Correct vs Incorrect Usage
Consumr.AI is a specialist tool. Like any precision instrument, it produces poor results when misused, and clients will sometimes blame the tool rather than the technique.
Anti-Patterns & Corrections
Section titled “Anti-Patterns & Corrections”Segmentation
Section titled “Segmentation”Anti-pattern: Making 100–200 segments
Clients sometimes believe more segments equals more insight. It does not. At 100+ segments, audiences overlap heavily, and overlapping audiences produce identical-looking twins who give similar answers. You end up with more noise, not more signal.
Instead: Build 6–10 segments that collectively cover 80–90% of your market. Segment quality matters far more than segment count. If a client insists on many segments, show them two overlapping ones and demonstrate that the twins respond almost identically; that usually resolves the debate. (S2 95–97)
Anti-pattern: Defining a segment as a single individual or a single city
The twin represents a cohort: a meaningful population exhibiting shared behaviours. A cohort requires sufficient population size (approximately 500,000+) to produce reliable signals. Restricting a segment to one city (e.g., Memphis, TN) risks falling below the minimum threshold, triggering a “population insignificant” or failed segment result.
Instead: Define segments by behaviour, psychographic profile, and broad geography before narrowing to location if needed. A single individual is a persona exercise, not a Consumr.AI use case. (S2 71–83)
Research Design
Section titled “Research Design”Anti-pattern: Asking the twin to rank N concepts in order of preference
A human asked to rank five concepts in order of preference will give a different ranking on Tuesday than they gave on Monday, depending on how the question was framed and what they had for breakfast. The twin behaves the same way: each conversational instance is statistically independent. Asking for a ranked list produces an output that looks precise but is not reproducible.
Instead: Use concept testing or message prioritisation features, which are designed for comparative evaluation and produce statistically valid outputs. Do not use Converse for ranking exercises. (S2 594–598)
Anti-pattern: Logo / gestalt testing
Clients occasionally want to test whether an abstract or gestalt logo “reads” correctly to consumers. AI cannot do this. The platform processes visual data, lines, shapes, colour values, and cannot reconstruct the culturally learned meaning of a logo that depends on negative space or associative memory.
Instead: Do not pitch Consumr.AI for logo recognition testing. If a client raises this, explain the limitation directly (see Common Objections: logo objection). This is not a workaround situation; it is a genuinely unsupported use case. (S2 562–588)
Input Behaviour
Section titled “Input Behaviour”Anti-pattern: Pasting essay-length prompts into Converse
Converse is a conversation interface, not a prompt engineering box. Clients who come from an LLM background sometimes paste multi-paragraph strategic briefs expecting the twin to synthesise a research report. The platform applies a cap (approximately 250 words) for a reason: long inputs dilute focus and produce diluted outputs.
Instead: Keep inputs conversational. If a client has a complex strategic question, coach them to break it into a conversation: ask, get a reply, follow up. The dialogue format surfaces better insights than a monologue does. (S1 288–300)
Anti-pattern: Pasting sensitive customer data into chat
Clients may attempt to paste CRM extracts, PII, or customer-identifying information into the Converse interface to “give the twin more context.” This is a governance risk with no upside: the platform does not process that information in any useful way, and if something goes wrong, accountability is unclear.
Instead: Never enter PII, CRM data, or customer-identifying information into the chat interface. If a client asks about it, the answer is a clear no. Governance is the answer, and the platform is designed so that first-party data never touches Consumr.AI systems in the first place. (S1 312)
Survey & Brand Track
Section titled “Survey & Brand Track”Anti-pattern: Expecting to swap or customise brand-track questions
Clients sometimes want to modify the brand-track survey questions to match their internal frameworks or add proprietary metrics. The questions are standardised and locked by design: results are comparable over time and across brands precisely because the questions do not change. Poorly formed custom questions contaminate the dataset.
Instead: Explain that standardisation is the feature, not a limitation. If a client needs bespoke quantitative research, that is a different instrument. The brand track is for longitudinal, comparable measurement. (S2 467–475)
Anti-pattern: Running the large-sample reverse methodology casually
The 10,000-respondent reverse method is computationally intensive. Triggering it without a clear, justified reason consumes significant resources and is not appropriate for exploratory use.
Instead: Reserve this method for situations where you have a specific hypothesis that warrants high-confidence validation at scale. Confirm with your team that the research question justifies the compute cost before initiating. (S2 345–347)
Role Configuration
Section titled “Role Configuration”Anti-pattern: Asking questions in the brand-owner role when you want unbiased category insights
When you identify yourself as a brand representative within the platform, the twin contextualises all answers within that brand relationship. Every response is filtered through the lens of how the consumer relates to your brand, which is correct for brand research, but wrong for category-level intelligence.
Instead: Switch to the “category researcher” role. In that mode, the twin does not know who you represent. It gives unfiltered, category-level perspective: how consumers think about the space, not how they think about your brand specifically. Use the right role for the right question. (S1 148–156)