Inputs & Audience Sizing

Every piece of intelligence Consumr.AI produces starts with inputs. Getting the inputs right determines whether a report will be rich, generic, or fail entirely.

The two fundamental input types

Consumr.AI works with exactly two types of inputs:

1. Audiences: behavioral and demographic cohorts sourced from ad platforms. This includes lookalike audiences built from client data, or platform-native audiences based on interests and behaviors. (The platform treats this as “first-party” in conversation even though Consumr.AI never handles client data directly; clients upload it to the platform.)

2. Keywords: terms that define what people are searching for, talking about, or intending to act on. Keywords power the Intent and Mentions report types. Audiences power the Behavior/Persona report.

Input helpers

Because most users don’t arrive knowing which exact interests or keywords to select, Consumr.AI provides three helpers:

Brief. Describe your target customer in plain language. The platform’s models translate that description into the specific interests that exist in Meta, Snapchat, TikTok, and other platforms. You don’t need to know the taxonomy; the brief does the mapping for you.

Key categories. When a specific audience is too narrow to reach the minimum population needed for behavioral intelligence, key categories let you step back to a broader cohort group. For example, “credit cards” as a category will always include enough people, whereas a very specific sub-segment might not. The category gives you a platform to work from when niche inputs would fail.

URL / landing page. Paste a landing page URL and the platform extracts relevant interests and keywords from the page content. This is useful when a client has an existing asset (campaign landing page, product page) and wants to align audience inputs with the message they’re already putting out.

The recommendation engine and Omni box

Once a segment definition is in place, the recommendation engine picks inputs automatically: audiences, keywords, and keywords for exclusion. The Omni box is the editable version of this, letting you refine the recommendations. For example, a segment defined as “digital-savvy, budget-conscious drivers” might auto-populate with inputs like “online car insurance” and “cheap car insurance,” then automatically exclude other insurance categories so the data doesn’t bleed into competitors.

Date range is another input parameter: it controls the recency of the comments and conversations the platform draws from. Fresher date ranges reflect current sentiment; older ranges can capture historical baselines.

Exclusions prevent the audience from drifting into adjacent categories. If you’re building a segment around one brand’s customers, excluding competitor brands keeps the signal clean.

Audience-size indicators

Every input combination is scored before a report runs. The platform shows one of four states:

Indicator	Color	Meaning
Insignificant	Red	Below usable threshold: the report will fail. Do not run.
Niche	Orange	Borderline; may work but there is a meaningful risk of failure.
Ideal	Green	The target state. Balances affinity (behavioral specificity) with reach (enough people).
Broad	Blue	Enough people, but so generic that results will be superficial, little better than asking an LLM.

The minimum population for behavioral intelligence is approximately 500,000 people. Below that, there isn’t enough behavioral signal to construct a meaningful report. The “ideal” range is the sweet spot where the audience is specific enough to be interesting and large enough to be statistically meaningful.

If a client restricts their audience too narrowly, say, to a single city like Memphis rather than a broader geographic range, they risk falling into “insignificant” territory and having the run fail. Cohorts represent behavioral types, not individuals in specific zip codes.