure that keywords, tracked prompts and content all hang from in RankX AI. What a good cluster looks like, and why it is worth getting roughly right.