Verifiable
Every claim, sourced.
We’re the company that argues for verifiability. So here’s ours.
Three layers, kept separate: empirical claims (prevalence, adoption, survey shares) carry source, date, and sample size; ethical and pastoral conclusions draw on separate reasoning; advocacy framing (Movemental’s thesis, formation-before-deployment) is labeled as argument, not as survey fact. Percentages from different studies are not interchangeable — see the findings page for paired comparisons only.
For Movemental’s full source-quality rubric (evidence vs. formation language, question router, maintenance cadence), see authoritative-sources-ai-nonprofits-faith-formation in the graded-high research corpus (maintainer reference, not a public reader).
- 1Virtuous & Fundraising.AI — The 2026 Nonprofit AI Adoption Report92% use AI in some form; only 7% report major organizational capability improvements; 47% have no governance policy.
- 2Barna & Pushpay — State of Church Tech 202660% of church leaders use AI monthly; only 5% of churches have an established AI policy.
- 3McKinsey — The State of AI in 202588% use AI in at least one function; 6% qualify as high performers; 55% of high performers redesigned workflows.
- 4BCG — The Widening AI Value Gap (Build for the Future 2025)60% generate no material value from AI; 5% qualify as future-built with AI-first operating models.
- 5MIT NANDA — The GenAI Divide: State of AI in Business 2025Roughly 95% of corporate gen-AI pilots fail meaningful financial returns — integration and governance, not model quality, drive outcomes.
- 6Lifeway Research — Pastors' Views on Artificial Intelligence42% of pastors use AI for ministry; 62% concerned about non-disclosure of AI involvement.
- 7Lifeway Research — Churchgoers and AI in Sermon Preparation43% disagree with their pastor using AI for sermon preparation (24% strongly).
- 8Stanford HAI — AI Index 2026Documents broad organizational uptake with uneven depth — transformation lags adoption.
- 9Dell'Anna et al. — Navigating the Jagged Technological FrontierLarge gains inside the frontier (+12% tasks, 25% faster); 19 points less accurate on one outside-frontier task.
- 10Pew Research Center — Artificial IntelligenceGold-standard US public opinion; use for disclosure reactions and trust trends, not vendor ROI.
- 11NIST — AI Risk Management FrameworkVoluntary US baseline for AI risk management; widely referenced by boards and CIOs.
- 12Forvis Mazars — AI Governance for Nonprofit BoardsAI governance as a fiduciary issue — Care, Loyalty, and Obedience extend to technological oversight.
- 13TechSoup & Tapp Network — State of AI in Nonprofits 202585.6% exploring AI; 24% with formal strategy; larger orgs adopt at roughly twice the rate of smaller ones.
- 14FBI IC3 — Internet Crime Complaint Center (AI-enabled fraud)$893M in AI-attributed losses in 2025; voice-cloning fraud up sharply — sector underestimates exposure.
- 15NPR — Clergy grapple with the ethics of using AI to write sermons12% comfortable with AI-written sermons; 43% see merit for preparation and research.
- 16NTEN — 2025 Data Empowerment ReportAI data policies among the least implemented governance items as tool adoption rises.
- 17Barna & Gloo — State of the Church 2026 (AI & spiritual authority)Public attitudes on AI and spiritual authority — pair with Lifeway pastor data for leader/congregation gap.
- 18Edelman Trust Barometer 2025 — AI flash pollUneven comfort with corporate AI use; geographic divergence in AI trust.
- 19Authors Guild — 2023 Author Income SurveyMedian book income $10,000 for full-time authors; $2,000 for whole sample including part-time; half of full-time authors below minimum wage on all writing work.
- 20Substack — How much does Substack cost?10% platform fee on paid subscriptions plus Stripe processing (2.9% + $0.30) and 0.7% recurring billing on renewals as of July 2024.
- 21Pew — U.S. streaming and subscription habits83% of U.S. adults use streaming services — reminder that attention and wallet share are contested even when creator tools are cheap.
- 22Google — Search Quality Rater GuidelinesPrimary reference for Experience, Expertise, Authoritativeness, and Trustworthiness — raters align systems, not hand-rank pages.
- 23Aggarwal et al. — GEO: Generative Engine OptimizationIntroduces GEO framing and benchmark; reports material but bounded improvements with domain dependence — not a settled spec from Google or OpenAI.
- 24Reuters Institute — Digital News Report 202540% overall trust in news (stable three years); 58% globally worry about true vs false online (73% U.S.); AI expected to make news cheaper but less trustworthy.
- 25Stanford HAI — Labeling AI-Generated Content May Not Change Its PersuasivenessLabels can shift attribution without reliably reducing persuasion for some message types — disclosure alone is insufficient for credibility protection.
- 26Pew — Reactions to learning AI was involved in content71% would view a political candidate less favorably if AI helped write a speech they liked; 56% negative for a news article — post-hoc disclosure matters.
- 27Hu et al. — Parameter-efficient fine-tuning of large language modelsSurvey of PEFT methods (LoRA, adapters); tenant-specific adaptation is feasible but entangled with base-model updates and data rights.
- 28Ippolito et al. — Automatic detection of generated textHuman-likeness and machine detectability are not aligned — optimizing for plausibility can diverge from statistical consistency.
- 29COPE — Authorship and AI toolsAI cannot be an author; disclosure required; humans accountable for every line.
- 30Dwivedi et al. — So what if ChatGPT wrote it?ChatGPT-era authorship stock-take — governance, accountability, and epistemic risk remain open; voice preservation not solved.
- 31OpenAI — Introducing ChatGPTPublic launch 30 November 2022 — fair marker for how recent lived generative-AI practice is at scale.
- 32BCG — Generative AI and knowledge workers (Sept 2024)Follow-on experiments with consultants — gains on some tasks, failure modes when junior staff use AI without senior verification.
- 33Fernandes et al. — AI-assisted reasoning and calibrationAI-assisted work can improve performance while worsening calibration — overconfidence sometimes higher among users with greater AI literacy.
- 34Ahrefs — What percentage of new content is AI-generated?74.2% of newly detected English pages contained some AI-flagged content; 2.5% pure AI, 25.8% pure human — detector-dependent, not peer-reviewed.
- 35Cooke et al. — As Good As A Coin TossRealistic multimodal synthetic stimuli — mean human discrimination accuracy near chance (~50%); do not rely on unaided perception alone.
Sources current as of mid-2026. Verify against each provider before relying on them.