A quant attention job is a map plus a confirmation rule. The map turns a listed name into the strings each source accepts. The rule requires more than one source to move before a flag. POST https://api.trendsapi.ai/api. Factor jobs should fail if body is left as a string. Field names: the API reference. Search as a source: the Google Trends guide.

Map the name, then confirm

POST https://api.trendsapi.ai/api
Authorization: Bearer <api_key>
Content-Type: application/json

{"mode":"get_growth","source":"google search, wikipedia, youtube","keyword":"ChatGPT","percent_growth":["3M","12M"]}

Mapped-name check on 27 Aug 2026 for ChatGPT:

  • google search: 78 on 22 Aug 2026, 3M −4.88%, 12M −10.34%.
  • wikipedia: stored Chatgpt, 47.5 on 1 Jul 2026, 3M +38.08% (volume 6574), 12M −52.5% from 100.
  • youtube: 49 on 22 Aug 2026, 3M +44.12%, 12M +68.97%.

Search cooled. Video search heated. Encyclopedia traffic rose on the quarter and crashed on the year from a peak. A rule that flags "up" when any source is up will always fire. A rule that flags when all three share a sign will stay quiet. Write the rule down. Event-driven notes in 2026 still treat multi-source spikes as the confirmation step. This pull is the opposite case: the sources disagree, so the flag stays off.

Units

recent_value is 0-100. growth is a signed percent between two preset dates. volume is a count when the source has one. None of those is a return, a share, or a dollar. volume_weighted_growth in this pull was +38.11 and used only Wikipedia. Google and YouTube dropped out of the weight. Logging 38.11 as "ChatGPT attention" without weight_coverage 1/3 is a data error.

percent_growth presets are two-date comparisons (7D, 1M, 3M, 12M, YTD). They are not OHLCV bars. Difference get_time_series in the notebook when the model needs a daily factor. Wikipedia titles: the Wikipedia guide.