Honest comparison

Bothy vs. Candid's AI Fundraising Assistant

In 2026, Candid quietly shipped an AI Fundraising Assistant inside its Premium and Ultimate plans. We paid for Premium and stress-tested it against something most reviews don't have: a nonprofit whose complete institutional funding history we had already verified from IRS filings. This is what we found — including the places Candid is genuinely strong.

Test conducted August 20, 2026 on a paid Candid Premium account · Product may have changed since · Reading time ~6 minutes

The TL;DR

Candid's data is real. Ask it about one named funder and the depth is genuinely strong — it surfaced a current grant amount that we then verified, correctly, against the funder's e-filed 990-PF.
Recall is the problem. Asked to find the funders of an org with 72 known institutional grantmakers, it returned 4 — and the four smallest. The org's largest relationships never appeared.
Zero cycle intelligence. No deadlines, no windows, no "this closed last month." For working fundraisers, timing is half the job.
The meter is aggressive. Each substantive answer consumed ~12–13% of the monthly credit allowance — about eight real questions a month on a $1,199/yr plan.
Generated facts drift. Its LOI draft invented a website URL for the org, cited years-stale program numbers, and told the org to "apply immediately" to a program that is invitation-only.

How we tested

Most software comparisons check features. We checked answers. Our ground truth was a Louisville pediatric-therapy nonprofit whose funding history we had built the slow way: its funders' e-filed 990-PFs read line by line, its own published donor honor roll, and a geographic sweep of the IRS Business Master File — 72 known institutional grantmakers, over $1.6M in verified grant lines. Then we asked Candid's assistant the questions a real development director would ask, and scored every answer against the filings.

That cuts both ways, and we report both: where the assistant was right, we say so — one of its findings was real, current, and material, and we verified it against the IRS e-file and acted on it. Where it was wrong, the errors weren't matters of opinion. They were checkable facts.

This was a single-org test on one paid account in August 2026. It is our test, not an audit; Candid's product may have improved since. Full test notes available on request.

Side by side

Based on our August 2026 test and each product's public materials. = does it well, = with caveats, = doesn't do it.

Capability	Candid AI Assistant	Bothy
Funder recall (find the org's existing institutional funders)	4 of 72 known grantmakers in our test — and the four smallest	Multi-engine research: filings reverse-index + the org's own donor honor roll + geographic IRS sweep, verified before it ships
Named-funder depth (ask about one specific funder)	Genuinely strong — surfaced a real, current grant amount we verified against the IRS e-file	Same facts, from the filings directly — with the source object ID cited
Deadline / cycle intelligence	None in our test — “not available in Candid's data”	Every dated entry web-verified against the funder's own site; stale dates are a release-blocking check
Eligibility screens	None — recommended an invitation-only program as “apply immediately”	Built-in: invitation-only flags, budget-cap screens, geography checks, grantmaker-vs-grantseeker detection
Accuracy of generated drafts	LOI draft invented the org's website URL; profile facts were years stale	No generated facts — every claim traces to a filing or the funder's own site, or it doesn't ship (23 automated release gates)
Renewal strategy per funder	LLM-generated at query time; polished prose, no sourcing	Computed from the funder's verified filing history — trajectory, restriction pattern, process type
Ask calibration (what does this funder actually give?)	Sometimes offered, unsourced	The funder's real grant distribution (median, quartiles, max, n) from its own filings — never an estimate
Usage limits	Credit-metered — each substantive answer cost ~12–13% of the monthly allowance in our test (~8 real questions/month)	No meter. Your workspace is yours
Underlying dataset	Excellent — tens of millions of grants, including public-charity grantmakers invisible to 990-PF-only tools	Public IRS filings + funder websites + the org's own published donor lists — smaller, but verified per org
What you get at the end	Chat answers you copy out before the meter resets	A permanent workspace: pipeline, calendar, stewardship tasks, reporting, board packet
Price	Assistant included in Premium ($1,199/yr nonprofit rate) and Ultimate ($1,699/yr)	Solo $99/mo; curated research included on Pro + Premium tiers
Tenure and brand	The sector standard for nonprofit data; decades of trust	Pilot 2026 — 69 nonprofits, early SaaS access

Funder recall (find the org's existing institutional funders)
Candid4 of 72 known grantmakers in our test — and the four smallestBothyMulti-engine research: filings reverse-index + the org's own donor honor roll + geographic IRS sweep, verified before it ships
Named-funder depth (ask about one specific funder)
CandidGenuinely strong — surfaced a real, current grant amount we verified against the IRS e-fileBothySame facts, from the filings directly — with the source object ID cited
Deadline / cycle intelligence
CandidNone in our test — “not available in Candid's data”BothyEvery dated entry web-verified against the funder's own site; stale dates are a release-blocking check
Eligibility screens
CandidNone — recommended an invitation-only program as “apply immediately”BothyBuilt-in: invitation-only flags, budget-cap screens, geography checks, grantmaker-vs-grantseeker detection
Accuracy of generated drafts
CandidLOI draft invented the org's website URL; profile facts were years staleBothyNo generated facts — every claim traces to a filing or the funder's own site, or it doesn't ship (23 automated release gates)
Renewal strategy per funder
CandidLLM-generated at query time; polished prose, no sourcingBothyComputed from the funder's verified filing history — trajectory, restriction pattern, process type
Ask calibration (what does this funder actually give?)
CandidSometimes offered, unsourcedBothyThe funder's real grant distribution (median, quartiles, max, n) from its own filings — never an estimate
Usage limits
CandidCredit-metered — each substantive answer cost ~12–13% of the monthly allowance in our test (~8 real questions/month)BothyNo meter. Your workspace is yours
Underlying dataset
CandidExcellent — tens of millions of grants, including public-charity grantmakers invisible to 990-PF-only toolsBothyPublic IRS filings + funder websites + the org's own published donor lists — smaller, but verified per org
What you get at the end
CandidChat answers you copy out before the meter resetsBothyA permanent workspace: pipeline, calendar, stewardship tasks, reporting, board packet
Price
CandidAssistant included in Premium ($1,199/yr nonprofit rate) and Ultimate ($1,699/yr)BothySolo $99/mo; curated research included on Pro + Premium tiers
Tenure and brand
CandidThe sector standard for nonprofit data; decades of trustBothyPilot 2026 — 69 nonprofits, early SaaS access

Where Candid is honestly better

Candid's dataset is the best in the sector, full stop. Tens of millions of grant records, updated as filings land, including Schedule I coverage of public-charity grantmakers that 990-PF-only research can't see. If your question is "tell me about this one specific funder," the assistant's depth is real — and Premium includes the whole Candid suite, not just the chat.

And Candid has decades of earned trust. We're a 2026 pilot with 69 nonprofits. If institutional tenure is what your board needs to see, they win that row without contest.

Where we think we're honestly better

The gap isn't the data — it's the method. Candid's assistant generates answers at query time with a language model over its database. Bothy does the research once, the slow way, verifies every claim against a filing or the funder's own site, and refuses to publish anything that fails one of 23 automated checks. Same category of feature, opposite epistemics:

· Recall by method, not retrieval — the org's own donor honor roll and a geographic sweep of the IRS Business Master File catch the funders a database query misses. In our test that was the difference between 4 funders and 72.
· Deadlines are first-class data — every dated entry is verified against the funder's site, and a stale date blocks release.
· Renewal strategy from filings, not vibes — our per-funder strategy blocks are computed from the funder's actual multi-year giving to you: trajectory, restriction pattern, whether an application process even exists.
· Ask calibration you can defend to a board — the funder's real grant distribution from its own filings, never a model's estimate.
· No meter — research lives in a permanent workspace with your pipeline, tasks, and reporting, not in chat answers you ration.

The honest bottom line

If you want the sector's biggest funder database and a research tool for occasional deep-dives on named funders, Candid Premium is a defensible buy — it was useful enough in our own test that we verified and acted on one of its findings. If you want to know who funds organizations like yours, when the windows open, and what to ask for — verified, deadline-aware, and yours to keep — that's the job Bothy was built for.

See the difference on your own funders

Bring us your funder list. We'll show you what verified looks like.

Send us the funders you already know about. We'll come back with what the filings actually say — including the relationships and deadlines you're missing.