Most grant writers I talk to use AI every day and are still not sure they are allowed to. They draft a paragraph in ChatGPT, clean it up so it sounds like them, and then feel a small pull of worry that a funder will somehow know. They are using the tool and hiding the tool at the same time.
Instrumentl's 2026 survey of 176 grant professionals puts numbers to that tension, and they're more interesting than a simple adoption story. Grant writers are misjudging AI. The data show that we underestimate how much we already rely on it, fear the wrong risks, and guard the wrong parts of the work. In each case, the line between what belongs to the machine and what belongs to the human is drawn in the wrong place.
Below, we’ll dig into the survey results and their implications for the state of AI in grants work. Then, I’ll share tangible ways to re-draw the lines around how and where AI shows up in your personal and organizational work.
We're using more AI than we admit
When asked which tools they had used for work in the past three months, only 31% of respondents checked "AI embedded in software." Compare that with 68% for ChatGPT, 30% for Gemini, 28% for Claude, and 27% for Microsoft Copilot.
I do not believe 31% is the real figure for embedded AI, and neither should you. Questions you might ask yourself to get to your answer:
- Do you use any editing software? Built into Word, or Google Docs, or Grammarly? Powered by AI.
- Are you using software to find grants? Whether the company is explicit about it or not, there is almost certainly AI running in the background.
Why does this matter? The grant writer who believes they are a non-user is making decisions from a false starting line. They may decide they have nothing to disclose if a funder asks whether they are using AI, or that building AI skill is a someday/tomorrow project, even though they are already three or four AI-assisted tools deep.
The adoption paradox in the survey tells the same story from the other side. 71% of respondents use AI regularly or heavily, yet only 27% say it has changed how they work.
Grant professionals report in the Instrumentl survey that the top use, by a wide margin, is editing and rewriting their own drafts, at 71%. So AI is everywhere in the field, and for most people it is polishing the last 5% of a document rather than changing the grant production workflow. We have adopted the tool without adopting a new way of working.
We're afraid of the wrong thing
The single loudest emotion in the survey is fear of getting caught. 72% of respondents said it was very or somewhat true that they worry funders will penalize AI-assisted proposals. That is the highest-scoring anxiety in the entire dataset, above concerns about accuracy, ethics, and fear of being replaced. We are a rule-abiding bunch, aren’t we? Even among the heaviest AI users, 62% share the same worry, suggesting this is not a beginner's fear that fades with skill.
Part of that fear is well-founded and stems from two major challenges.
- Funders have not provided broad reassurance that the use of AI is okay. And the funders who don’t want you to use AI haven’t clarified what that means. Do we mean, don’t use AI to correct grammar/spelling? Do we mean, don’t use AI to score and revise your application? Ostensibly, both of those use cases would improve the output.
- AI detection tools return false positives, and they do it often. In my own work, my writing routinely scores between 30% and 50% "likely AI" on these detectors, even when I wrote every word myself (I’ve run several AI detectors on my work from 2022 or earlier). Clean, well-organized, plainly worded prose is exactly what these tools flag. So the strongest human technical writers are suspected at roughly the same rate as people who use AI badly, and there is no fair adjudication process behind any of it.
When 42% of survey respondents expect funders to be "split and inconsistent" about AI over the next few years, they are reading the situation correctly. That’s a heck of a landscape to try to navigate.
But the fear is also overblown. Very few funders have banned AI outright. The survey shows the field expects inconsistency, not prohibition: only 18% expect funders to grow more skeptical or try to detect AI, while 26% expect funders to increasingly expect or accept it.
For most grant writers, the bigger risk is letting a vague, unaddressed fear keep them from using AI well or disclosing it plainly. Hiding your use of AI does not protect you. It reinforces the myth that the use was shameful to begin with.
I hold a clear position on this, and I will state it plainly: the use of AI in grant writing should be normalized and disclosed.
And in three to five years when we all know AI is being used regularly in grant writing, we won’t have to disclose it any longer. It will be assumed.
Until then, read each funder's actual AI policy, if they have one. Keep careful documentation of what was AI-assisted and what was not, because in a year or two the source material itself, the logic models, evaluation plans, and boilerplate you sample from, will increasingly have been touched by AI, and you will want a record of what you can safely reuse if you aren’t allowed to use AI for a particular funder. I wrote a lengthy piece on this topic if you want to read more. AI Certifications in Grant Applications: What This Really Means for Grant Professionals.
And, always, always, own final responsibility for everything you submit. Do those things, and the fear has a productive outlet; use the fear to drive accountability through documentation, something grant professionals already know a thing or two about.
We're guarding the wrong tasks
The survey asked which human skills AI cannot replace.
The field gets it mostly right. 78% named building and maintaining funder relationships, and that belongs firmly in the human sector.
From here things get rocky:
48% say AI can't deeply know your programs and impact story. I'd argue that this is something it can do. Mine certainly does! What it can't do is understand a story you’ve never told it. 47% noted that reviewing and editing for accuracy and voice is something they don’t trust AI to do. But, when AI is trained on a thorough voice print (essentially a training manual on your organization’s voice), the human review process shifts from a full-on editing session to a quick confirmation. I’ve collaborated with Instrumentl on a training on writing an efficient voice print tailored to your organization - join us live on July 28!
34% named "ensuring data privacy and ethical compliance" as a human-only skill. I understand the instinct, but I think it is misplaced. AI is a capable partner for exactly this kind of work. It can cross-check a proposal against a NOFO's requirements, flag a missing certification or non-compliant budget item, and surface inconsistencies. However, governance around inputting sensitive information into the tool and accountability for the result remain the human user’s responsibility.
Now, the one that should worry us.
Only 18% named "judgment calls about whether to apply at all" as a human skill AI cannot replace. I think that is the most consequential decision a grant professional makes, and one where a human touch is most necessary.
The go/no-go call is upstream of everything else. When properly trained and given access to the right tools, AI can help inform the decision. It can surface opportunities, summarize a funder's priorities, and lay requirements side by side. It cannot own the choice to spend your organization's finite time on a specific pursuit. That judgment is high stakes, and requires human discernment.
Redraw the human-AI line by consequence, not by comfort
Draw the human-AI line around what carries consequence and cannot easily be reversed, rather than around what feels personal. Drafting a narrative section, summarizing an RFP, running a first-pass compliance check, generating a budget narrative you will verify line by line: these are reversible, low-stakes-to-check tasks where AI earns its keep and a human confirms the result. The decision to pursue a funder, the choice of which data to expose, and the final accountability for what gets submitted: these are human decisions because they are strategic, high-stakes, hard to reverse, and yours to answer for.
That redrawn line has a useful side effect. It aligns with what the survey found: grant professionals are willing to own the work. They want to remain responsible for what they submit. Drawing the line by consequence gives that instinct a clear home. You own the consequential decisions, and you let the tool carry the reversible production work in between.
Where to start this week
You can correct your line this week with an honest inventory. No new tool or course required.
- List every place AI already touches your work, including the embedded features you never counted, so you are working from a true starting point.
- Take one funding opportunity currently on your list and treat the go/no-go decision as the real work it is: write down why you are applying, whether the fit is genuine, and what the effort will cost.
- Pick one funder and read their actual stated position on AI rather than assuming the worst.
None of that requires you to keep up with every model release or assess every AI tool out there. The foundational moves (working in projects with instructions, giving AI good context, learning why AI fails) have not changed much in the last year and seem unlikely to change much in the next few. If you haven’t yet learned the basics, come check out the affordable 3-hour GPCI-approved course, AI for Grant Writers.
To move through this period of transition, grant professionals must not underestimate the seriousness of the stakes and still approach the tools with curiosity. Draw your line by consequence, protect the spaces where human judgment is essential, and let the tool do the reversible work in between. That is a line you can defend, to a funder and to yourself*.
*Of course, if a funder says not to use AI, don’t. And if a fund asks if you used AI, don’t lie.
