AI Chat Tools in Nonprofits: Project Planning and Impact Evaluation
Vendor pages document different prices, upload limits, research options, and data controls for ChatGPT, Claude, and Gemini, so this test plan shows how a nonprofit can compare them against its own acceptance rules.
Sources checked 2 Oct 2026
For nonprofit project planning and impact evaluation, OpenAI’s pricing page, Anthropic’s pricing page, and Google’s US AI plans document different prices, upload rules, research tools, and data controls across ChatGPT, Claude, and Gemini. Those pages do not establish that one assistant produces better plans or evaluations. Chat Picker has not tested the assistants for this work, as its method states, so this page provides an afternoon test plan rather than a winner. The prior version’s unsourced statistics and test results have been removed.
Know these limits before you start
Use redacted or synthetic files. These vendor limits were as read on October 1, 2026. OpenAI’s File Uploads FAQ says Free users can upload 3 files per day and every file has a 512MB hard cap. Claude’s upload guide allows 500MB per chat file and 30MB per project file; Claude analyzes text and visual elements in PDFs of 100 pages or fewer, but only text in PDFs from 101 to 1,000 pages. Gemini’s file guide allows up to 10 files per prompt, 100MB for non-video files, and 2GB per video. For a work or school account, an administrator must enable Gemini access to Drive uploads.
ChatGPT’s pricing page lists an opt-out from training on Free, Go, Plus, and Pro. Claude’s pricing page lists an opt-out on Free, Pro, and Max and says Team is not trained on by default. The Gemini plan page states no training setting, so mark that item unverified rather than assuming either result.
OpenAI’s accuracy note warns that ChatGPT may sound confident when wrong and says important information needs verification. Anthropic’s accuracy note says not to rely on Claude as the only source of truth and to inspect original sources. Google’s Gemini guidelines say outputs may reflect training-data limits or overgeneralizations; its source-viewing guide says a Sources button appears when sources are available.
OpenAI’s Usage Policies prohibit automated high-stakes decisions in sensitive areas without human review. Anthropic’s Usage Policy calls for relevant human expertise in elevated-risk work and disclosure of AI involvement in direct-to-consumer sessions. These are vendor rules, not legal advice.
The test plan
Run each task in every assistant with the same prepared files, a fresh conversation, and your normal account controls. Use a shared log covering the assistant, plan, date, prompt, file names, time taken, usage-counter change, sources shown, errors, corrections, and unresolved issues.
-
Grant proposal: speed and compliance. Prepare a fictional Green Skills Access Grant package with eligibility, deadline, budget, and reporting rules, plus a draft response. Use this prompt: “Create a compliance table with requirement, exact source location, draft response, evidence, gap, and revision. Do not invent missing facts; label every inference.” Check traceability and invented terms. Log omissions, unsupported additions, needed edits, time, and counter change.
-
Logic-model accuracy. Prepare a fictional youth-employment program using a goal, activities, outputs, outcomes, assumptions, and external factors. Use this prompt: “Build a logic model; label each causal link, assumption, and evidence gap, then provide monitoring questions.” Check for output-outcome confusion and hidden assumptions. Log broken links, unsupported claims, and corrections.
-
Stakeholder mapping. Prepare a fictional partner list with roles, interests, influence, access needs, and objections. Use this prompt: “Map stakeholders, identify gaps and representational risks, then propose an engagement sequence with an owner and purpose for each task.” Check for invented motives and generic tactics. Log missing voices, accessibility gaps, and follow-ups.
-
Impact evaluation. Prepare a de-identified CSV, its codebook, and a short set of evaluation notes. Use this prompt: “Use only measures defined in the codebook; show filters and transformations; tie each finding to rows or a note; separate observations from interpretation; draft a funder narrative with limitations.” Check traceability, missing-data handling, claim strength, and visible sources. Log corrections and unsupported statements.
-
Cost and access. Prepare current vendor pricing pages, your seat count, usage pattern, and storage need. Use this prompt: “List every documented charge, billing condition, usage limit, and unverified item. Do not total them unless all required inputs and tax treatment are provided.” Check checkout, region, billing term, and limits. Log each dated price, access condition, unknown, and counter change.
-
Security and privacy. Prepare redacted records, a data classification, consent status, and retention rule. Use this prompt: “Classify each field; propose minimization, access, retention, deletion, and human-review steps; flag unresolved points and avoid exposing unnecessary personal data.” Check settings against each claim. Log the training setting, retention choice, connected sources, permissions, and staff reviewer.
How to read your results
Before testing, define pass, conditional, or fail rules for source traceability, invented facts, file and usage fit, privacy settings, and known purchase terms. A pass means every required condition was met in your log; conditional identifies a required fix; fail identifies a blocker.
Review cited material in the original grant, dataset, policy, or vendor page rather than trusting only the assistant’s summary. Compare correction burden, time, counter changes, and verified cost only for the same task. Keep an unknown price or setting as unknown, and resolve privacy or policy blockers before adoption. Your decision may differ by task; do not turn the log into an overall winner.
Where the plans differ
The three-way comparison matrix tracks the rows that matter here: free plan, low-cost tier, main paid plan, heavy-use plans, team plan, context window, usage limits, training on chats, and ads. The vendor figures below were as read on October 1, 2026, except ChatGPT Business, which was as read on October 2, 2026. Recheck vendor checkout before subscribing.
| Matrix row | ChatGPT | Claude | Gemini |
|---|---|---|---|
| Free and low-cost | Pricing: Free has unlimited text chats but limited uploads and tools. Go shows A$13/month on the Australian page; the US price is unverified, and the plan may include ads. | Pricing: Free is available; no low-cost tier is listed between Free and Pro. | US plans: Free requires a Google Account and includes 15 GB storage. AI Plus is $4.99/month with 2x Free limits and 400 GB storage. |
| Main paid plan | Plus is $20/month, billed monthly. | Pricing: Pro is $20/month or $17/month with annual billing and $200 up front. | US plans: Google AI Pro is $19.99/month with 4x Free limits and 5 TB storage. |
| Heavy use | Pro tiers cost $100, $200, or $500/month. | Pricing: Max starts at $100/month and offers 5x or 20x Pro usage; the 20x price was not shown as a number. | US plans: Ultra is $99.99/month with 5x AI Pro limits or $199.99/month with 20x. |
| Team | Business: standard seat is $20/month on annual billing or $25 monthly; premium seat is $100 or $125. At least 2 seats. | Team: standard seat is $20/month on annual billing or $25 monthly; premium seat is $100 or $125. For 2 to 150 people. | The team-plan price is not verified. |
OpenAI and Anthropic do not publish exact message counts on their pricing pages. Google describes compute-based limits that refresh every 5 hours, subject to a weekly limit. The matrix leaves unverified items blank rather than treating them as zero.