Documentation

Why these two numbers matter

Confidence and Idea are the headline signals on the report hero: they compress an entire simulated panel into two percentages you can track over time. They are among the first metrics leadership and investors look at—use them as the scoreboard for “did this session believe us?” and “did they like the idea?”—then drill into personas and tabs for the why.

Standalone runs (no GEO link): a snapshot of this panel only

These scores are not fixed constants for your product worldwide. Every run produces its own Confidence and Idea values because they reflect how this specific focus group reacted—this mix of personas, this brief, this day. If you run again with different people or prompts, expect the numbers to move; that is normal, not a broken model.

Product–market fit and buyer portraits

Compare sessions across runs: for which persona set, niche, or scenario were Confidence and Idea highest? That is a directional signal for product–market fit in that slice—not a guarantee, but a strong hypothesis. When the idea “lands” for one cluster of personas, study who they are on the page (role, industry, seniority, tone, risk appetite). Those profiles approximate who is most likely to buy, who will hesitate, and who will never engage—because the simulation is written to embody those traits. Low Idea with a given archetype is often as informative as high Idea: it tells you who you are not for yet.

GEO-linked runs: weak spots from GEO, plus AI- and consumer-shaped perception

When the session is started from a GEO analysis, the scenario is anchored in weaknesses the GEO pass already surfaced (visibility, positioning, trust, narrative gaps, etc.). The panel is still synthetic, but the question framing is aligned with “where the company is weak in AI-mediated discovery.”

In that mode, Confidence and Idea summarize how strongly this simulated panel—after seeing that GEO-shaped context—buys the story and the offer. That bridges two worlds: (1) how large language models tend to describe and recommend your category, and (2) how buyers like those personas react when the same weaknesses are in play. In practice, the same LLM systems that power GEO-style answers also influence what end users see in AI assistants; personas are a proxy for “the kind of end user” those models implicitly optimize for. So: if your personas score high, you are looking at a segment where both human-style skepticism in the session and AI-driven recommendation patterns are more likely to work in your favor; if they score low, the same profile is a warning for both sales and generative discovery.

What each score measures (technical)

Confidence reflects whether the simulation produced a stable, internally consistent discussion—personas stayed in character and the narrative did not collapse into noise. Idea score reflects how strong the evaluated concept or offer appeared to the panel in aggregate.

How they are built

Scores are model-generated from the full session context—they compress many qualitative turns into headline percentages for tracking over time. Individual persona ratings (where shown) drill into aspects such as trust or recommendation likelihood.

How to read them

• Low confidence with normal Idea: the idea may be fine but the brief, personas, or moderation were noisy—re-run before changing strategy.
• Low Idea with high confidence: the simulation understood the offer and still disliked it—prioritize offer, packaging, or proof.

History charts

When history exists for the same company or brand, compare runs after you change site, pricing page, narrative, or GEO inputs. Direction matters more than single-point absolutes. The highest historical pair points at the niche and messaging where you had the strongest PMF signal in simulation.