---
source: marlvel.ai
type: methodology
title: Marlvel.ai Methodology
date: 2026-06-03
license: CC-BY-NC-4.0
canonical_url: https://marlvel.ai/methodology
citation: Marlvel.ai, Methodology, June 2026. https://marlvel.ai/methodology
---

# Marlvel.ai Methodology

> How Marlvel.ai provides independent mobile app intelligence reports for the US market. We continuously improve our analysis, accuracy, and coverage.

Last updated: June 3, 2026

## Our Mission

Marlvel.ai's mission is to help mobile builders improve their existing apps and create new ones, so they can unleash their creativity. We provide intelligence reports across 60 App Store categories, covering 14,800+ top apps in the US market. Our reports offer an objective, data-driven view of the mobile app landscape. No app publisher pays for coverage, influences our analysis, or reviews reports before publication. Our goal is simple: give builders the insights they need to make better decisions.

## Data Sources

Our analysis is built on publicly available signals from multiple independent sources. We cross-reference data points to reduce bias and increase confidence in our findings.

- **App Store Listings:** Metadata, descriptions, version history, screenshots, and pricing from iOS App Store and Google Play.
- **User Reviews:** Ratings and written reviews from both platforms, analyzed for sentiment patterns, recurring themes, and evolving user perception.
- **Developer Websites:** Official websites, about pages, press releases, and public documentation from app publishers.
- **Public Market Signals:** App store rankings, chart positions, category trends, and competitive landscape data from public sources.
- **Community Signals:** Public discussions and user-generated content that reflect real-world user experience and market perception.
- **Store Metadata:** Technical metadata including bundle identifiers, platform availability, content ratings, and update frequency.

## Analysis Pipeline

Each intelligence report is produced through a multi-stage analysis pipeline that combines AI-powered processing with structured analytical frameworks.

### 1. Signal Collection & Normalization
We aggregate data from all available public sources for a given app. Raw signals are cleaned, deduplicated, and normalized into a structured dataset that can be analyzed consistently across thousands of apps.

### 2. Feature & Market Positioning Analysis
Our AI identifies the app's core features, monetization model, target audience, and competitive positioning. Each feature is classified as either a market standard or a differentiator based on category benchmarks.

### 3. User Sentiment Analysis

We analyze user reviews across platforms to extract sentiment patterns, including recurring praise themes, pain points, and emerging trends. We require a minimum of 5 reviews to generate sentiment data. Below this threshold, we flag confidence as low rather than guessing.

Each app receives a **Sentiment Score (0–100)** and a **Sentiment Label** based on our proprietary 5-level taxonomy:

| Label | Score Range | What It Means |
|-------|------------|---------------|
| **Thrilled** | 81–100 | Users love the app. Few complaints, strong loyalty. |
| **Excited** | 61–80 | Users are satisfied and vocal about it. Some friction but overall positive. |
| **Mixed** | 41–60 | Real praise coexists with real frustration. Could go either way. |
| **Frustrated** | 21–40 | Pain points outweigh the positives for most users. |
| **Upset** | 0–20 | Widespread dissatisfaction. Critical issues reported frequently. |

The sentiment score combines star ratings and volume with AI analysis of review text (theme extraction, evidence quoting). This captures both the rating and the reasoning behind it, which is a richer signal than ratings alone.

### 4. Competitive Landscape Analysis

Each app's competitive environment is mapped using our **4-tier competitive taxonomy**:

| Tier | What It Means |
|------|---------------|
| **Nemesis** | The single closest rival. Same sub-genre, same user, maximum overlap. If this app vanished, most users would switch here. |
| **Contenders** | Strong direct competitors actively competing for the same users. Each represents a distinct competitive angle (price, features, UX, ecosystem). |
| **Same Space** | Apps in the broader ecosystem. Same category, overlapping audience, but not head-to-head. |
| **New Kids on the Block** | Emerging or fast-growing apps that represent potential future threats through innovation or momentum. |

Competitors are identified through AI analysis of features, category, audience, and market signals. We prioritize the same **sub-genre** over broad category. A surf simulation game gets compared to other surf games first, not to all sports games. We also filter for apps that are actively maintained, recently updated, and have real user bases.

### 5. Intelligence Synthesis

All collected signals are cross-referenced and synthesized into a structured intelligence report. Each app is compared against category peers and direct competitors to identify competitive advantages and gaps. The output includes SWOT analysis, market outlook, strengths and weaknesses derived from real user feedback, and actionable insights. Reports follow a consistent format designed for both human readers and machine consumption.

## Quality Assurance & Expert Review

Our team of experienced mobile industry professionals, with over 15 years of expertise in app development, product management, and mobile market analysis, continuously reviews the generated content to ensure quality.

Reviewers check reports for factual accuracy, analytical coherence, and relevance. They flag and correct incoherencies, outdated information, and misleading conclusions that automated analysis may produce. This ongoing human-in-the-loop approach ensures that our reports meet the standard of quality that mobile builders rely on.

### Confidence Scoring

Every report includes a transparent confidence score (0.0 to 1.0) that reflects how much data was available to produce the analysis.

| Level | Score | What it means |
|-------|-------|---------------|
| High | 0.7 to 1.0 | 100+ reviews, diverse data sources, strong sentiment signal |
| Medium | 0.4 to 0.69 | 20-99 reviews, limited source diversity, moderate signal |
| Low | 0.0 to 0.39 | Fewer than 20 reviews, limited data, or very recent launch |

Confidence is calculated from review volume (5 tiers: 5+, 20+, 50+, 100+, 500+), website availability, about page content, sentiment data quality, and feature documentation depth (3+ or 5+ features). Additionally, our team evaluates reports and may downgrade the confidence score if the generated information appears inaccurate or inconsistent.

### Sentiment vs. Store Rating

Our sentiment score (0-100) and the store rating (1-5 stars) are complementary but independent signals. The store rating is a cumulative historical average reflecting all-time user scores. Sentiment analysis is based on recent user reviews processed by our AI pipeline, capturing the current user experience. When they diverge (e.g., a 4.5-star app with "Frustrated" sentiment), it typically means recent reviews are more negative than the historical average. Reports include an explanatory note when this occurs.

## Update Frequency

Reports are updated on a continuous basis, with refreshes running every hour. Our target is that no report should be older than 15 days. Each refresh re-collects all public signals, re-runs the analysis pipeline, and regenerates the report when new reviews, version updates, or ranking changes are detected. Market pulse data (rankings, top movers) is refreshed daily. All reports display their last audit date prominently.

## Independence & Ethics

- **No pay-for-play:** No app publisher can pay for a favorable report, higher ranking, or removal of negative findings.
- **No editorial influence:** Publishers do not see or approve reports before publication.
- **Public data only:** We analyze publicly available information. We never access, store, or process any private or personal user data, internal analytics, or confidential business information.
- **Transparent methodology:** Our confidence scoring and data sources are fully disclosed.
- **Corrections policy:** Each report includes the ability to flag an error. We commit to reviewing all requests promptly and applying corrections when justified.

## Corrections Policy

Every intelligence report on Marlvel.ai carries a **Report an issue** link. If you spot a factual error — wrong rating, wrong publisher, misattributed feature, outdated pricing, incorrect version history — submit it via that link or the [support page](https://marlvel.ai/ai-policy#contact) with a short description and, when possible, the source that contradicts what we show.

We review reported issues within a few hours during business days. When a correction is warranted, we update the underlying data and regenerate the affected report from scratch so the intelligence reflects the corrected inputs end-to-end. We do not silently patch individual sentences: if the source changes, the analysis is rerun.

If a report is withdrawn pending correction, we return a noindex placeholder rather than a stale version, and we publish the corrected report once the inputs are verified. We do not maintain a public changelog of individual corrections, but the `dateModified` field in the report's metadata reflects the most recent meaningful update.

## AI Usage & Editorial Policy

Marlvel.ai intelligence reports are produced by a pipeline that combines publicly available data (App Store and Google Play metadata, user reviews, developer websites, release notes, chart rankings) with large language models that structure, synthesize, and write the report sections. The AI does not invent data: it only organises and summarises what the input signals contain.

To keep outputs grounded, every generation prompt enforces an explicit discipline: no LLM-jargon filler (a blacklist of filler words is injected into every prompt), no speculation when the input does not support a claim, and no verbatim quoting of user reviews (we paraphrase recurring patterns to comply with store terms of service and copyright). When a section does not have enough input data, we display a transparent fallback saying so rather than padding with generic prose.

A separate AI judge scores every report along ten axes (factual grounding, analytical depth, editorial clarity, originality, actionability, and structural coverage of overview, features, sentiment, competitive position, and outlook). Reports that score poorly on grounding or originality are flagged for regeneration before they are published. The [Reliability Index](https://marlvel.ai/about/reliability) exposes a public-facing composite score (0-100) summarising the trust we place in each report based on data solidity, freshness, and completeness.

AI generation does not replace judgement. Our team reviews prompts, taxonomy, and scoring calibration on a rolling basis, and we retire or rewrite any prompt that produces consistently low-quality output. We explicitly do not make medical, financial, legal, or political recommendations — those categories receive hedged phrasing and skip the FAQ module entirely.

## Content Compliance & Takedown

Before publishing, content is filtered with awareness of the laws and regulations applicable to the subjects we cover, to the best of our knowledge, so that what we publish stays in line with legal norms and platform policies. Sensitive categories — health, finance, politics, minors, hate, weapons, gambling, adult content — are treated with additional guardrails (hedged phrasing, disclaimers, exclusion from AI retrieval for kids content).

If a concern is raised — factual error, privacy issue, sensitive topic, rights complaint, or suspected non-compliance — we review it at our discretion and, where warranted, remove or amend the content within a few hours. Use the [support page](https://marlvel.ai/ai-policy#contact) or the "Report an issue" link on any report.

We operate under a best-effort obligation (obligation de moyens), not a guarantee of result. Reports are strictly informational: they aggregate publicly available information, we are not accountable for the underlying public sources themselves, and we do not provide professional advice (medical, financial, legal, or otherwise). Always verify critical information independently.

## See It in Action

Check out the [Candy Crush Saga intelligence report](https://marlvel.ai/apps/candy-crush-saga) as an example of a high-confidence report including user sentiment analysis with real quotes, key feature breakdown with competitive positioning, store rankings history, pros and cons, market outlook, pricing analysis, and related apps comparison.

## Known Limitations

- Apps with fewer than 5 user reviews do not receive sentiment analysis.
- Very recent app updates or launches may not be reflected immediately.
- AI-powered analysis may occasionally produce inaccurate or incomplete conclusions. Our expert review process catches most issues, but some may persist.

## Disclaimer

All intelligence reports published on Marlvel.ai are provided strictly for informational purposes, to help mobile builders improve their apps and make more informed decisions. They do not constitute guaranteed advice, recommendations, or endorsements. Marlvel.ai declines all responsibility for any decisions made based on the information contained in our reports. Use at your own discretion.

## FAQ

### Does any app publisher pay Marlvel for coverage?

No. No app publisher pays for coverage, influences our analysis, or reviews reports before publication. All analyses are fully independent.

### How is the confidence score calculated?

Every report carries a confidence score from 0.0 to 1.0 based on review volume, source diversity, and data completeness. High (0.7–1.0) means 100+ reviews and a strong sentiment signal; Medium (0.4–0.69) is 20–99 reviews; Low (below 0.4) is fewer than 20 reviews, limited data, or a very recent launch.

### How often are reports updated?

Reports refresh continuously — our pipeline runs every hour and our target is that no report is older than 15 days. Market pulse data such as rankings and top movers refreshes daily.

### Where does Marlvel's data come from?

We analyze publicly available signals only: App Store and Google Play metadata and reviews, developer websites, and public market signals such as chart rankings. We never access private or personal user data.

### Can I get a report corrected?

Yes. Every report has a "Report an issue" link. We review reported issues within a few hours on business days and, when warranted, regenerate the affected report from the corrected inputs.

## Machine-Readable Access

All intelligence reports are available in machine-readable formats for AI systems, researchers, and developers:

| Endpoint | Format | Description |
|----------|--------|-------------|
| /llms.txt | Text | Index of top apps per category with links to reports |
| /llms-full-{category}.txt | Markdown | Complete reports for a single category |
| /llms-full.txt | Markdown | Directory of all per-category report files |
| /api/llm/apps/{cat}/{slug} | Markdown | Individual app report with YAML frontmatter |
| /api/llm/categories | Markdown | Dynamic index of all categories with app counts and links |
| /api/llm/pulse | Markdown | Live US App Store rankings with rank changes and top movers |
| /api/llm/compare/{slug1}/{slug2} | Markdown | Head-to-head comparison of two apps with YAML frontmatter |
| /api/llm/index | JSON | Structured index of all apps with bundle_id, rating, sentiment, and report URLs |
| /api/llm/publishers/{slug} | Markdown | Publisher portfolio — all apps by a publisher with aggregate metrics |
| /api/llm/niche/{slug} | Markdown | Per-niche category analysis — ranked apps, verdict, awards, and FAQ |
| /api/llm/rankings | Markdown | Top store charts by country, platform, and category |
| /.well-known/ai.json | JSON | AI discovery manifest with endpoint catalog and methodology |
| /methodology.md | Markdown | This page in markdown |
| /ai-policy.md | Markdown | AI policy in markdown |

All data is freely accessible. Only fair use is accepted, under [CC-BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) license.

---

Cite as: Marlvel.ai, "Methodology," June 2026. https://marlvel.ai/methodology

All data is freely accessible under [CC-BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) license.