Methodology

GiveRadar's records come from two tiers of sources, and every charity page says which tier it is built on.

Tier 1: official government sources. Government charity registries, tax authorities, and regulatory filings: the IRS (US), the Charity Commission for England and Wales, OSCR (Scotland), CCNI (Northern Ireland), the CRA (Canada), the ACNC (Australia), and dozens more. These provide the core record: registration, legal status, filings, and officers.

Tier 2: vetted charity directories. Some countries publish no machine-readable registry data, or their registry allows no public access. For those countries we source from vetted charity directories (for example humanitarian coordination directories), and the charity page says "listed on GiveRadar" rather than claiming registry verification. Where a country's official registry provides no public data access, the integrity assessment breakdown says so explicitly instead of penalizing the charity for its country's data policy.

On top of both tiers, we enrich profiles with information charities publish themselves: websites, contact details, social profiles, and financial documents. Enriched data is labeled as such, and verified charity owners can add self-reported information, which is always marked self-reported. We do not use third-party estimates for core data, and nothing on GiveRadar is paid placement.

For every data point we track provenance: which source it came from and when. The complete country-by-country source list, with links to each original register, is on our data sources page.

Raw registry data is messy, incomplete, and stored in dozens of formats and languages. Turning it into usable information takes six steps:

Import. We ingest each source using official data feeds, bulk downloads, or public API access, with a dedicated parser per source. We do not circumvent access controls.

Matching. A single charity may appear in multiple registries (a large NGO registered in the US, UK, and Canada). We match records across registries using legal name similarity, registration number lookups, and address verification to build one unified profile where possible.

Cleanup and exclusion. Government registers contain entries that are not charities: government bodies, businesses, dissolved entities, duplicates. We exclude non-charities and merge duplicate listings. Exclusion is reversible (records are hidden, not destroyed), and dissolved charities keep an informational page stating their removal from the register.

Enrichment. Registry data often lacks websites, emails, or phone numbers. We cross-reference public sources, including the charity's own website, to fill gaps, and we label what came from the original registry versus what we added.

Normalization. Financial data is converted to consistent units (with original currency preserved), dates standardized, country codes harmonized, categories and cause areas mapped to one shared vocabulary.

Scoring and monitoring. Every profile then receives the integrity assessment (section 3), red flag checks (section 4), badges (section 5), and GiveRadar Analysis (section 6), and enters the monitoring cycle (section 7).

Every charity receives a 0-100 integrity assessment computed from public data. It measures one thing: how much verifiable information about this charity is available. It does not measure impact, effectiveness, or whether the finances look "good".

The five components, signal by signal

For countries whose official registry allows no public data access, an admin-reviewed evidence check (the organization submits its registration ID plus an official document, which we verify against the country's official lookup) scores like registry-sourced data and is labeled as a manual check. Where no such check exists, the breakdown states that the missing points reflect the country's data policy, not the charity's failure.

This component measures whether a charity has disclosed its finances, never whether the finances look good. There are no overhead-ratio points, no program-spending tiers, and no revenue-trend judgments: a new charity that publishes one full audited statement can earn the full 30 points.

Measured from the most recent dated evidence we hold: the latest financial filing year, the latest financial document the organization published, the latest verification date, or, when none of those exist, the most recent refresh of the profile itself.

Adjustments

Red flag penalty (up to -20). Red flags (section 4) subtract from the score: high executive pay -3 (-5 if severe), stale filings -3, bare-minimum disclosure -2. The combined penalty never exceeds -20.

Self-reported data rule. If a charity has NO official source at all (neither an official registry nor official financial data), its Financial Transparency and Governance points are halved. A charity with at least one official source is never penalized this way.

Score labels

We deliberately use grey, not red, for low scores. Red is reserved for the red flag indicator, which surfaces concrete concerns independent of the score.

How this scoring system has changed

We document scoring changes because a score you cannot audit is a score you should not trust.

What the score does NOT measure

• Whether a charity's programs are effective at solving the problems they address
• Whether donating here produces more impact per dollar than donating elsewhere
• The quality of the scientific evidence behind the charity's interventions
• Long-term outcomes for beneficiaries

These are questions of impact, not disclosure. Deep impact evaluation requires methodology that effective-giving researchers like GiveWell, Giving What We Can, Animal Charity Evaluators, and Stichting Effectief Doneren apply to a small number of charities with strong evidence bases. GiveRadar covers millions of charities those evaluators will never reach; we provide the best disclosure signals public data supports, as a starting point for research, not a substitute for impact evaluation.

Alongside the score, we automatically detect three patterns that may warrant a closer look. Exact rules:

These are the only red flag types. We deliberately do not penalize charities for things outside their control: we removed earlier flags for revenue swings and news coverage (revenue can drop for legitimate reasons; a charity does not control the media), and we do not perform sanctions, AML, or watchlist screening (an earlier sanctions flag was removed because it produced almost entirely false positives on common charity names). Per-charity news is displayed for context and never affects the score.

A red flag is a signal, not a verdict. A thin profile may reflect a jurisdiction with minimal reporting requirements; high compensation may have a defensible explanation. We surface signals so donors can ask informed questions, not to accuse organizations.

Charity pages display badges where an organization holds a recognized status. There are three kinds, and they follow different rules:

Government statuses. Tax-deductibility and registration statuses granted by the state: 501(c)(3) recognition (US), ANBI (Netherlands), DGR (Australia), section 8A approval (Denmark), CRA registration (Canada), ACSO registration (Ethiopia), and equivalents elsewhere. These come from the issuing authority's public register.

Independent quality seals. Sector accreditations such as CBF-Erkend (Netherlands) and the DZI Spenden-Siegel (Germany), displayed where a charity holds them. Quality seals are shown for information only: since April 2026 no third-party rating or seal feeds the integrity assessment.

Evidence-verified registration. For countries whose official registry allows no public data access (currently Indonesia, the Philippines, Malaysia, and Vietnam), a verified profile owner can submit their official registration ID together with an official document (a decree or certificate). We check the submission manually against the country's official lookup before it is accepted, and the result is labeled as a manual check. This feeds the Registration component exactly like registry-sourced data (section 3), because it is the same fact established by a different route.

The rules that apply to every badge: badges cannot be bought, they are granted only on the issuing body's say-so, and where the issuing register publishes status changes we re-check on a recurring schedule and remove the badge if the status is revoked. The charity page shows when a status was last checked. Badges add no points to the integrity assessment beyond the registration signals described in section 3.

Each charity page carries a written GiveRadar Analysis that puts the organization in context. Its method:

Comparison groups. Every charity is compared against the tightest group that is large enough for a fair comparison (at least 30 organizations), walking a ladder: same category and country, then same country, then same category worldwide, then all charities. The analysis always names which group a comparison used, so a claim like "more filing history than most education charities in Kenya" is never silently a worldwide comparison. The peer box on charity pages additionally matches by revenue size band (under $100K, $100K to $1M, $1M to $10M, $10M to $100M, above $100M), so a village clinic is not listed next to an international hospital network.

What it compares. Observable disclosure facts from the same data documented above: years of filing history, leadership disclosure, contact coverage, number of independent sources, founding year, revenue size band. The analysis reports where a charity stands relative to its group (for example, more consecutive filing years than most peers).

Honesty rules. The analysis is evidence-first and deliberately conservative. It only states what the underlying data supports, links each observation to where you can verify it, and never converts a percentile into a recommendation. When even the broadest group is too thin, or when a metric barely varies within the group, the analysis drops the comparison and states plain facts instead of forcing a misleading percentile. Benchmarks are recomputed weekly from the full dataset, and each analysis states the date its comparisons were generated.

What it is not. Not a rating, not a ranking of merit, not a donation recommendation. It answers "how does this organization's public footprint compare with similar organizations", nothing more.

Profiles are maintained, not archived:

• Re-imports. Sources are re-imported on a per-country cadence; new filings appear after charities submit them and registries publish them.
• Registry re-checks. For registers that publish status changes, we re-check registration status on a recurring schedule that varies per register (weekly for the most active registers, up to semi-annual for others), including accreditation and certification statuses.
• Nightly change detection. A nightly process detects meaningful changes: deregistrations, new filings, new red flags, and integrity score changes of 10 points or more. Users who follow a charity can receive these as email alerts.
• Dates on every page. Each charity page shows when its data was last updated and, where an official re-check applies, when its registry status was last checked.
• Dissolved organizations. When a register removes a charity, its page becomes an informational notice stating the removal, rather than silently disappearing: visitors arriving from old links deserve to know.

GiveRadar is extending this methodology with AI-assisted capabilities:

AI translation of underserved languages. We are translating charity data from non-English registries (Japanese, Korean, Thai, Chinese, Arabic, Hindi, and others) into accurate English. Low-resource translation is an area where AI performs unevenly; we mitigate with native-speaker quality auditing, conservative confidence thresholds, and provenance flags showing when a description is AI-generated versus human-curated.

AI-assisted research context. For each charity, a research layer that identifies the problem and intervention it works on, synthesizes what the scientific literature says about effectiveness, and flags evidence strength as strong, limited, or absent. This is not an impact scoring system: where data is sparse we say so and do not produce confident scores the evidence cannot support. The methodology will be published openly when the first version is complete.

Open data and an AI-readable interface. Methodology documentation, data schemas, and research outputs released under open licenses (CC-BY 4.0 for documentation and research, MIT for reference code), so AI assistants, researchers, and developers can build compatible systems. Live data and the API continue under tiered access: free for individual donors and researchers, paid commercial access for organizations.

This work is informed by conversations with Goede Doelen Nederland, Stichting Effectief Doneren, and Kenniscentrum Filantropie, who have reviewed and informed our approach.

Today we use AI for data normalization, entity matching across registries, translation support, and text cleanup of corrupted registry data. The risks of AI in charity research are real: translation errors can misrepresent small charities, and generated text can produce confident-sounding mistakes that misdirect donor money. We mitigate through:

• Native-speaker translation auditing with conservative confidence thresholds that flag uncertain translations for human review
• Explicit "insufficient data" statements rather than synthesized confident assessments when the evidence is too thin
• Provenance tracking on every data point so users see where information came from
• A public correction process so any organization can dispute or update its profile
• Data minimization on research pilot participants, aggregate-only reporting
• Documented failure modes, so the sector learns where AI helps and where it should not be trusted

We make mistakes. Registry data is sometimes wrong at the source, enrichment sometimes misidentifies information, and assessments sometimes miss context.

If information about a charity on GiveRadar is inaccurate, outdated, or missing important context, contact info@giveradar.com with:

• The charity name and its URL on GiveRadar
• What specifically is inaccurate
• What the correct information should be
• A link to a public source supporting the correction, where possible

Corrections are reviewed and typically processed within 5 business days. Where relevant we consult sector partners to verify contested claims. If we update information based on your correction, the profile's update date reflects it.

Organizations can also claim their profile to maintain their own description, contact information, and donation link, and to add self-reported financial documents (always labeled as such). Claiming cannot modify official financial data or buy a better score; the one score effect is transparent and positive: a claimed and verified profile earns the 4-point verification signal in the Registration component, because verified ownership is itself a disclosure signal.

Honesty about limitations is as important as the methodology itself:

• Coverage is uneven. Some countries publish detailed financial data (US, UK, Canada, Australia); others publish basic registration only; some publish nothing, where we rely on vetted directories. Our methodology is only as good as each country's source data.
• Small local charities have less data. A village NGO with minimal regulatory requirements will score lower on data availability than a large US nonprofit. That is a statement about available data, not about the organization. The grey label and its note exist for exactly this case.
• Financial year conventions vary. Fiscal years, deadlines, and filing formats differ by country; cross-country year comparisons should be read carefully.
• Enrichment is best effort. Added contact details and links come from public sources and are labeled; we do not guarantee them.
• The integrity assessment is not impact evaluation. Said elsewhere on this page, worth repeating: we measure disclosure, not program effectiveness.
• The analysis inherits every limitation above. GiveRadar Analysis compares the data we hold; where a cohort's data is thin, comparisons are correspondingly weaker, which is why it declines to rank in thin cohorts.
• Languages outside current coverage. Where we have not yet built translation, we store original-script data (see section 8).

• Questions about methodology: info@giveradar.com
• Questions about a specific charity: use the correction process in section 10
• Programmatic access for developers and researchers: see the API page
• Journalists and academic researchers who need expanded free access: contact us directly