Our methodology
This is not a chatbot.
cleargiving.io uses a structured evidence synthesis pipeline built around your foundation's specific scoring model. The result is a scored, traceable, human-reviewed assessment. Not a generic AI summary.
Generic AI produces generic answers.
Large language models are remarkable synthesisers. Given a charity name, a capable model can return a plausible-sounding summary. But plausible is not the same as scored. Synthesised is not the same as traced. And a summary that applies the same lens to every charity is not the same as an assessment that reflects your foundation's values.
Without a defined scoring model, a structured research pass, and a mandatory human review step, AI-generated assessments are impressionistic: useful for orientation, not for funding decisions.
"Plausible is not the same as scored."
A five-stage evidence pipeline.
Every assessment runs the same stages, in the same order, with the same checkpoints.
Scoring model & context
Your foundation's mission, values, priorities, and weighted criteria are baked into every assessment from the first line.
Research pass
Before scoring, the model cross-references Charity Commission, Companies House, the charity's website, and published accounts. Sources are recorded.
Parallel scoring
Each scoring model criterion is scored independently against the evidence, with a rationale, confidence level, and cited sources returned per criterion.
Synthesis
A separate synthesis pass produces an overall summary, key strengths, principal concerns, and open questions for trustee discussion.
Human review
Every score, rationale, and summary can be read, challenged, and overridden. No funding decision leaves the platform without a documented human judgement.
Your scoring model. Not ours.
Before the evidence pipeline runs, cleargiving.io builds a complete context document from your workspace configuration: your foundation's mission, your strategic priorities, your geographic focus, and every scoring model criterion with its weighting. This is injected into every assessment as the scoring framework.
Two foundations assessing the same charity through cleargiving.io will produce different scores, because their criteria, weights, and strategic contexts differ. That difference is the point. An objective score is meaningless without a defined objective.
- Mission statement shapes the system prompt for every assessment
- Scoring model criteria and weights define what counts as a strong or weak score
- Geographic and strategic context adds nuance the model would otherwise miss
"Two foundations. Same charity. Different scores. That's correct."
Evidence quality is graded independently of the score.
A charity can score well on a criterion and have limited evidence, because the available evidence is consistent, even if sparse. A charity can score poorly and have strong evidence, because the evidence clearly shows a gap.
cleargiving.io surfaces both: the score and the quality of the evidence behind it, so trustees understand not just what the assessment found, but how confident to be in it.
Strong
Multiple corroborating sources; evidence is consistent and complete
Limited
Some evidence present but sparse, dated, or from a single source
For Review
Evidence is ambiguous, conflicting, or requires trustee clarification
Missing
No evidence found; criterion cannot be scored from available public sources
Built-in guardrails against invented facts.
AI models can confabulate, generating plausible-sounding details that are not supported by sources. In charity assessment, that is unacceptable.
cleargiving.io's research pass includes explicit anti-default instructions: if a charity has a registration number, that is a findable fact and the model is instructed to surface it or flag it as missing, not estimate it. If governance information is unavailable publicly, the model must record that gap as a 'Missing' evidence grade, not substitute an assumption.
Transparent scoring mechanics.
Each criterion is scored on a 0–2 scale. Scores are then weighted by the importance you assigned each criterion in your scoring model. The weighted total is normalised to a 0–10 scale and mapped to a tier band.
0
Not evidenced or significantly below expectations
1
Partially met: some evidence but gaps or inconsistencies
2
Well evidenced and meets expectations for this criterion
AI confidence levels
High
Multiple strong, independent sources corroborate the criterion assessment. The model is confident the score reflects what evidence shows.
Medium
Evidence exists but is partial, indirect, or from a single source. The score reflects available evidence with some inferential weight.
Low
Evidence is sparse, dated, or tangential. The score should be treated as provisional and may warrant trustee enquiry.
AI recommends. Humans decide. Always.
Every criterion score produced by the pipeline can be overridden by a reviewer. Every override requires a documented reason. The audit trail preserves both the AI's original position and the human's final judgement, side by side.
No assessment can be marked as a funding decision without a human review step. This is not a feature that can be disabled. It is how the platform works.
"No funding decision leaves cleargiving.io without a documented human judgement."
What the AI records
- Score
- Confidence level
- Rationale
- Sources cited
- Evidence grade
What you can add
- Override reason
- Final score
- Decision outcome
- Named signatory
Independent by design. Secure by default.
cleargiving.io has no commercial relationship with the charities it assesses. There are no placement fees, no featured listings, no referral arrangements. Our only commercial relationship is with the foundations who use the platform.
Technically: workspace data is isolated, encrypted at rest on Google Cloud infrastructure, and never used to train AI models. Your assessments remain yours.
- No commissions or placement fees from assessed charities
- Workspace data isolated and encrypted at rest
- Assessments never used for AI model training
See the methodology in practice.
Request access and run your first assessment.