AI humanizer for grant writing: funder rules, monthly pools and what the tools keep
Eight tools, ordered by words a month on the cheapest plan, largest first. A full federal package, narrative, aims, budget justification and biosketches, commonly runs 10,000 to 15,000 words, so the pool column is read against that.
What do NIH and NSF say about AI in applications?
Two notices, one restrictive and one procedural, and both read on 9 November 2026.
| Funder and notice | Dated | What it states (funder's wording) | What that means for a humanizer |
|---|---|---|---|
| NIH, NOT-OD-25-132, Supporting Fairness and Originality in NIH Research Applications | Released 17 July 2025; effective for the 25 September 2025 receipt date onward | "NIH will not consider applications that are either substantially developed by AI, or contain sections substantially developed by AI, to be original ideas of applicants"; six applications per principal investigator per calendar year; post-award detection may be referred to the Office of Research Integrity | The ideas and the text must be the applicant's; a humanizer pass on an AI-drafted section does not change what the section is |
| NSF, Notice to the Research Community on generative AI in merit review | 14 December 2023; last updated 21 May 2026 | Proposers "are encouraged to indicate in the project description the extent to which, if any, generative AI technology was used and how"; reviewers are prohibited from uploading proposal content to non-approved AI tools; information uploaded to tools outside NSF's firewall "is considered to be entering the public domain" | Disclose the use; and treat a humanizer's paste box as a public place for the proposal's text |
NIH adds that it "will continue to employ the latest technology in detection of AI-generated content" and lists enforcement actions from disallowing costs to termination (NIH, NOT-OD-25-132, read 9 November 2026). NSF frames its concern as confidentiality and authorship rather than detection, and states that the source and accuracy of AI output "is not always clear, which can lead to research integrity concerns including the authenticity of authorship" (NSF, Notice to the Research Community on AI, read 9 November 2026).
The two notices together draw the line this page follows. A humanizer used to adjust the register of a proposal the organisation wrote is a writing tool, disclosed where the funder asks. A humanizer used to disguise a section a model wrote is the thing NIH's notice is about, and no rewrite changes that. Private foundations publish their own rules, and many have none yet; the honest default is NSF's: say what the tool did. The wording of such a statement is on the how to disclose AI use guide.
What does a grant package do to a request cap and a privacy page?
A proposal is long, structured and confidential, which stresses the two fields vendors print least prominently.
| Document in the package | Typical length | Pieces on a 750-word cap | Pieces on a 1,000-word cap | What a humanizer can change that a reviewer will notice |
|---|---|---|---|---|
| Specific aims or summary | 500 to 1,000 words | 1 to 2 | 1 | A hedged aim becoming a promised outcome |
| Narrative or research strategy | 6,000 to 12,000 words | 8 to 16 | 6 to 12 | A defined term drifting to a synonym between sections; a cited figure rounded |
| Budget justification | 1,000 to 3,000 words | 2 to 4 | 1 to 3 | A line item's rationale generalised until it no longer matches the budget table |
| Biosketch or organisational history | 500 to 1,500 words | 1 to 2 | 1 to 2 | A date or a role changed; a named program renamed |
Every piece is a join, and a reviewer reading for consistency will find a term that changed halfway through faster than a detector will find anything. The confidentiality side is simpler: NSF's notice treats text uploaded to open AI tools as public, and the privacy pages above show which vendors say nothing about training, which offer a control, and which state there is none. For a proposal under embargo or with partner data in it, the practical rule is to strip names, figures and identifying detail before pasting, or to rewrite those sentences by hand and run only the connective text through the tool.
StealthGPT's Pro plan has the largest monthly pool of the 8 tools on this page, 50,000 words at $29.99 monthly, read 9 November 2026. Walter Writes AI's Starter has the largest under $10 a month, 30,000 words at $8 billed yearly. Both numbers come second to a funder's rule. An AI humanizer for grant writing rewrites the register of a proposal, and the two largest US research funders have each published what they expect of AI in an application. This page sets those two policies out, then orders 8 tools by the monthly pool on the cheapest plan, with the request cap and the privacy page's statement on pasted text beside each. Long proposals in pieces are covered on the long documents page, and the other audiences on the best-for hub.
Updated
Short answer
StealthGPT's Pro plan has the largest monthly pool of the 8 tools on this page, 50,000 words at $29.99 monthly, read 9 November 2026, and Walter Writes AI's Starter the largest under $10, 30,000 words at $8 a month billed yearly. Those numbers come second. NIH notice NOT-OD-25-132, released 17 July 2025 and read 9 November 2026, states that NIH will not consider applications substantially developed by AI to be the applicant's original ideas and caps a principal investigator at six applications a year; NSF's notice, updated 21 May 2026, encourages proposers to state how generative AI was used. An AI humanizer for grant writing is therefore a tool for a nonprofit's own drafts, used within the funder's rule and disclosed where asked, and this page orders 8 tools by the monthly pool on the cheapest plan with the request cap and privacy statement beside it.
Which AI humanizer for grant writing fits a proposal cycle?
Grammarly sits last because its pool cannot be stated in words, not because of price. Privacy wording is the vendor's own as read on 9 November 2026, set out with links on the AI humanizer privacy guide. Price is reported and never scored; detector readings and meaning-kept ratings under the published method join every row when the first run publishes.
| Order | Tool | Best for | Main limitation | Free tier | Cheapest plan (checked 9 November 2026 unless dated) | Request cap | Privacy page on pasted text (read 9 November 2026) | Review |
|---|---|---|---|---|---|---|---|---|
| 01 |  StealthGPT | A large pool for a team drafting several proposals a month | 1,000 words a request splits every narrative; no free tier; $1.00 trial rolls into the plan | None; $1.00 for 3 days | Pro, 50,000 words a month, $29.99 monthly or $299.99 a year | 1,000 | Kept no longer than three months past account termination; no training statement | Review of StealthGPT |
| 02 |  Walter Writes AI | A 30,000-word month under $10 with a strength setting, and a team plan on the page | 750 words a request on Starter, the smallest paid cap here after Grubby AI and WriteHuman | 300-word trial, no card (checked 8 November 2026) | Starter, 30,000 words a month, $8 a month billed $96 yearly | 750 | Processes text on the vendor's own self-hosted models; no training statement | Review of Walter Writes AI |
| 03 |  GPTHuman | One full package a month in 1,000-word pieces | Training is on by default until the opt-out is set; trial is 300 words once | 300 words once | Starter, 25,000 words a month, $8.25 a month billed $99 yearly or $15.00 monthly | 1,000 | May use content to improve services; opt-out in account settings | Review of GPTHuman |
| 04 |  Netus AI | A 15,000-word month in one of 36 named languages | Cap stated in characters, 18,000; credits shared with a paraphraser and summarizer | 50 credits a month, about 500 words | Basic, 15,000 words a month, $8.08 a month billed $97 yearly or $14 monthly | 18,000 characters | Not read for this page | Review of Netus AI |
| 05 |  Grubby AI | Short sections under 500 words | 10,000 words a month and a 500-word cap; the $8.99 rate is a $107.88 yearly charge | 300 words a month | Essential, 10,000 words a month, $8.99 a month billed $107.88 yearly or $19.99 monthly | 500 | Not read for this page | Review of Grubby AI |
| 06 |  Undetectable AI | A low entry price with no stated request cap | 10,000 words a month is less than one full package; the $5.00 rate is a $60.00 charge | 250 words in total (checked 8 November 2026) | 10,000 words a month, $5.00 a month billed $60.00 yearly or $9.99 monthly | Not stated | General retention clause on a policy dated 18 March 2023; no training statement | Review of Undetectable AI |
| 07 |  WriteHuman | A fixed number of 600-word sections a month, through the API for confidential text | A request is a request; web submissions train the model with no opt-out | 3 requests a month, 250 words each | Basic, 80 requests a month, $12 a month billed $144 yearly or $20 monthly | 600 | Trains on web submissions, no opt-out; API text not used to train | Review of WriteHuman AI |
| 08 |  Grammarly AI Humanizer | A house voice built from the organisation's own writing, free | No word allowance published; plans are metered in AI prompts; training on by default with a control | Free tool and free account, 100 AI prompts a month | Pro, $12, billing period not stated on the card, 2,000 AI prompts a month | Not stated | Trains AI models on collected information; control in account settings | Review of Grammarly AI Humanizer |
Which humanizer should a grant writer choose?
One line per job. Each name links to its full review.
- Choose StealthGPT if the organisation drafts several proposals a month and can work in 1,000-word sections.
- Choose Walter Writes AI if one or two packages a month fit 30,000 words and a strength setting matters more than request size.
- Choose GPTHuman if one package a month is the volume and you will set the training opt-out first.
- Choose Netus AI if the proposal is in one of its 36 languages and you will convert characters to words.
- Choose Grubby AI if you rewrite short sections and will pay a year up front.
- Choose Undetectable AI if the month's volume is under 10,000 words and $60.00 up front is acceptable.
- Choose WriteHuman only through its API, whose text the vendor states is not used for training.
- Choose Grammarly if a voice built from the organisation's past proposals is the point and you will set the training control before pasting.
How the ranking works
- Inputs
- 24 texts24 AI-written inputs across six genres plus 6 human-written controls, the same set for every tool.
- Detectors
- 5 namedGPTZero, Originality.ai, Copyleaks, Pangram and ZeroGPT.
- Weights
- Meaning kept35
- Errors introduced20
- Detector pass rate30
- Limits and access15
- Run date
The eight tools on the same fields
Every free tier gets the same fields, in rank order. The disc is the sample number it was poured under in the full run.
- Best for
- a large pool for several proposals a month.
- Main limitation
- 1,000 words a request, no free tier, and a $1.00 trial that rolls into the plan.
- Free tier
- none; $1.00 for 3 days.
- Cheapest plan
- Pro, 50,000 words a month, $29.99 monthly or $299.99 a year, read 9 November 2026 (StealthGPT, pricing).
- Request cap
- 1,000.
- Privacy page on pasted text
- retention no longer than three months past account termination; no training statement.
Sample 02: Walter Writes AI
Vendor site: Walter Writes AI.
- Best for
- a 30,000-word month under $10 with Simple, Standard and Enhanced modes.
- Main limitation
- 750 words a request on Starter, so a 6,000-word narrative is eight pieces; 1,500 needs Pro at $13 a month billed yearly.
- Free tier
- 300-word trial, no card, checked on 8 November 2026.
- Cheapest plan
- Starter, 30,000 words a month, $8 a month billed $96 yearly, checked on 8 November 2026; a Teams plan for 10 members at $99 a month billed yearly is priced on the same page, read 9 November 2026.
- Request cap
- 750.
- Privacy page on pasted text
- processes text on its own self-hosted models; no training statement.
- Best for
- one full package a month in 1,000-word pieces.
- Main limitation
- training is the default until the opt-out is set, and the trial is 300 words once.
- Free tier
- 300 words once.
- Cheapest plan
- Starter, 25,000 words a month, $8.25 a month billed $99 yearly or $15.00 monthly, read 9 November 2026 (GPTHuman, pricing).
- Request cap
- 1,000.
- Privacy page on pasted text
- may use content to improve services; opt-out in settings.
- Best for
- a 15,000-word month in one of 36 named languages.
- Main limitation
- an 18,000-character cap with no word equivalent, and credits shared across several tools.
- Free tier
- 50 credits a month, about 500 words.
- Cheapest plan
- Basic, 15,000 words a month, $8.08 a month billed $97 yearly or $14 monthly, read 9 November 2026.
- Request cap
- 18,000 characters.
- Privacy page on pasted text
- not read for this page.
- Best for
- short sections under 500 words.
- Main limitation
- 10,000 words a month and a 500-word cap, with the advertised rate a yearly charge.
- Free tier
- 300 words a month.
- Cheapest plan
- Essential, 10,000 words a month, $8.99 a month billed $107.88 yearly or $19.99 monthly, read 9 November 2026.
- Request cap
- 500.
- Privacy page on pasted text
- not read for this page.
- Best for
- a low entry price with no stated request cap.
- Main limitation
- 10,000 words a month is under one full package, and the $5.00 rate is a $60.00 charge.
- Free tier
- 250 words in total, checked on 8 November 2026.
- Cheapest plan
- 10,000 words a month, $5.00 a month billed $60.00 yearly or $9.99 monthly, read 9 November 2026.
- Request cap
- not stated.
- Privacy page on pasted text
- general retention clause on a policy dated 18 March 2023; no training statement.
- Best for
- a fixed number of 600-word sections a month, through the API for anything confidential.
- Main limitation
- a 100-word fix costs a full request, and web submissions train the model with no opt-out.
- Free tier
- 3 requests a month, 250 words each.
- Cheapest plan
- Basic, 80 requests a month, $12 a month billed $144 yearly or $20 monthly, read 9 November 2026.
- Request cap
- 600.
- Privacy page on pasted text
- trains on web submissions with no opt-out; API text not used to train.
- Best for
- a voice built from the organisation's own past proposals, free.
- Main limitation
- no word allowance published, plans metered in AI prompts, and training on by default with a control.
- Free tier
- free tool on the product page and a free account with 100 AI prompts a month.
- Cheapest plan
- Pro, $12, billing period not stated on the card, 2,000 AI prompts a month, read 9 November 2026.
- Request cap
- not stated.
- Privacy page on pasted text
- trains AI models on collected information; control in account settings.
Why do lists of humanizers for grant writers disagree?
Most are lists of humanizers for students with the word "grant" added, ranked on detector readings that no funder's notice mentions. This page starts from the two notices, orders by the pool because a proposal is long, and prints the cap and the privacy statement because a proposal is confidential. Price is reported and not ranked; the budget arithmetic across every plan is on the budget humanizer page. No detector score should be the only evidence in a misconduct case, and no humanizer pass makes an AI-drafted section the applicant's original idea.
Questions about AI humanizers for nonprofits and grants
Can I use an AI humanizer on a grant proposal?
Only within the funder's rule: NIH's notice NOT-OD-25-132, released 17 July 2025, states it will not consider applications substantially developed by AI to be original and may refer post-award detection for a misconduct review, while NSF encourages proposers to indicate how generative AI was used, both read 9 November 2026.
Does NIH detect AI in grant applications?
NIH states in NOT-OD-25-132 that it will continue to employ the latest technology in detection of AI-generated content, read 9 November 2026, and names enforcement actions from disallowing costs to termination if AI is identified after an award.
Does NSF ban AI in proposals?
No: NSF's notice, dated 14 December 2023 and updated 21 May 2026, prohibits reviewers from uploading proposal content to non-approved AI tools and encourages proposers to state in the project description the extent to which generative AI was used, read 9 November 2026.
Which humanizer fits a full grant package in one month?
StealthGPT's Pro plan allows 50,000 words a month and Walter Writes AI's Starter 30,000, read 9 November 2026, so both take a 15,000-word package twice over; GPTHuman's Starter takes it once at 25,000, and Grubby AI's and Undetectable AI's 10,000-word entry plans do not.
Is it safe to paste a confidential proposal into a humanizer?
Not without reading the privacy page: WriteHuman states web submissions train its models with no opt-out, GPTHuman and Grammarly state training with a control, and the other 5 tools here say nothing about training, read 9 November 2026; NSF's notice treats text uploaded to open AI tools as entering the public domain.
Should a grant writer disclose humanizer use?
Where the funder asks, yes: NSF encourages a statement in the project description, and NIH's policy turns on whether the application is the applicant's original idea, so a disclosure that a tool edited register rather than content is the honest form, read 9 November 2026.
What is the cheapest plan on this page for a small nonprofit?
Price is reported and not ranked here: Undetectable AI's entry plan is $5.00 a month billed $60.00 yearly for 10,000 words, Walter Writes AI's is $8 billed $96 yearly for 30,000, and Grammarly's humanizer is free with a plan metered in AI prompts, all read 9 November 2026.
Does this page score the tools?
No: it lists the same fields for every tool from the vendors' own pages, and detector readings and meaning-kept ratings join each row when the published method's first run is complete.
Sources
- NIH, NOT-OD-25-132, read 9 November 2026: Applications substantially developed by AI not treated as original; six-application cap; detection and enforcement
- NSF, Notice to the Research Community on AI, read 9 November 2026: Disclosure encouraged; reviewer prohibition; uploaded text treated as public; updated 21 May 2026
- StealthGPT, pricing, read 9 November 2026: Pro at 50,000 words, 1,000 per use, $29.99 monthly or $299.99 a year
- GPTHuman, pricing, read 9 November 2026: Starter at 25,000 words, 1,000 per output, $8.25 billed $99 yearly or $15.00 monthly
- Vendor pricing and privacy pages, read 8 and 9 November 2026: Plans, caps and privacy wording for the other six tools, cited in full on their reviews
Revision log
| Page published. |
Related reading
AI humanizer for product descriptions: catalogue volume, specs keptAI humanizer for product descriptions, ordered by the route each vendor states for a whole catalogue: a.
Originality AI humanizer: what Originality.ai publishes and the tools listed on one set of fieldsOriginality AI humanizer, answered with what Originality.ai publishes about its models, accuracy and false.
AI humanizer for long documentsAI humanizer for long documents, ordered by the per-request word cap on each vendor's cheapest plan, with the.
Best AI humanizer by useBest AI humanizer by use: academic writing, work checked by Turnitin, content writing and more, each page.
Is it safe to use an AI humanizer? What each vendor's privacy page says about your textIs it safe to use AI humanizer tools? What 12 vendors' privacy pages say about keeping your text and training.