tokconnectUse a returned sequence to choose what to investigate next. A larger last value can be useful, but its meaning depends on the starting value, the shape of the series, the topic wording, and the relevance of the proposed video. Your finished output should name the candidate, show the calculation, state the missing fields, and prescribe one small validation.
Define the trend question
“What is trending?” produces a list with no decision rule. A café team can ask, “Which coffee topic should we validate for one answer-led video this month?” Record the audience, market or language filter, period requested, seed, and the minimum evidence for an angle. The question makes clear that a trend signal must compete with relevance. A café can serve an in-store ritual; a brewing-equipment seller needs a question about making coffee.
Collect seeds from customer questions, product vocabulary, and existing content research. Record where each phrase came from. Search them under the same filters and compare like with like. TikTok’s Creator Search Insights offers manual topic exploration. Save screenshots, visible period, and filter selection if you add an app observation to a tool record.
Work the coffee record
The September 18, 2026 coffee retrieval used keyword_research with keyword coffee, language en, expansion 3, and videos false. Its selected record reads coffee shops and solo coffee, with query ID 7668817526511370260, search-volume estimate 4,806,180, listed video count 789, and seven returned values: 10,000; 110,511; 301,330; 456,845; 1,123,812; 1,655,865; 4,806,180.
The evidence labels these as values in trend7d. It does not supply a date attached to each point, a time unit for the estimate, or a definition of an individual searcher. Keep those unknowns in the result rather than converting the number into monthly volume or global demand.
Calculate with the baseline
The first-to-last change is (4,806,180 − 10,000) ÷ 10,000 × 100 = 47,961.8%. The final point is about 2.9 times the preceding 1,655,865 value. Those calculations describe this stored sequence. The percentage looks extreme because it begins at 10,000; it does not describe a sustained rate of growth, a forecast, or likely reach.
Put three candidates in one sheet with the same columns: first returned point, last returned point, absolute difference, percentage difference, whether dates exist, whether the topic fits, and a manual evidence link. Sort by relevance first, then inspect the scale and shape. A small relevant topic with a clear answer can outrank a large unrelated one.
| Interpretation | Evidence needed | Decision |
|---|---|---|
| Sharp late rise in returned order | Repeat retrieval and inspect topic context | Research candidate |
| Large estimate, weak subject fit | Product question does not match | Reject |
| Series lacks dates or values | Record the missing field | Do not chart as a dated trend |
Handle the missing follow-up
The saved follow-up requested a seven-day global search_popularity series for the coffee query ID. It returned one global item with points: null and latest: 0. There is no chartable series in that response. Zero may mean a returned numeric zero in another schema; alongside null points it cannot establish a collapse in interest. Preserve the response and write “series unavailable” in the analysis.
Check the topic manually in Creator Search Insights if you need current context. Record the date, filters, visible wording, and screen capture. Do not combine manual chart values with the stored retrieval unless the topic, scope, and periods match. Run a second retrieval later under the same input to see whether the late rise repeats.
Make a small relevance test
A café can test a source-supported “solo coffee-shop ritual” angle after it picks a real drink and a setting it can show. Search the phrase, collect a fixed sample of related videos, and code captions for questions about ordering, seating, or spending time alone. Those questions are targets for the next sample, not findings from the topic record.
A seller of brewing equipment should treat the same topic as a possible rejection: a solo outing topic may offer no useful way to demonstrate its product. The team should search a making-coffee seed and repeat the protocol. The outcome is a reasoned choice, not a demand headline.
For [seed] and [language], return the selected topic, query ID, estimate, listed video count, each returned trend value, and whether individual dates are present. Calculate first-to-last change and flag unavailable follow-up series. Rank only candidates that fit [reader problem]. For the top candidate, collect five public video links and identify questions for a source-backed test.
Candidate: coffee shops and solo coffee
Evidence: workflow-coffee-trends.json, retrieved 2026-09-18
Calculation: 10,000 to 4,806,180 = 47,961.8%
Limit: returned sequence has no per-point dates; popularity follow-up has null points
Decision: café may validate a solo-ritual angle
Next validation: repeat the retrieval, inspect five linked videos, and verify the offer
Status: research candidate