Funders read hundreds of proposals and can tell within a paragraph whether an organisation knows its own community. AI-written proposals fail on a single dimension: they describe a problem in general terms when the funder needs your specific one, with your numbers.
The phrases, and what to write instead
| Avoid | Write instead |
|---|---|
| in today's rapidly changing landscape | The specific change, with the year. |
| unprecedented challenges | The actual challenge, with a number. |
| a holistic, comprehensive approach | List the actual programme steps. |
| leveraging synergies between stakeholders | Name the partner organisations. |
| underserved communities | Name the community, the neighbourhood, the count. |
| we are uniquely positioned to | Say what you have that others do not. |
| evidence-based best practices | Cite the specific practice and the study. |
| capacity building | Say what capacity, and how much more of it. |
| transformative impact | The metric you expect to move, and by how much. |
| This funding will allow us to continue our vital work | State what specifically it buys. |
| A needs statement with no names, places or numbers | Add all three. This is the single biggest tell. |
| Bland, voiceless prose throughout | Write the way your staff talk about the work. |
Score this draft before you send it
Paste your text into Slop Score. It checks every rule on this page plus the full 200+ list, marks each hit in place, and cleans the mechanical ones. Runs in your browser; nothing is uploaded.
Check my draft →The rule underneath all of them
Every replacement above does the same thing: it swaps a claim for evidence. That is the whole method, and it is why no tool can finish the job for you. A model can delete results-driven. Only you know the number that belongs there.
Density is what readers notice, not any single word. One phrase from this table is a coincidence. Four in a short piece is a pattern, and people now recognise the pattern quickly. The full 200+ tell list covers the vocabulary, sentence patterns and machine artifacts that apply everywhere, whatever you are writing.
Questions
Can funders tell if a proposal was written by AI?
Programme officers usually describe it as a proposal that could have come from any organisation in any city. That is the real signal, not vocabulary. A needs statement with your neighbourhood, your waiting list and last year's numbers cannot be produced by a model that has never seen your files.
Does using ChatGPT hurt my chances of getting funded?
Using it is not the problem. Submitting its default output is, because it writes the average of every proposal ever published. Use it to structure sections, then fill every claim with your own data.
What makes a needs statement sound generic?
The absence of proper nouns. If the statement names no place, no partner, no number and no date, it is describing a category of problem rather than the one you are asking to be funded for.