Trend Intelligence sources, signals, and comparable history
Understand what a trend observation represents, how topics and duplicates are handled, and why AI attention is different from search demand or sales.
On this page
Start with the signal familyWhich sources can contributeTopics and the basis of a findingDeduplication across modulesShares and volumesHistory, freshness, and thresholdsA confidence label is not purchase probabilityFrom trend to merchant actionStart with the signal family
Trend Intelligence organises observations into topics and comparable histories. An AI recommendation signal measures topic presence in recorded answers. Search demand uses supported search observations. Merchant performance uses supported attributed commerce records. Market and competitor-supply evidence describe other observed conditions. These families can reinforce a hypothesis but must not be pooled into one undocumented volume.
Which sources can contribute
The reviewed signal implementation can consume stored AEO and Commerce answers, its benchmark observations, connected Search Console reports, directly identifiable imported commerce attribution, and supported competitor or shopping observations. A connector must be available, authorised, populated, and relevant. Seeing a source name in a settings screen does not establish that it supplied usable data for this topic.
| Signal | Observed unit | Limit |
|---|---|---|
| AI attention | Deduplicated answers containing an applicable topic | Not total consumer demand |
| Search demand | Reported impressions associated with applicable queries or pages | Limited to the connected property and available report |
| Merchant performance | Eligible directly attributed imported product quantities | Not necessarily total shop sales |
| Competitor or market evidence | Supported observations with their source and date | Not an exhaustive live market inventory |
Topics and the basis of a finding
A topic can be a category, attribute, preference, intent, or brand. Structured extraction and a textual alias match provide different evidence. The signal layer distinguishes measured, inferred, AI-interpreted, recommended, and forecast information. A category inferred from a product mapping should not be presented as if a shopper explicitly typed that category.
Deduplication across modules
The same provider answer can be stored by more than one module. The reviewed source layer recognises normalised answer-text identity so a duplicate is not independent corroboration merely because two dashboards contain it. Questions, providers, models, locale, and method identify comparable panels. Repeated answers to one question remain correlated evidence rather than new questions.
Shares and volumes
For a share signal, a topic contributes a hit within an eligible observation population. For a volume signal, quantities such as reported impressions are accumulated under the applicable source definition. A share can rise while absolute observations fall. Search volume and directly attributed units cannot be added to answer shares. Read the unit beside every chart.
History, freshness, and thresholds
The deployed trend-signal-scoring-v1 profile uses complete UTC ISO weeks, comparing four-week windows with a longer history where available. It requires sufficient samples and distinct clusters for share comparisons and sufficient covered weeks for volume comparisons. Its reviewed share minimums are twenty samples and five clusters per applicable window; volume comparisons require at least three covered weeks. A new topic can show an initial signal while a reliable trend remains unavailable. Future method versions may use different gates, so retain the displayed version.
A confidence label is not purchase probability
The reviewed confidence index combines significance, sample support, persistence, freshness, and comparability. It is a heuristic summary of evidence quality, not the probability that demand will grow or a product will sell. A spike concentrated in one week differs from a persistent rise. Forecasts, when available, have additional history and fit gates and should remain visibly identified as forecasts.
From trend to merchant action
Use a supported trend to ask a narrower question: is the topic relevant to the shop, do current recommendations feature products you carry, and is there an unresolved mapping or content issue? Verify the catalogue and source evidence before creating a task. An AI-attention trend alone does not establish a supplier order, a missing product, or a future revenue amount.
Still need help?
Tell us what you expected and what happened.