Donor education · 5 min read
How the match score is calculated: the actual math.
Most platforms that show donors an "AI-generated" percentage match refuse to explain what the number means. That's the wrong trade. The percentage is the most persuasive thing on the screen, and if it isn't auditable it's just a confidence trick. Here is the line-by-line breakdown of every score GIVE→ALIGN shows you, including the rows we hide behind the popover, so the next time you see a 91% on a card you can rebuild it on a napkin.
Published June 15, 2026 · By the GIVE→ALIGN team
Why we publish the formula
The match score is shown next to the donate button. If we got the math wrong, or if we secretly inflated it, the consequence would land on a donor's decision — and on the nonprofit on the other side of the click. Publishing the weights, the components, and the worked example is the cheapest accountability mechanism we have. The source function lives in our backend at match.py · _score_match() and the donor-facing breakdown popover is the same data, rendered.
Step zero: the hard pre-filter
Before any scoring happens, every nonprofit must pass four non-negotiable gates. These aren't weighted — they're on/off. An organisation that fails any one of them never reaches your screen and never receives a score at all.
- ·Moderation status — manually approved by our team. Any nonprofit with a pending or rejected status is excluded.
- ·Verified governance — either a U.S. 501(c)(3) status with an EIN on the IRS Business Master File, or a citation to a public charity registry (UK Charity Commission, ACNC, India 12A/80G, and so on). "Trust us" is not a category.
- ·Region compatibility — the nonprofit either operates in the region you selected, or operates globally, or you left the region open. A children's nonprofit working only in Southeast Asia will not appear if you selected Europe.
- ·At least one cause overlap — if you selected "environment" and "animal welfare", the nonprofit must have at least one of those in its tagged cause set. Causes aren't free-text on our end; they come from a closed vocabulary so the overlap check is reliable.
The six rows of the score (sum = 100)
Once a nonprofit has cleared the pre-filter, the displayed percentage is the sum of six components, capped at 100. Every row corresponds to a public field on the nonprofit's detail page; nothing here is computed from a hidden signal.
Made the cut
50 ptsEvery NGO that reaches the scoring step has already passed four hard pre-filters: approved by our moderation team, verified governance (IRS 501(c)(3) with EIN or a public state/national registry citation), region-compatible with what you asked for, and at least one cause overlap with what you care about. Appearing on your screen is already a strong signal, and the 50-point base reflects that.
Cause depth
up to 15 ptsThe pre-filter requires one cause overlap. Each ADDITIONAL overlap beyond the first earns +3 points, capped at +15 for five or more matching causes. A nonprofit working on both climate and disaster relief for a donor who selected both will score higher than one that matches only on climate.
Region precision
up to 15 ptsExact region match earns 15. If you left the region open (or selected Global), every match gets 13 — the field literally doesn't constrain the result. A nonprofit operating globally when you picked a specific region gets 12. A nonprofit operating in a different region than you asked for can't appear here at all; the pre-filter already excluded it.
Verification
up to 10 ptsU.S. 501(c)(3) status with a verified EIN on the IRS Business Master File earns 10. A public state or national registry citation (UK Charity Commission, ACNC, etc.) earns 8. A nonprofit that has passed our manual moderation without a public registry ID — rare — earns 5. The 10/8/5 ladder is deliberate: a researchable government record is worth more than our internal review, even when we trust the internal review.
Transparency
up to 5 ptsEach nonprofit carries a 0–100 transparency score derived from the completeness of its public disclosures (annual report on site, audited financials, named board, mission clarity). We scale that proportionally to a 0–5 contribution: a score of 90 earns 4-5 of these points, a score of 60 earns 3. The same transparency value is shown on the nonprofit's detail page so you can audit how the multiplication is done.
AI ranking
up to 5 ptsClaude returns its top five picks in a strict relative order. The number-one pick earns 5 points, the second pick 4, and so on down to 1 point for the fifth. This is the only component of the score that isn't pure math against database fields — it's the AI's contribution. It's also intentionally capped at five points so the AI can never override a strong objective signal. We expose this row so you know exactly how much of your match is heuristic and how much is large-language-model.
| Component | Points | How it's computed |
|---|---|---|
| Made the cut | 50 | Every NGO that reaches the scoring step has already passed four hard pre-filters: approved by our moderation team, verified governance (IRS 501(c)(3) with EIN or a public state/national registry citation), region-compatible with what you asked for, and at least one cause overlap with what you care about. Appearing on your screen is already a strong signal, and the 50-point base reflects that. |
| Cause depth | up to 15 | The pre-filter requires one cause overlap. Each ADDITIONAL overlap beyond the first earns +3 points, capped at +15 for five or more matching causes. A nonprofit working on both climate and disaster relief for a donor who selected both will score higher than one that matches only on climate. |
| Region precision | up to 15 | Exact region match earns 15. If you left the region open (or selected Global), every match gets 13 — the field literally doesn't constrain the result. A nonprofit operating globally when you picked a specific region gets 12. A nonprofit operating in a different region than you asked for can't appear here at all; the pre-filter already excluded it. |
| Verification | up to 10 | U.S. 501(c)(3) status with a verified EIN on the IRS Business Master File earns 10. A public state or national registry citation (UK Charity Commission, ACNC, etc.) earns 8. A nonprofit that has passed our manual moderation without a public registry ID — rare — earns 5. The 10/8/5 ladder is deliberate: a researchable government record is worth more than our internal review, even when we trust the internal review. |
| Transparency | up to 5 | Each nonprofit carries a 0–100 transparency score derived from the completeness of its public disclosures (annual report on site, audited financials, named board, mission clarity). We scale that proportionally to a 0–5 contribution: a score of 90 earns 4-5 of these points, a score of 60 earns 3. The same transparency value is shown on the nonprofit's detail page so you can audit how the multiplication is done. |
| AI ranking | up to 5 | Claude returns its top five picks in a strict relative order. The number-one pick earns 5 points, the second pick 4, and so on down to 1 point for the fifth. This is the only component of the score that isn't pure math against database fields — it's the AI's contribution. It's also intentionally capped at five points so the AI can never override a strong objective signal. We expose this row so you know exactly how much of your match is heuristic and how much is large-language-model. |
A worked example: 86%, line by line
Say you tell us you care about climate and disaster relief, you leave the region open, and the American Red Cross comes back as Claude's top pick. The card displays an 86% match. Here's exactly where those 86 points come from.
| Made the cut | +50 |
| Cause depth (1 extra overlap × 3) | +3 |
| Region precision (open to any) | +13 |
| Verification (IRS 501(c)(3) + EIN) | +10 |
| Transparency (92/100 × 5/100) | +5 |
| AI ranking (#1 of 5) | +5 |
| Total displayed on the card | 86% |
Swap any input and the score moves in a predictable direction. Pick a specific region instead of leaving it open and you replace the 13-point "open to any" row with either 15 (Red Cross operates in that region) or 12 (Red Cross operates globally), shifting the total by one or two points. Pick a nonprofit Claude ranked #4 instead of #1 and you lose three of those AI points. There is no unobservable knob.
Why an AI ranking row exists at all
If five candidate nonprofits all share the same causes, the same region precision, and the same verification source, our deterministic math will return identical integer scores for all of them. Donors quite reasonably distrust five cards in a row that all read 84%. The AI-ranking row uses Claude Sonnet's relative ordering of the top five picks (5 points for #1 through 1 point for #5) as the tie-breaker. It's deliberately the smallest column in the table: the AI cannot override a strong objective signal, but it can break a tie among objective signals that are already equivalent. The click-out donation model means the same Claude credit is paid for whether you give $5 or $5,000, so we have no incentive to inflate any score in either direction.
What the match score deliberately won't tell you
The score is an alignment number, not an impact number. It measures how well a nonprofit fits the causes, region, and governance constraints you gave us — nothing more. A 96% score does not mean a 96-cents-on-the-dollar guarantee, and it does not encode third-party impact ratings. For impact diligence the right next step is the nonprofit's most recent IRS Form 990 — linked from every detail page — plus a third-party evaluator like GiveWell or ImpactMatters for the cause area.
The score also has no memory. Two donors with identical inputs get identical scores. That's a feature, not a limitation: a personalised algorithm that nudged you toward whichever nonprofit could pay for placement would defeat the entire point of an independent donor-matching platform. We don't take placement money, and the deterministic math is how we prove it.
See it in action
Every score on every card has a popover. Open one.
On every match card, the percentage chip in the top-right corner opens the same six-row breakdown you just read, populated with the actual numbers for that nonprofit. No black box, no marketing copy — just the six rows summing to whatever appears on the card.