GrantPanel's team of more than a dozen purpose-built AI agents runs a mock review panel on your funder's own criteria, and hands back a recommended revision for every weakness they find, with how to fix it. The result: better odds, before you submit.
Compliance & solicitation fit Literature & prior funded work Technical check of claims & math Three reviewers, three rounds Purpose-built agents, real tools The right model for each task Deleted after review · never trained on 3 runs: start with your draft
Works with any solicitation: NSF ·NIH ·DOE ·DOW ·NASA ·USDA ·DOT ·foundations & beyond
Sample review
CAREER: Learning Sediment Transport from Low-Cost Coastal Sensor NetworksNSF 22-586
Sample panel review
1/1700:00
Progress Status
Step 1 is yours. Everything after it is GrantPanel.
PI · 01
PI (user) uploads
Required
Proposal: the project description, plus any other parts, in one PDFDrop a PDF here, or browse
Solicitation: the agency's call and its review criteriaDrop a PDF here, or browse
Optional
+
Supporting materialsRelated articles, prior reviews, other related materials
Name this run It labels the run in your history, on the report, and in the email about it.
e.g. NSF DMREF: autonomous catalyst discovery
The proposal and solicitation are all that is required. The panel works from your documents alone.
North Carolina State UniversityOklahoma State UniversityTennessee Tech UniversityTexas A&M UniversityUniversity of Alabama in HuntsvilleUniversity of Houston-Clear LakeUniversity of Illinois ChicagoVirginia TechNorth Carolina State UniversityOklahoma State UniversityTennessee Tech UniversityTexas A&M UniversityUniversity of Alabama in HuntsvilleUniversity of Houston-Clear LakeUniversity of Illinois ChicagoVirginia Tech
Institution names shown for identification only, not as an endorsement.
IN ONE MINUTE
GrantPanel in one minute.
IN THEIR WORDS
What it's like to face our panel.
Researchers who put their proposals in front of the panel, in their own words.
“I was skeptical, so I tested GrantPanel on the original version of my NSF CAREER proposal, before it was funded. It came back Competitive, with every reviewer at Very Good or Good, and the feedback was mostly spot on. It even flagged the weakness the real NSF panel raised: inadequate measurement of demonstrated computational thinking learning. I fixed that, and the proposal was funded.”
Tugce AldemirAssistant Professor · Teaching, Learning and CultureTexas A&M University
“I read a lot of reviews as an editor, and these were substantive: specific, tied to the text, and willing to disagree with each other. When I ran the revised proposal again, the panel showed what had actually improved. It’s a fast, private way to pressure-test a proposal before the real panel does.”
Deb MishraAssociate Professor · Civil and Environmental EngineeringVirginia Tech
“I study how independent agents reach consensus, so I was curious how GrantPanel’s panel would do it. The reviewers wrote their assessments independently, then argued, and the final summary kept the dissent instead of averaging it away. That dissent told me more about my proposal’s risks than a single score would have.”
Ceyhun EksinAssociate Professor · Industrial and Systems EngineeringTexas A&M University
“I ran my proposal on learning-based control through GrantPanel before submitting. Watching three reviewers argue over it for three rounds showed me which weaknesses would actually move a panel and which were noise. The recommended revisions pointed to specific pages, so I knew exactly where to spend my time.”
Avimanyu SahooAssociate Professor · Electrical and Computer EngineeringUniversity of Alabama in Huntsville
“What won me over was the evidence. Every strength and weakness pointed to a section and page, so I could check the panel’s reasoning instead of taking it on faith. It caught gaps in my evaluation plan that I had stopped seeing after the tenth read. And my proposal was deleted when the review ended, which matters for unpublished work.”
John HuAssociate Professor · Electrical and Computer EngineeringOklahoma State University
“Early in a career, you rarely get to see how a panel will read your work before it counts. GrantPanel checked my proposal against the solicitation’s own criteria, not a generic checklist, and the revision list was specific enough to act on the same week.”
Hamid NazaripouyaAssistant Professor · Electrical and Computer EngineeringOklahoma State University
“Within just a few minutes of uploading my proposal, I had three independent reviews, a panel discussion, and a prioritized list of revisions. The compliance check flagged missing supplementary documents. For an interdisciplinary hardware proposal, seeing the solicitation’s criteria mapped section by section was most useful.”
Muhammad Faeyz KarimAssociate Professor · Engineering Technology and Industrial DistributionTexas A&M University
PRIVACY & TRUST
Your proposal is never kept, and never used to train AI models.
Your proposal is held in encrypted storage only while the review runs, and deleted automatically the moment it finishes. Only the panel's written report remains, in your account, so you can revise against it, and you can delete it whenever you like. The solicitation is kept too, for your proposal's 60-day revision window, so revision runs need not upload it again. Your unpublished work stays yours.
Never used to train AI
Hover for detailsTap for details
Never used to train AI
Your proposal and solicitation are sent to our model providers, Anthropic and OpenAI, only to generate your review. They are not used to train any model, by us or by the providers.
Tap to close
Auto-deleted after your review
Hover for detailsTap for details
Auto-deleted after your review
Your proposal and supporting files are deleted automatically as soon as your review finishes. Your report stays in your account until you delete it.
Tap to close
A practice tool, not an agency product
Hover for detailsTap for details
A practice tool, not an agency product
This is a simulation for your own preparation. It is not affiliated with, endorsed by, or a guarantee of any outcome from any funding agency or foundation.
Tap to close
MODEL TRANSPARENCY
Why not just ask a chatbot? One AI model is not a panel.
Ask a chatbot like ChatGPT or Claude and you get the opinion of one AI model, made by one company. GrantPanel convenes a mock review panel instead: a team of more than a dozen purpose-built AI agents, working independently, then together.The agents use AI models from two companies, OpenAI (the maker of ChatGPT) and Anthropic (the maker of Claude), each on the task it does best. Other tools quietly downgrade to cheaper AI models to cut costs; we never downgrade to save money. A single AI model, however well prompted, tends to agree with the framing it is handed, so the panel is built to argue. The full answer in the FAQ.
The modelsThe AI models behind both ChatGPT and Claude. No hidden downgrades.
A chatbot: one of them, never both.
Hover for detailsTap for details
The AI models behind both ChatGPT and Claude. No hidden downgrades.
Your review never rests on one AI model or one company. GrantPanel uses AI models from both OpenAI, which makes ChatGPT, and Anthropic, which makes Claude, and the two catch different kinds of errors. We chose the AI model for every stage by running both through the full panel on real proposals whose outcomes we know, and keeping whichever caught more of the weaknesses already documented in them. A newer model goes through the same test before it replaces one. A model is never swapped down to protect our margin.
Tap to close
01
The agentsMore than a dozen agents, each equipped for its job
A chatbot: one generalist, with the tools it ships with.
Hover for detailsTap for details
More than a dozen agents, each equipped for its job
Each agent gets the tools and data its job needs, not a chatbot’s defaults: your proposal and the call read natively as PDFs, the solicitation’s own rules and rubric, Semantic Scholar, OpenAlex and Crossref to confirm that every cited work exists, read its abstract and mark any retraction or preprint, live web search with Europe PMC and arXiv for the state-of-the-art scan, NSF Award Search, NIH RePORTER and USAspending.gov for the work already funded in your area, and for the technical analysis every page read as an image and a code sandbox to run your arithmetic. Checks in our own code then pin every passage the recommended revisions quote to your text and inspect the panel’s summary before you see it.
Tap to close
02
The panelIndependent reviews, then an argument
A chatbot: one reading, which tends to agree with your framing.
Hover for detailsTap for details
Independent reviews, then an argument
Three reviewers each read your proposal in their own context and file a complete review before any other position exists, so none takes its cue from the others. Then they argue across three rounds: each pushes back or concedes with evidence, and a turn that only repeats its review is sent back. The chair writes the consensus from the reviews and the discussion, and a disagreement that survives round three is reported as a disagreement, not averaged away.
Tap to close
03
The approachEngineered process, not fine-tuned weights
A chatbot: a prompt you write, and rewrite, yourself.
Hover for detailsTap for details
Engineered process, not fine-tuned weights
Solicitations change every cycle, and so do the models best suited to reading them closely. Fine-tuning would freeze today’s program knowledge into weights we could not inspect and you could not audit — and it would make your proposal training material. Instead the review process itself is the engineering: a team of more than a dozen purpose-built agents, each calibrated to its role and to your agency’s own criteria, working independently, then together, on whichever models are strongest at the time. The rulebook stays readable, the method stays current, and your work is never trained on.
Tap to close
04
The peopleShaped by the people it is built for
A chatbot: general-purpose, not shaped by grant reviewers.
Hover for detailsTap for details
Shaped by the people it is built for
GrantPanel was developed with, and repeatedly revised on the advice of, former federal program managers, experienced panel reviewers, and faculty researchers at several universities. They ran their own proposals through it and told us what a real panel would have done differently. Who built it has more about them.
Tap to close
05
OpenAI (ChatGPT) and Anthropic (Claude) are referenced only to identify compatible AI providers and products. GrantPanel is independent and is not affiliated with, sponsored by, or endorsed by either company. No third-party logos are used.
WHO BUILT IT
Built by people who've been on every side.
GrantPanel's multi-agent review process is designed by faculty researchers, panel-side reviewers, and former program managers: people who've written, won, reviewed, and funded grants.
THE APPLICANT SIDEFaculty researchers01Hover for detailsTap for details
Faculty researchers
Experience turning complex research ideas into competitive, compliant proposals under real submission constraints.
Proposal strategy and positioning
Evidence, aims, and narrative coherence
Revision under deadline pressure
Tap to close
THE REVIEW SIDEPanel-side reviewers02Hover for detailsTap for details
Panel-side reviewers
Experience reviewing proposals, interpreting agency criteria, and understanding how panel judgments are formed.
Agency review criteria
Panel deliberation and consensus
Decision-relevant weaknesses
Tap to close
THE AGENCY SIDEFormer program managers03Hover for detailsTap for details
Former program managers
Experience directing funding programs: shaping solicitations, guiding panels, and deciding what a portfolio funds.
How solicitations are written
What panels are asked to prioritize
What separates funded from declined
Tap to close
ONE REVIEW SYSTEMWe know what reviewers look for because we've sat on every side of the process.
Built by researchers and former program managers who wanted a second opinion before submitting, not a startup selling AI hype. We built the tool we wished existed for our own proposals.
WHY REHEARSE
Funding favors the rehearsed.
Everything here exists for one blunt reason: better odds.Panels fund a fraction of what they read, and proposals are usually declined for avoidable reasons: an unsupported claim, a vague evaluation plan, a missed requirement. Some universities already convene mock panels for their flagship submissions. GrantPanel makes that rehearsal available for every proposal. And an AI panel's critique is worth having: independent research finds AI pre-submission review can match or beat strong human reviewers, for example on journal articles. See the evidence.
1 in 5proposals is fundedHover for detailsTap for details
1 in 5 proposals is funded
Roughly the odds at NSF; closer to 1 in 8 at NIH. The margin between funded and declined is rarely brilliance; it is usually a handful of fixable weaknesses.
Tap to close
6–12 mobefore you can act on itHover for detailsTap for details
6–12 mo before you can act on it
A real panel's critique arrives only after the deadline, and the next cycle may never come. Programs change, topics are not always re-announced, and an idea can be overtaken while you wait.
Tap to close
5–15 minto hear it earlyHover for detailsTap for details
5–15 min to hear it early
A mock panel delivers the same kind of critique before submission, while every weakness it finds is still yours to fix.
Tap to close
26 of 30blind picks for the AI reviewHover for detailsTap for details
26 of 30 blind picks for the AI review
Independent research finds AI pre-submission review can match or beat strong human reviewers, for example on journal articles. Former journal editors, not told that any review was AI-written, compared an AI review of a draft research paper with a paid expert's and chose the AI review in 26 of 30 judgments (Harvard working paper, October 2026). A 2026 study in a peer-reviewed cardiology journal points the same way.
Tap to close
Can we promise funding? No, and we never will. What a rehearsal changes is the part you control: whether the weaknesses in your proposal are found by you before submission, or by the panel that decides after it.
SAMPLE REPORTS
What you actually get.
Every sample is a fictitious proposal against a real call. Read one before you pay for one. A complete run, start to finish: compliance, alignment, a technical analysis, three independent reviews on the funder's own scale, the discussion transcript, the chaired consensus, how to fix each weakness, and the supporting research, readable in the browser or saved as HTML or PDF.
No subscriptions and no plans to compare. Every purchase runs the same full panel. Each proposal gets up to 3 runs: your initial review plus two re-reviews, so the first can be a mostly complete draft rather than the final version. Nothing counts down until you start: the first run opens a 60-day window, and runs 2 and 3 are yours at any point inside it.
Proposal review
Up to 3 panel runs in 60 days, so start with your draft.
$59$79per proposal
Launch price through March 31, 2027
First proposal guaranteed
First paid proposal not helpful? We refund it in full, tax included, if you ask within 14 days of its first run and before its second. No questions asked. Refund policy
Buying codes for a department, or paying by university card, purchase order or tax-exempt? How your university can pay →
A run is one full panel review. Buy whenever you like; nothing counts down until you start. A failed run is never charged. Prices are in US dollars; sales tax is added at checkout where it applies. See a full sample report →
Institutional
For a department or research office buying proposals for its faculty.
Data-flow summary, subprocessor list, and retention schedule on request
DIRECT ANSWERS
Common questions.
How is this different from pasting my draft into a chatbot like ChatGPT or Claude?
Fair question, and honestly the first one we asked ourselves. We know how much of yourself goes into a proposal. If a few prompts could give it the read it deserves, we would have shared them and gone back to our own research. We are researchers ourselves, and we built GrantPanel to help researchers.
Here is the difference. A chatbot gives you one model's reading, and a single model, however well prompted, tends to agree with the framing it is handed. GrantPanel convenes a mock review panel instead: a team of more than a dozen purpose-built AI agents, each with its own role, tools and data, working independently, then together, to judge your proposal on your funder's own criteria.
Not one model, and not one company's. A chatbot runs on a single company's model. GrantPanel uses different models from two companies, OpenAI (ChatGPT) and Anthropic (Claude), and the two catch different kinds of errors. Each stage runs on whichever model did it best when we tested both on real proposals whose outcomes we know.
GrantPanel was developed with, and repeatedly revised on the advice of, former federal program managers, experienced panel reviewers, and faculty researchers at several universities. They ran their own proposals through it and told us what a real panel would have done differently.
In short: The value is in the engineered process, not a single prompt, and in the people behind it. GrantPanel grew out of real departmental mock review panels, and its founder has written, won and reviewed federal proposals. You watch the panel work live, you get three runs per proposal, and your proposal is never trained on and is deleted when the review finishes. More than a dozen purpose-built AI agents, on models from both OpenAI and Anthropic, do the work: each run checks compliance, any budget you upload, every reference, the current literature and prior funded work, adds a technical analysis, has three independent reviewers argue over three rounds, and ends with a chair who keeps dissent and a recommended fix for every weakness.
Item by item:
A panel you can watch. Follow your run live, stage by stage: the literature searches as they go out, the three reviews as they are written, and the panel's argument round by round.
Three runs per proposal. Runs 2 and 3 read your revised draft against what the earlier run found, so you see which concerns you addressed and what moved.
Never trained on, and deleted. Unlike a personal chatbot account, which may use your conversations to train future models, your proposal is never used for training, by us or by our providers, and it is deleted the moment the review finishes.
More than a dozen purpose-built agents. Each agent has one job and the tools and data that job needs, not a chatbot's defaults: reference indexes for your citations, live literature search for the state of the field, and federal award databases for prior funded work.
Models from two companies, not one. Different models from OpenAI (ChatGPT) and Anthropic (Claude), which catch different kinds of errors, each stage on whichever did it best when we tested both on real proposals whose outcomes we know, and never downgraded to cut costs.
Your funder's own criteria and scale. Read from your solicitation and the agency's own review frame (NSF, NIH, DOE, DOW, NASA, USDA, DOT), with pre-proposals judged at their stage. Any other call is reviewed on its stated criteria, rated on GrantPanel's own 1-5 scale, labeled as ours.
Compliance and solicitation fit. Required sections, page and format limits, required sub-headings and attachments, the triggers for a return without review; then whether the proposal answers the call: its priorities, eligibility and required plans.
A budget check. When you upload a budget, every sum is computed in code and held to the call's ceiling, cost-sharing and indirect-cost rules, and to the work the narrative describes.
Every reference looked up. Each cited work is checked in Semantic Scholar, OpenAlex and Crossref, with retracted works and preprints marked.
A deep literature search. Your novelty claims are checked against the field as it stands, with a live search that includes Europe PMC and arXiv, not just what a model remembers.
Prior funded work. Related awards from NSF Award Search, NIH RePORTER and USAspending.gov, so the panel sees what is already funded and how yours differs.
A technical analysis. Before the panel meets, an expert reader in your field checks the key claims, the math, every figure and table, the methods and the feasibility.
Three independent reviewers. Each works in its own context and files a complete review before any other position exists, so none takes its cue from the others.
Three rounds of panel discussion. Positions, counter-points and final votes: reviewers push back or concede with evidence, and a turn that only repeats its review is sent back.
A chair who keeps dissent. The consensus says where the panel agreed and, under Points of disagreement, where it did not, instead of averaging the readings into one.
A recommended revision for every weakness. The steps, the passage as it stands, and suggested wording where the fix is wording, with every quoted passage checked against your text. GrantPanel never edits your document.
Findings you can trace. Every finding points to the page and passage it comes from, and the full deliberation transcript comes with the report.
Your whole package, read in full. Up to 200 pages across the proposal, the solicitation and supporting documents, each page as text and as an image, so figures, tables and signed letters are read too. An optional document you leave out is noted as not supplied, never counted as a weakness.
Exports for co-PIs and your research office. Save the report as PDF or HTML with the sections you choose, and a compliance checklist as PDF or CSV for Excel.
What happens to my proposal, and is it stored anywhere?
We are researchers ourselves, and we would not hand our own unpublished work to a tool that kept it. Your proposal and supporting files are held only while the review runs and are deleted automatically the moment it finishes. The solicitation is kept for your proposal's 60-day revision window, so revision runs need not upload it again. Beyond the solicitation, only the panel's written report is kept, in your account, so you can revise and run a revision review, and you can delete it at any time. Your files are never used to train models, by us or by our providers.
Do I need a finished proposal before I run it?
No, and waiting for one is the easiest way to lose the review's value. One proposal comes with three runs, so the first can be a mostly complete draft. The panel reviews it as the proposal you would submit, so its findings are the ones a real panel would raise, and you still have time to act on them. Fix what it finds, run it again, and keep the third run for the week before submission. The three runs share a 60-day window that opens on the first, so start about two months before your deadline.
What counts as a run, and how long do I have to use all three?
A run is one complete panel: compliance and solicitation alignment, a reference check, a deep literature search, prior funded work, a technical analysis, three independent reviewers, three rounds of deliberation, a chaired consensus, and recommended revisions. One proposal buys three of them. Runs 2 and 3 read your revised proposal against what the earlier run found, so you see what moved, which is why the first run can be a draft rather than the final version. Nothing counts down while a proposal sits unused, so you can buy well before you are ready. The first run opens a 60-day window, and the other two are yours at any point inside it. Sixty days is deliberately longer than most revision cycles: it is meant to cover reading the panel's findings, rewriting, and running again before your deadline, without becoming an allowance you bank indefinitely against a different submission.
Which agencies does it support?
Any program with a written solicitation. NSF, NIH, DOE, DOW, NASA, USDA, and DOT each get their own review frame: the funder's criteria, its rating scale, and its decision, such as NIH's 1-9 impact score, NASA's adjectives and relevance, DOE Early Career's 1-6 rating, NIFA's ranking, SCRI's relevance points, or a DOW BAA's select or do not select. Your call's own criteria and scale come first, except that NSF reviews always use NSF's own ratings. Foundations, institutional competitions, and any other call are reviewed on the criteria the call states, rated on GrantPanel's own 1-5 scale, labeled as ours.
Can my university or department pay?
Yes, with a purchasing card. Checkout runs on Stripe and takes a university p-card like any other card. Each purchase comes with an invoice and a receipt, with your department's name and address on them if you add them, for your records or for reimbursement.
For a whole department. A department, centre or research office can buy a block of proposal codes, at least ten, from its own GrantPanel account under Proposal codes, at less per proposal than buying them one at a time. It sends each colleague a code, from that page or by email from GrantPanel, and each colleague redeems it into their own account. The office sees whether and when each code was used, never the proposal or its report. Proposal codes are a prepaid purchase by your institution for its faculty. Each code must be used (its proposal's first run started) within one year of purchase. Until 30 days after that date, we refund any code that hasn't been used, at the price paid; after that, unused codes lapse.
By purchase order. Purchase orders are for proposal codes. If your department pays by purchase order, email hello@grantpanel.ai for a quote for a block of codes. We invoice your institution, and the codes are added to your account once it is paid.
Tax-exempt. Sales tax is charged only on purchases billed to a Texas address. If your institution is tax-exempt and pays with its own funds, email us its exemption certificate (Texas Form 01-339, or a university purchase voucher) before you buy: we mark your account tax-exempt, no tax is charged at checkout, and the invoice notes the exemption. Bought before sending it? Send it now and we refund the sales tax on that purchase. Paying by purchase order, send the certificate with the order and the invoice is issued without the tax.
Vendor setup. If your purchasing office needs to set GrantPanel up as a vendor, we send a W-9 and the details it asks for.
Can I try it before I pay?
There is no free run: every run is a full panel and costs the same to run either way. What you can do first is read a full sample report, in the same form yours will take, for a fictitious proposal. And your first paid proposal is covered: if its review doesn't help, tell us within 14 days of its first run, before you start its second, and we refund it in full, sales tax included.
Will GrantPanel write or edit my proposal?
No. It never edits your document. For each weakness, the report says how to fix it: the steps, the sentence as it stands, and suggested wording where the fix is wording. Where the fix is research content (your aims, methods, or results), it says what to write and the question to answer first, and leaves the writing to you. NIH, for one, does not consider applications substantially developed by AI to be the applicants' original ideas.
Can I use it on a proposal I'm reviewing for an agency?
No. Upload only work you wrote, or that its authors have authorized you to submit. A proposal assigned to you for review is confidential, and NIH and NSF both bar reviewers from uploading proposals to AI tools. The same goes for a colleague's proposal without their permission.
Will it predict my score?
No, and the distinction matters, because you will see ratings in your report. Each reviewer scores your proposal on the scale your solicitation states, or else on the funder's own (NSF's adjectives, NIH's 1-9 impact score, DOE's and NASA's scales), or else on GrantPanel's own 1-5 scale, labeled as ours, and the chair records where the panel came out. Those are this panel's judgement of this reading of your draft, and they are there for a reason: they are how you tell whether a revision actually moved anything. What they are not is a forecast. A real panel turns on who is assigned your proposal, what else is in the pile that day, and how the discussion goes on the morning. GrantPanel is practice, not prophecy: read the rating as a signal about your draft, not an estimate of your odds.
Why is the review so strict?
On purpose. Before submission, a tougher read than you are likely to get is the more useful one, because every weakness it finds is one you can still fix. Your actual review may be milder, and its outcome depends on the reviewers assigned, the other proposals in the panel, and the funds available.
Is there independent evidence that an AI review is worth reading?
Yes. Independent research finds AI pre-submission review can match or beat strong human reviewers, for example on journal articles. In an October 2026 Harvard working paper, researchers compared AI-written reviews of 124 draft research papers with reviews from paid expert reviewers. Three former journal editors, not told that any review was AI-written, preferred the AI review in 26 of 30 judgments.
Other 2026 work points the same way. In a peer-reviewed cardiology journal, two editors who did not know which reviews were which rated the AI reviews higher overall on 40 manuscripts. When the AAAI-26 conference added an AI review to each of nearly 23,000 submissions, more of the participants surveyed found it useful than not.
Those studies used a single AI review. GrantPanel gives your proposal a full panel: three independent reviewers who argue over three rounds, on your funder's own criteria, while you can still act on what it finds.
Can it review a white paper, concept paper, or pre-proposal?
Yes. The panel recognizes a preliminary submission and judges it at its stage: against the call's rules for that stage, with the stage's own decision, such as invite or not invite. Documents the call asks for only with a full proposal are not expected.
What do I upload?
PDFs. Your proposal and the solicitation are required; on a revision run the solicitation is already on file. The proposal is one PDF: only the project description is required, and the rest of the package goes in the same file if you have it, such as the summary, references, letters, biosketches, current and pending support, budget documents, and facilities and equipment. You can also add up to 10 supporting PDFs, such as related articles or reviews from an earlier submission. One run reads up to 200 pages in all, and the upload page keeps a tally as you go.
What if my program has no written solicitation, only a web page?
Save or print the program's web page as a PDF and upload it as the solicitation. If the page states review criteria, the panel uses them; if not, it falls back to the funder's general criteria, or else GrantPanel's own.
How long does a review take?
A full run (compliance and alignment, the reference, literature, and prior-funding checks, a technical analysis, three reviewers, three rounds of discussion, and the summary with its recommended revisions) usually takes about 5 to 15 minutes. Most of that time is the deep literature search: the panel checks your novelty and prior-work claims against the current literature instead of relying on what a model already remembers. We could return a report in seconds by skipping that step and save API credits and costs, but it would cost the technical depth and rigor that make the review worth reading.
Can I share the report with my co-PIs?
Yes. Export it as a PDF, or as an HTML file that opens in any browser, and send it to whoever you like. They don't need a GrantPanel account.
What does the technical analysis check?
Before the panel meets, a technical analyst, a senior researcher in your field, reads every page, figures and tables included, and checks your technical case: the claims, the math, and the arithmetic, which it runs rather than eyeballs. It does not rate the proposal. The reviewers read its findings, and the report prints them.
Which AI models run my review?
Models from both OpenAI (ChatGPT) and Anthropic (Claude), not one model with default settings. The two catch different kinds of errors, so each stage of the panel runs on whichever model did that stage best when we ran both providers through the full review on real proposals whose outcomes we know, scored against the weaknesses already documented in them. We follow both providers' releases, and a newer model goes through the same test before it replaces one. GrantPanel is independent and is not affiliated with, sponsored by, or endorsed by either company.
Why not use cheaper or open-weight models to bring the price down?
We tested them extensively, across the whole pipeline. Smaller and open-weight models handle the mechanical stages competently: spotting a missing section, checking a page limit, extracting citations. Where they consistently fall short is the part that makes a review worth reading. They summarize a proposal instead of interrogating it, agree with each other instead of holding distinct positions through three rounds of argument, miss the gap between what a claim asserts and what the cited evidence actually supports, and produce weaknesses too generic to act on. A cheaper review isn't a bargain if its findings are ones you could have written yourself. So your review runs on a mixture of models from OpenAI (ChatGPT) and Anthropic (Claude), each stage on the model that did it best when we tested both providers on real proposals, with frontier models for the reviews, the debate, and the chair's synthesis. None is ever downgraded to cut costs. We would rather charge honestly for that than quietly serve you a weaker panel.
Other tools charge less. Why does GrantPanel cost more?
Some do, and if what you want is a quick read-through of your draft, one of them may suit you. GrantPanel is doing a different amount of work. Every run is a full panel of more than a dozen purpose-built AI agents: compliance and solicitation alignment, a reference check, a deep literature search, prior funded work, a technical analysis, three independent reviewers, three rounds of deliberation, a chaired consensus, and recommended revisions. Each stage is its own model call, not a single pass over your document. The most expensive part is the deep literature search: it queries the current literature on every run rather than relying on what a model happens to remember, and that search is a real cost we pay each time. We also move to newer frontier models as they are released instead of settling on an older, cheaper generation to protect our margin. For a proposal that represents years of work, and at a fraction of what a human grant consultant charges for the same read: we would rather be worth the price than be the cheapest.
Why is there no subscription?
Because we are from academia and we know the work is not continuous. Proposal writing arrives in bursts around solicitations and deadlines, and in between you are teaching, running a lab, reviewing, and doing the research itself. A monthly fee would quietly bill you through every month you never open the tool. So you buy a proposal review when you have your proposal ready, an unused proposal review never expires, and nothing charges you for the semesters you do not need us.
What if GrantPanel doesn't help?
Your first paid proposal is covered. If its review doesn't help, email us within 14 days of its first run, before you start its second, and we refund it in full, sales tax included. After that, a run that fails is returned to you automatically, and a report that misread your proposal is refunded or credited (see the Refund Policy). For a report that fell short, write to hello@grantpanel.ai with the proposal and solicitation you submitted and the report you received: your files are deleted after each run, so we'll need you to send them again.
Can I upload patient data or controlled material?
No. A narrative describing a clinical study is fine; the data behind it is not. Do not upload identifiable human-subject data, patient records, controlled unclassified information, or ITAR, export-controlled, or classified material. We do not sign BAAs.
Can my university license GrantPanel?
Yes. Research-development teams can contact hello@grantpanel.ai for institutional access, privacy review, and purchasing.
Is GrantPanel affiliated with a funding agency?
No. GrantPanel is an independent preparation tool. It is not affiliated with, endorsed by, or connected to any funding agency or foundation.
Why are your reviewers called A, B and C?
Because nobody needs to see “Reviewer 2” before their Monday coffee. (All three read the whole proposal.)
Put your proposal in front of our panel first.
Upload your proposal and the agency's solicitation, and get a full simulated panel review in about 5–15 minutes.