How ChatReady
Scores Work
If you're evaluating ChatReady, this page is the methodology behind every number we show you. The formulas, weights, models, and data sources below match our production code.
We publish this because, as far as we know, no other tool in the GEO/AEO category publishes its methodology at this level of detail. “Methodology you can verify” is not a marketing line for us — it is the reason this page exists. Where our method rests on judgment rather than measured outcomes, we say so plainly below.
The ChatReady Score
Your ChatReady Score is a 0–100 composite of two halves:
ChatReady Score = (Brand Visibility × 0.40) + (AI Readiness × 0.60)- Brand Visibility — are AI engines actually mentioning your brand today?
- AI Readiness — is your website's content structured so that AI engines can understand and cite it?
The 40/60 weighting is our judgment, not an empirical finding: AI Readiness is the factor you can directly control — you can fix your site's content this week — whereas Brand Visibility lags, because it depends on when the engines next refresh. We weight the controllable factor higher. We are working toward calibrating this weighting against real customer outcomes; we do not have that calibration today.
Brand Visibility
Brand Visibility is itself a 0–100 score built from three measured components:
Brand Visibility = (Mention Rate × 0.50) + (Rank Score × 0.30) + (Sentiment Score × 0.20)- Mention Rate — the share of (prompt × engine) pairs in which your brand was mentioned. Test 25 prompts across 6 engines (150 pairs), appear in 30, and your Mention Rate is 20.
- Rank Score — when your brand appears in a ranked answer, how high. Position maps to a fixed curve: 1st = 100, 2nd = 80, 3rd = 60, 4th = 40, 5th or lower = 20. A brand mentioned without an explicit ranking scores 50; a brand not mentioned contributes 0. Averaged across every response, not only the ones that mention you.
- Sentiment Score — when your brand is mentioned, the tone: positive = 100, neutral = 50, negative = 0, averaged across every response.
The component weights (50/30/20) and the rank-to-score curve are methodology decisions, not measured findings. We chose them so that being mentioned at all dominates — appearing on more engines matters more than placing 1st on one. The rank curve is uncalibrated against external click-through data; refining it is on our roadmap.
AI Readiness — the 7 dimensions
AI Readiness is the weighted average of seven page-level dimensions, each scored 0–100 by analyzing your site's content:
AI Readability
Can an AI model parse and summarize the page — clear structure, concise paragraphs?
Entity Clarity
Are people, products, and places named and given context?
Fact Density
Verifiable facts, statistics, dates — citable data points?
E-E-A-T Signals
Experience, Expertise, Authoritativeness, Trustworthiness — author credentials, sourcing?
Citation Worthiness
Original information an AI would cite as a source?
Schema Completeness
Structured-data signals (schema.org) an AI can extract?
Content Freshness
Dates, recency signals, update markers?
The weights sum to 100%. They are hand-set judgments — heavier on the dimensions we believe most affect whether an AI will cite a page — not weights fitted to customer outcome data. Calibrating them empirically is future work.
Sample sizes
Brand Visibility is computed from a finite sample, and we show you the size on every report:
- Each analysis tests N prompts across 6 AI engines — ChatGPT, Claude, Perplexity, Google Gemini, Google AI Overviews, Bing Copilot.
- N depends on your plan: Pro tests 15 prompts; Business, Agency, and Agency Plus test 25.
- A Business analysis is therefore 25 × 6 = 150 (prompt × engine) measurements.
Brand Visibility is a point-in-time snapshot. AI engines search the live web; their answers change as the web changes. Re-running your analysis next week may move your score — not because the measurement is noisy, but because the engines genuinely returned different results. Treat each score as a reading taken on a given day, not a fixed property of your brand.
Which models we use, and why
GEO Audit scoring — 7 dimensions, E-E-A-T, platform readiness
Claude Haiku 4.5 · temperature 0
Fast and low-cost for structured rubric scoring; temperature 0 makes scores as reproducible as the model allows.
Prompt generation
Claude Sonnet 4.6
Generating realistic, varied consumer prompts benefits from the stronger model.
Brand-mention detection
Deterministic regex — not an LLM
Whether your brand name appears in an answer is a factual string match; we do not ask a model to judge it.
temperature 0 on the scoring calls means the same page scored twice gets the same — or near-identical — result. See the variance note below.
The Prompt DNA corpus
The prompts we test are not generic “best X” queries — they are modeled on how real consumers actually talk to AI. The Prompt DNA corpus was built in April 2026 from approximately 9,249 patterns extracted from Reddit threads and Google Maps reviews across 15 industry verticals. Because AI search behavior shifts over time, a corpus like this can age. Our policy is to review the corpus every quarter — asking whether its patterns still reflect how people query AI — and to rebuild it from fresh data when that review concludes a refresh is warranted. There is no automatic refresh; a rebuild is a deliberate decision, not a calendar event. Whenever we rebuild, we update the build date and pattern count on this page.
What ChatReady does not claim
We think the honest part of a methodology is the limits. ChatReady does not:
- ✕We do not predict citation lift or traffic outcomes. We measure visibility and readiness as they are today; we do not promise a score change will produce more AI citations or more traffic.
- ✕We do not use empirically calibrated weights. The 40/60 composite, the 50/30/20 Brand Visibility split, and the 7 dimension weights are reasoned judgments, not weights fitted to observed customer outcomes. Calibration is roadmap work.
- ✕We do not use an LLM to decide whether your brand was mentioned. Mention detection is deterministic regex — a factual string match, not a model’s opinion. (Sentiment, separately, is model-classified.)
- ✕We do not treat Brand Visibility as a fixed score. It is a point-in-time snapshot; re-running it may shift the number because the web changed, not because the tool is inconsistent.
- ✕We do not claim run-to-run identical GEO Audit scores. Even at temperature 0, dimension scores can vary by roughly ±1–2 points on identical content — a small residual non-determinism from logprob ties in the model. We disclose this rather than hide it.
- ✕We do not track your brand continuously. Each analysis is a sample of 15 or 25 prompts, not an always-on monitor. Tools that poll continuously will have larger sample sizes than a single ChatReady run.
Versioning
Last updated May 20, 2026. This page is revised by hand whenever the methodology changes. Detailed methodology decisions — every formula change, weight choice, and the reasoning behind them — are tracked internally in our state-docs system; we are happy to share specifics on request.
Run a free analysis