f76cf2858b
Foundational public methodology for the first open public ranking of brand visibility in AI search results (ChatGPT, Perplexity, Gemini, Claude). This release establishes the framework — no rankings have been computed or published yet. First scan cycle: late May 2026 (private validation). First public ranking publication target: August 2026, after 3 validation cycles. Includes: - methodology.json: machine-readable formulas, weights, policies - README.md: human-readable overview + open/closed boundary - CHANGELOG.md: versioning policy + v1.0.0 release notes - taxonomy.md: tier system + 11 PL pilot categories - LICENSE: MIT - .gitignore: closed operational data (exact prompts, anti-gaming thresholds) - prompts/README.md: 6-stage prompt curation process - prompts/example-swiece-sojowe-pl.md: illustrative framework for first category Strategic principles: - Algorithm-first, no advisory board - Open methodology + closed exact prompts (Goodhart's Law defense) - No retroactive changes (FIDE 2024 lesson) - No pay-to-play, hard rule (Moody's / Forbes 30 Under 30 lessons) - Subjective opinion disclaimer (Gartner v. NetScout 2020 First Amendment shield) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
61 lines
4.6 KiB
Markdown
61 lines
4.6 KiB
Markdown
# Changelog
|
||
|
||
All notable changes to Citee Index Methodology are documented in this file.
|
||
|
||
The format is based on [Keep a Changelog](https://keepachangelog.com/), versioning follows [Semantic Versioning](https://semver.org/) adapted for methodology:
|
||
|
||
- **MAJOR** (`2.0.0`) — fundamental scoring formula change, weight rebalance, definition of categories
|
||
- **MINOR** (`1.1.0`) — new prompt types, new cross-signals, new model added, anti-gaming rule additions
|
||
- **PATCH** (`1.0.1`) — documentation fixes, clarifications, additional examples, typos
|
||
|
||
**Important:** No retroactive changes. Methodology updates apply to FUTURE cycles only. Cycles published before a version bump are not recomputed.
|
||
|
||
---
|
||
|
||
## [1.0.0] — 2026-05-03
|
||
|
||
Initial public release. Foundational methodology. **No public ranking yet** — first publication scheduled August 2026 after 3-month validation period.
|
||
|
||
### Added
|
||
|
||
- **Scoring formula:** `CiteeScore = sum(mention_score_per_model * model_weight) * (1 + cross_signal_bonus)`, normalized to 0-100 per category
|
||
- **Model weighting** for PL market: ChatGPT 0.45, Perplexity 0.25, Gemini 0.20, Claude 0.10 (Claude added Q4 2026 in pilot, see `methodology.json` for rationale)
|
||
- **Mention score per model:** position (0.4) + prominence (0.3) + sentiment (0.15) + citation depth (0.15)
|
||
- **5 prompt types** with weights:
|
||
- Buying intent (2.0) — 30% of pool
|
||
- Comparison (1.5) — 25%
|
||
- Specific need (1.5) — 20%
|
||
- Informational (0.3) — 15%
|
||
- Brand-direct (0.3) — 10%
|
||
- **4 cross-signals** with maximum total bonus +20%:
|
||
- Wikidata entry (≥90 days, ≥5 triples): +5%
|
||
- Trustpilot/Opineo (>50 reviews, ≥4.0 average, no review bombing): +5%
|
||
- Reddit organic mentions (>10 in niche subreddit, account age + karma weighted): +5%
|
||
- Google AI Overviews presence (verified via SerpAPI): +5%
|
||
- **Anti-gaming protections:** rank-jump flag (>30), fresh Wikidata exclusion (<90 days), review bombing exclusion, sock puppet detection (Reddit), prompt injection scrape filters (CSS hidden text, off-screen content, font-size:0)
|
||
- **Honeypot brand** mechanism for detecting AI training data circular logic and unauthorized scraping
|
||
- **Statistical methodology:** 95% confidence intervals via bootstrap resampling, overlapping CIs reported as tied (no false precision), 100 prompts × 3 models × 2 repetitions = 600 queries per category per cycle in pilot
|
||
- **Tier system:**
|
||
- Tier 1 — large markets (>1000 brands, >100M PLN GMV) — monthly scan
|
||
- Tier 2 — medium markets (100-1000 brands, 10-100M PLN GMV) — quarterly scan
|
||
- Tier 3 — niche markets (<100 brands, <10M PLN GMV) — semi-annual scan
|
||
- **11 pilot categories (PL, all Tier 2):** kosmetyki naturalne, suplementy / nutricosmetyki, diety pudełkowe, premium pet food, kawa specialty, czekolada rzemieślnicza, kursy programowania / IT bootcampy, kliniki estetyczne / dermatologia, fitness studios premium, kosmetyki dla mężczyzn, świece sojowe
|
||
- **Publication policy:** 3-month validation period before first public ranking. Hybrid format — Top 10 public HTML (SEO indexed), full ranking 100 brands as PDF behind email gate. `robots.txt` disallow for GPTBot, ClaudeBot, PerplexityBot, CCBot, Google-Extended on full-data endpoints.
|
||
- **Right to reply:** each brand profile page includes "Brand response" section, moderated for factual accuracy, 30-day response window per cycle
|
||
- **Monetization policy:** ranked brands NEVER pay Citee directly (hard rule). Revenue from Citee Pro SaaS (paid by shops optimizing visibility, not ranked brands), Industry Reports (paid by agencies/media), and Sponsored Custom Research (commissioned for category research, not brand-specific)
|
||
- **Prompt curation process** (6 stages): persona generator → prompt brainstormer → reality check (Google Trends, Reddit, Quora) → multi-agent validation (3 critics) → pilot test run → human approval
|
||
|
||
### Notes
|
||
|
||
This is **v1.0.0 — methodology release only**. No ranking has been computed or published. Foundational document establishing the framework.
|
||
|
||
First scan cycle planned: late May 2026 (private validation).
|
||
First public ranking publication target: August 2026 (after 3 validation cycles).
|
||
|
||
---
|
||
|
||
## Pre-history
|
||
|
||
Project began as "AIO Visibility" module within LMW Pulse SaaS in March 2026. Pivoted to standalone product `citee.ai` in May 2026 after market analysis showed no global competitor publishing public AI visibility rankings (27+ tracked SaaS dashboards but zero public rankings).
|
||
|
||
Strategic shift from advisory-board-driven model (Gartner / Forbes 30 Under 30 pattern) to algorithm-first model (Glassdoor / Trustpilot / FIDE / PageRank pattern) decided 2026-05-03 based on principle: "the tool must defend itself, not by authority."
|