Glossary
A quick-lookup reference for every named term used across these docs. Definitions are intentionally short — follow the cross-links for the full story.
A/B testing — Running two creative variants simultaneously to measure which performs better, traditionally by splitting ad spend. Consumr.AI can predict which message or creative will win before you spend (see Test Creative & Concepts).
Accuracy (statistics) — How close results are to the true value, on average. Distinct from precision (see below). High accuracy, low precision = scattered results centered around the target. See Accuracy vs Precision.
ACS data — American Community Survey; US Census Bureau data used to match respondent distributions to real population demographics. Used during survey creation so results extrapolate correctly. See Extrapolation & Census.
Ad copy — The written text accompanying an advertisement. Consumr.AI’s creative testing tools evaluate which ad copy resonates with a given segment.
Affinity (audience sizing) — A measure of how tightly a chosen audience matches a specific interest or behavior. Ideal-sized audiences balance affinity with sufficient reach. See Inputs & Audience Sizing.
AI twin — An AI persona built from three reports (behavior, intent, mentions) that represents a consumer cohort — not an individual. Used for qualitative research, conversation, and as a base for surveys. See AI Twins.
Audience network — A collection of external websites and apps (publishers) that agree to serve ads from a platform like Meta, extending reach beyond the platform itself.
Auto-update — An AI twin setting that reruns all three underlying reports once per month, keeping the twin current as the real-world audience shifts.
Brand attachment — A mindset metric measuring how strongly a cohort identifies with or feels loyal to a brand. One of six mindset metrics in Consumr.AI. See AI Twins.
Brand funnel — The staged model of brand awareness: awareness → familiarity → consideration → preference → intent → endorsement. Measured in brand track surveys.
Brand track — A standardized survey type that measures share of voice, brand funnel stages, NPS, and brand perception metrics against competitors.
Brief — A natural-language description of a target audience, used by the platform to translate segment intent into audience inputs and keywords.
Category awareness — A mindset metric measuring how familiar a cohort is with the product category, not just a specific brand.
Category twin — A broad AI twin representing an entire product category rather than a specific brand’s audience. Used as a starting point for the “build using portfolio” segment creation method.
Central Limit Theorem (CLT) — The statistical principle that the distribution of sample means approaches a normal distribution as sample size grows, even if the underlying population is not normal. See Central Limit Theorem.
Cohort — A group of people defined by shared behaviors, attitudes, or demographics — the unit Consumr.AI studies, not individual people.
Confidence interval (CI) — The plus-or-minus range around a result that captures the true value at a stated confidence level (e.g., ±2% at 95% confidence). Related to margin of error. See Margin of Error & Confidence Interval.
Confidence level — The probability (typically 95%) that repeated sampling would produce a result within the confidence interval. Not the same as confidence interval.
Custom audience — An audience built from first-party data (e.g., store visitors, CRM records) uploaded to a platform like Meta, enabling the platform to find and target those specific people.
Date range — An input parameter on each twin that controls how recent the social-signal data must be (e.g., last 21 days). Keeping it current is what makes twins reflect the audience right now, not months ago.
Degrees of freedom — In the context of confidence intervals, the flexibility in a statistical estimate; a more practical interpretation here is the margin of acceptable error around a stated confidence level.
Design effect (DEFF) — A multiplier that accounts for the clustering effect in a sample. DEFF of 1.2 is the Consumr.AI target; if it rises to 2, the required sample size grows by ~80%. See Design Effect.
DMAIC — Define, Measure, Analyze, Improve, Control; the Six Sigma problem-solving cycle. Referenced in Consumr.AI’s brainstorming research type. See Six Sigma / DMAIC.
Emotional affinity — A mindset metric measuring how emotionally connected a cohort feels toward a brand or category.
Exclusions — Audience inputs you remove so the recommendation engine doesn’t pull in competing signals (e.g., excluding other insurance brands when building a GEICO twin).
Extrapolation — Scaling results from a Meta-sourced audience (60–70% of US population) up to match census-defined population totals. See Extrapolation & Census.
Familiarity — A mindset metric measuring how well a cohort knows a brand through direct experience or repeated exposure.
First-party data — Data collected directly by a business about its own customers (e.g., loyalty card purchases, store visit records). Consumr.AI does not ingest first-party data directly; clients upload it to Meta, and Consumr.AI builds on Meta’s signals.
Funnel (ad) — The journey from broad awareness to conversion; also a pattern used to classify website visitors by how seriously they are engaging (see window shopper vs serious buyer).
Gestalt psychology — The principle that the mind perceives wholes rather than parts; referenced in the context of transparent/negative-space logos that AI cannot parse correctly when the background is stripped.
Inputs — The audience signals, keywords, date range, and exclusions fed into the platform to generate reports for a twin. Generated automatically from the brief by the recommendation engine, then editable via the Omni box.
Insignificant (audience size) — An audience too small (below ~500,000 people) to produce statistically reliable reports; shown in red. The run will fail.
Lookalike audience — An audience that a platform (e.g., Meta) builds by finding people who behave similarly to a given seed audience (custom audience). Used as a seeding method for Consumr.AI inputs.
Margin of error (MOE) — The plus-or-minus range on a survey result at a given confidence level. Also referred to as confidence interval in the platform output. See Margin of Error & Confidence Interval.
Mindset metric — One of six psychographic scores assigned to each twin during creation: brand attachment, emotional affinity, purchase intent, category awareness, familiarity, skepticism. Replaced “stages of conversion” as the primary characterization model.
Normal distribution — The bell-curve shape that many aggregated measurements approach. Mean = median = mode in a true normal distribution, which doesn’t exist in real-world data but is a useful model. See Normal Distribution.
NPS (Net Promoter Score) — A survey metric asking how likely respondents are to recommend a product, scored 1–10. 9–10 = promoters/loyalists; 7–8 = passives; 1–6 = detractors. See Brand Track.
Omni box — The input panel within the twin creation flow where you can view and edit the keywords and audience signals the recommendation engine has suggested.
Persona — The behavioral profile generated from the behavior report; the “face” of a twin (name, age, demographics). Represents the cohort’s median, not an individual.
Pillars — The strategic topic or product-category dimensions a research program is organized around. Consumr.AI discussions reference a “6 pillars × 10 segments = 60 research combinations” pattern.
Pixel — A tracking code (e.g., Meta pixel) placed on a website that sends behavioral signals back to the ad platform, enabling custom audiences and funnel tracking.
Portfolio — The pre-loaded brand context (brand name, category, competitors, description) that anchors every research session. Tells the platform who is asking questions and in what context. See Portfolio.
Precision (statistics) — How consistently results cluster together, measured by variance. Low variance = high precision. Independent of accuracy. See Accuracy vs Precision.
Product tags — Category or attribute labels assigned to products by ad platforms (e.g., “action game”) that drive subsequent ad targeting.
Publishers — Third-party website and app owners who allow an ad network’s ads to be served on their properties.
Purchase intent — A mindset metric measuring how likely a cohort is to buy from a brand or category in the near term.
Qual / qualitative — Research methods that produce subjective, narrative answers about motivations, feelings, and reasons (the “why”). Done via AI twins in focus groups, interviews, and brainstorming sessions.
Quant / quantitative — Research methods that produce numerical results about magnitudes and frequencies (the “how many” / “how much”). Done via respondent surveys.
Reach (audience sizing) — The total number of people in a given audience. Broad audiences have high reach but low affinity; ideal audiences balance both.
Recommendation engine — The platform component that takes a segment brief and automatically suggests audience signals, keywords, date ranges, and exclusions as inputs.
Report — One of three data products the platform generates per segment: behavior report (persona), intent report (purchase signals, long-term memory), mentions report (social commentary, short-term memory). A twin is built from all three. See Reports.
Respondent — A light or mini twin used in quantitative surveys; a simplified AI persona created at scale (typically ~10,000) to answer survey questions and produce statistical distributions.
Retargeting — Serving ads to people who have previously visited your site or app; made possible by pixel tracking.
RFM — Recency, Frequency, Monetary value; a segmentation framework for classifying customers by how recently they bought, how often, and how much. Used as a seeding method for Consumr.AI inputs. See RFM.
Round Robin — A structured brainstorming method where each participant contributes ideas in turns; available as a research type in Consumr.AI’s innovation research module.
Sample / sample size — A subset of the total population selected to represent it. For US-wide research, ~10,000 respondents is typically adequate; the right number depends on the universe size and desired margin of error. See Sampling & Sample Size.
Scatterplot — A chart plotting two variables; requires a minimum of ~30 data points to show meaningful patterns. See Extrapolation & Census.
Second-party data — Data shared between two organizations (e.g., a partner’s first-party data). Less common than first or third-party.
Segment definition — The written description of a cohort’s demographics, behaviors, attitudes, and context; the input that the platform translates into audience signals. See Segments.
Segment twin — An AI twin built from a specific, narrowly defined segment (as opposed to a category twin).
Share of voice — The proportion of total consumer conversation or awareness in a category that belongs to each brand.
Six Thinking Hats — Edward de Bono’s structured thinking method using six cognitive modes (data, emotion, caution, optimism, creativity, process); available as a brainstorming research type in Consumr.AI.
Skepticism — A mindset metric measuring a cohort’s doubt or wariness about a brand or category claim.
Skewness — Asymmetry in a data distribution (e.g., most survey responses cluster at the high end). The GEICO worked example uses skewness as a signal to investigate further. See GEICO Skewness Loop.
Third-party data — Data collected by an entity with no direct relationship to the consumer (e.g., data brokers, ad platforms aggregating cross-site behavior).
Universe size — The total population a survey result is intended to represent after extrapolation. Shown alongside margin of error in platform output.
Variance — The spread of data points around the mean; the unit of measure for precision. Lower variance = higher precision.
Walled gardens — Closed advertising ecosystems (Meta, Google, TikTok, Snapchat) that accumulate audience data but do not share it externally. Consumr.AI works within these constraints rather than trying to bypass them. See Data Foundations.
Window shopper vs serious buyer — A behavioral classification made from pixel signals: someone who visits a site without triggering a conversion goal (window shopper) vs someone who builds a car, fills a form, and books a test drive (serious buyer).