Is ChatGPT good enough to write a grant application?

Often, for short forms you'll check yourself. In our test it wrote strong prose but added plans and numbers the owner never gave, which you'd need to find and remove before you apply.

Who judged the drafts?

AI models from OpenAI, Google and xAI, scoring against each funder's own rules without knowing which draft was which. No human grant reviewers, and no Claude models.

Did WinAGrant win every grant?

No. ChatGPT won the Freed Fellowship. Ours won Shophand clearly and the Northeast SARE grant narrowly.

Can I see the method?

Yes. The grants, how both sides' drafts were made and judged, and the vote count for each grant are on this page, dated October 29, 2026.

WinAGrant vs ChatGPT

In a blind test, independent judges picked our grant applications 8 to 3.

Same grants, same owner answers, blind judges. Here's what happened.

ChatGPT can draft a grant application in minutes. So we tested OpenAI's top model against our writer on three real grants worth $500 to $50,000, with judges who couldn't see which draft was which. We publish the losses too.

8 : 3
ours : ChatGPT
Independent judges
Google + xAI
9 : 8
ours : ChatGPT
All 17 verdicts
1 : 5
ours : ChatGPT
OpenAI judges
same model as ChatGPT

Write my application for $25 or see the grant money you could win first

The result (October 29, 2026)

Across 17 blind verdicts, ours won 9 and ChatGPT won 8.

All judges

Ours 9ChatGPT 8

Independent judges (Google Gemini, xAI Grok)

Ours 8ChatGPT 3

OpenAI judges (same model as ChatGPT)

Ours 1ChatGPT 5

Judges from Google (Gemini) and xAI (Grok), whose models wrote neither draft, preferred ours 8 to 3.
OpenAI's judges, the same model that wrote the ChatGPT drafts, preferred ChatGPT 5 to 1. Research has found that AI judges tend to favor writing from their own model family (Wataoka et al., 2024, arXiv 2410.21819).
One more Grok verdict was lost to a network error, so it isn't counted.

Freed Fellowship Grant

$500, and Fellows are eligible for a $2,500 year-end grant

ours : ChatGPT 1 : 5

Shophand Small Business Boost Grant

$2,500 cash plus $2,500 in services

ours : ChatGPT 5 : 1

Northeast SARE Farmer Grant

up to $50,000

ours : ChatGPT 3 : 2

Blind verdicts per grant, ours : ChatGPT
Grant	What it pays	ours : ChatGPT
Freed Fellowship Grant	$500, and Fellows are eligible for a $2,500 year-end grant	1 : 5
Shophand Small Business Boost Grant	$2,500 cash plus $2,500 in services	5 : 1
Northeast SARE Farmer Grant	up to $50,000	3 : 2

The bigger difference: what each draft promised for you

All three judge families, OpenAI's included, found more promises the owner never made in ChatGPT's drafts than in ours.

Example (Northeast SARE Farmer Grant, up to $50,000)

The farmer told both tools:

"I haven't done the budget spreadsheet. I think the total ask would be well under $50,000, maybe $20,000-30,000, but I haven't added it up."

ChatGPT

"a provisional $23,460 planning budget," (a number the owner never gave)

ChatGPT wrote "a provisional $23,460 planning budget," with its own hourly rates and 24 paid hours at $125 for an advisor who hadn't discussed her time yet. It labeled the numbers provisional, but the farmer never gave them, and it entered $23,460.00 as the total request.

WinAGrant

Ours wrote:

"I have not completed the budget spreadsheet yet. My working estimate is that the total request will be well under the $50,000.00 cap, maybe $20,000 to $30,000, but I have not added it up."

It left the exact totals marked FILL IN for the farmer.

Example (Shophand)

ChatGPT added "Measure scheduling-related administrative time before the transition and again after adoption." The owner had said she never measured it.

Ours wasn't perfect either. Judges flagged a few small leaps in our drafts too, for example answering "Yes" to US residency because the business is in Ohio.

Where we lost, and why

1 : 5

ours : ChatGPT

Freed Fellowship

Freed Fellowship: 5 of 6 verdicts picked ChatGPT. All six flagged our "$19 application fee" as made up. The fee is real, but Freed's page loads the price by script, so the copy the judges got didn't show it.

Method

How we ran it

Grants: Three real grants from our catalog, picked and written down before any draft was made: one short form, one medium, one long proposal.
Businesses: Three made-up small businesses (a mobile dog groomer, a cleaning company, an organic farm) with notes written in advance, including things the owner hadn't decided yet.
Drafts: Both sides got the same answers from the owner. ChatGPT (OpenAI's top model, GPT-6 Astra) got the funder's guidelines pasted in, then one follow-up: "make it stronger and make sure it answers every question." Ours ran as a customer gets it, with the grant details from our catalog.
Judging: Judges saw two unlabeled drafts, in both orders, and scored them against the funder's own rules. No Claude judges, because our writer runs on Claude.
Limits: What this is and isn't: AI judges, not human grant reviewers. Three grants, one draft per side. A small test, not a prediction of what a funder will pick.
Raw data: Download the vote data (JSON)

When ChatGPT is enough

You already know the grant well and just want help with wording.
The application is one or two short questions and you'll check every line yourself.
You'll compare the draft against the funder's rules on your own and remove anything you didn't say.

If that's you, ChatGPT is a fine place to start.

What $25 gets you

A full application for one grant, written from a short interview about your business.
Built from the funder's actual rules and questions, in the funder's order.
Every draft is sent to Strayguard™, our in-house AI review. It flags claims it can't trace back to your answers, so you can confirm or cut them.
The writer is told to use only your facts and leaves a marked blank where one is missing.

Don't qualify for this grant? Email us within 14 days for a full refund (one per customer). If the grant wasn't real or had already closed when you bought, we refund the $25 too. No funding guarantee.

Write my application for $25

Or $97 for 10 applications.

How we got here

September 30, 2026: ChatGPT beat our first version, 8 of 9 judge votes. Our drafts dropped the owner's caveats, skipped form fields, and in one case changed a correct 50% match to a wrong number.
We fixed those, then re-ran the same grants and won 9 of 9. Same grants, so we didn't count that as proof.
October 1: three new grants, 8 of 9 for ours. Both of those wins had Claude judges, which could favor their own family.
This round removed Claude from the judging entirely.

Questions

Is ChatGPT good enough to write a grant application?: Often, for short forms you'll check yourself. In our test it wrote strong prose but added plans and numbers the owner never gave, which you'd need to find and remove before you apply.
Who judged the drafts?: AI models from OpenAI, Google and xAI, scoring against each funder's own rules without knowing which draft was which. No human grant reviewers, and no Claude models.
Did WinAGrant win every grant?: No. ChatGPT won the Freed Fellowship. Ours won Shophand clearly and the Northeast SARE grant narrowly.
Can I see the method?: Yes. The grants, how both sides' drafts were made and judged, and the vote count for each grant are on this page, dated October 29, 2026.

Refund details are in Section 3 of our Terms.

See the grant money you could win, then we'll write the application.

Find my grants