How the AI Showdown Works
Current methodology version: v1.0 · AI Verdict scoring: 5-4-3-2-1 points (v5-4-3-2-1-v1)
The setup
Every AI system participating in a run receives the same base question and setup. The profile and evaluation criteria remain consistent across all systems within that run, so the comparison is apples-to-apples.
AI responses are research inputs
AI responses are research inputs, not automatically published recommendations. Credit Capital Guide independently checks important product information and current terms, then makes the final editorial judgment.
How consensus works
Consensus measures how many tested systems independently recommended the same product. Consensus does not automatically determine Credit Capital Guide's final pick — the editorial team reviews each recommendation against verified terms before making its own selection.
AI Verdict & AI Consensus Score
The AI Verdict is a transparent, automatically calculated ranking based entirely on the participating AI systems' rankings. It is completely separate from Credit Capital Guide's editorial picks.
Point system — Each AI system's ranking awards points: #1 = 5 points, #2 = 4 points, #3 = 3 points, #4 = 2 points, #5 = 1 point, and not recommended = 0 points. A card's point total is the sum of points it receives across all participating AI systems.
AI Consensus Score — The point total is normalized to a 0–100 scale:
AI Consensus Score = (points earned / (participating systems × 5)) × 100
The number of participating AI systems is used in the denominator so future runs with a different number of systems remain consistent. The unrounded normalized value is used for ranking; only the displayed score is rounded so display rounding can never change the ranking order.
Consensus vs. AI Score
Consensus and AI Score answer different questions:
- Consensus — how many AI systems included the card (frequency of recommendation).
- AI Score — how strongly the card performed based on both frequency of recommendation and ranking position.
A card recommended by all systems at #3 will have high consensus but a lower AI Score than a card recommended by fewer systems but always at #1. Both metrics are shown so readers can see the full picture.
AI Verdict tie-breakers
Cards are ranked by AI Consensus Score. When two cards have the same score, the following tie-breakers are applied in order:
- More AI systems recommending the card.
- More #1 rankings.
- Better average rank among systems that recommended it.
If two cards are still exactly tied after all three tie-breakers, the result is displayed as a tie rather than inventing an additional arbitrary tie-breaker.
Validation safeguards
AI results are parsed from each system's raw response and cross-checked against the source text. When a parsed rank doesn't match the source, or a card can't be found in the source text, a validation flag is raised (for example, "Parsed rank doesn't match source, verify manually").
If any participating AI result has an unresolved validation flag, the final AI Verdict cannot be published. The verdict is marked "Incomplete / Needs Review" in the admin until an editor resolves the flagged results. Flagged contributions are never silently omitted to produce a partial score, because doing so could unfairly change the ranking.
Frozen published results
When a Showdown is published and all validation flags are resolved, the AI Verdict is calculated and stored as a historical snapshot tied to that run's methodology version. Published results are never recalculated live. If the scoring formula changes in a future methodology version, new Showdowns use the new formula while previously published Showdowns continue displaying the scores calculated under their original methodology.
If an editor intentionally corrects verified source or rank data on an already published Showdown, the verdict is not silently recalculated. An explicit "Recalculate AI Verdict" action is required so the change is deliberate and auditable.
AI Verdict vs. CCG editorial picks
The AI Verdict answers only one question: "What did the participating AI systems collectively rank highest?" It does not use verified product terms, annual fees, rewards, welcome offers, prequalification availability, affiliate relationships, or editorial judgment.
Credit Capital Guide's editorial picks remain a completely separate judgment made after reviewing the AI results and verifying important product information.
The AI Verdict is calculated entirely from the participating AI systems' rankings. Higher placements earn more points, while recommendations appearing across more systems naturally accumulate more points. Credit Capital Guide's editorial picks are determined separately after reviewing the results and verifying important product information.
What Showdowns don't do
Showdowns do not provide approval predictions or individualized financial advice. They are educational comparisons of AI-sourced research, verified and curated by Credit Capital Guide editors.
Methodology versions
Each Showdown is tied to the methodology version in effect when it ran. Superseded versions stay published so earlier results remain explainable.
One stage, one ranked list per system. Every participating AI system received an identical prompt describing a single credit profile and goal and was asked to rank its five best card recommendations. Those rankings were compared exactly as given: consensus counted how many systems included a card, and the AI Verdict was calculated from the same ranked lists using the 5-4-3-2-1 point scale. Preserved as published for Showdown #1 — its stored results, scores and verdict are never recalculated under a later methodology.
Two stages per system. Stage 1 (primary prompt) is unchanged: every system receives the same profile, ranks its top five cards, and those rankings alone drive the AI Consensus Score, the AI Verdict, ranking points and the winner. Stage 2 (supplementary prompt) asks each system a follow-up scenario question — for example which card it would choose for a large purchase or for a balance transfer. Scenario answers are recorded as separate card references and reported as a Scenario Consensus ("4 of 6 systems picked this card for this scenario"). Scenario picks are excluded by design from the AI Consensus Score, the AI Verdict, ranking points and winner determination, and are labeled as supplementary wherever they appear publicly.
Transparency over time
Methodology versions and prompt versions are stored with each run, so future changes remain transparent. Exact prompts, profiles, systems tested, and run-specific details are preserved with each individual Showdown.
Credit Capital Guide uses AI to gather and compare research. Credit Capital Guide independently verifies the material and makes the final editorial judgment. No user-facing financial recommendation is automatically generated or published by AI.
V1 does not provide approval predictions, personalized financial advice, live lender matching, bureau integrations, or automatic product recommendations.