Everything CRAA generates is a draft for you to check. You remain the author of your application. Before submitting anything to a funder, you are responsible for verifying every fact, figure, name, date, location, outcome, word limit, eligibility claim, and attachment.
How answers are produced
CRAA works in two stages, and it is worth knowing which is which.
- Reading your documents. Uploaded files are converted to text, then an AI model (Google Gemini) identifies the application’s questions, word limits, program areas, required attachments, and scoring criteria, and proposes reusable facts about your organization. If that model is unavailable, CRAA falls back to a pattern-matching parser that only recognises certain application formats.
- Drafting answers. For the built-in JCOD application, answers are composed from fixed templates filled with facts you have confirmed. For any opportunity you upload, answers are drafted by the AI model using only the facts you have confirmed.
What CRAA is designed not to do
- It does not invent facts. Drafting is instructed to use only confirmed information and to report what is missing instead of filling a gap. Where information is absent, you should expect an empty answer and a prompt, not an assumption.
- It does not auto-fill identity-sensitive fields. Contact details, leadership names, and community-representation claims are never taken from uploaded documents or websites. You enter those yourself.
- It does not submit anything. CRAA never contacts a funder or uploads to a grant portal. Attachment checklists inside CRAA are planning notes only.
- It blocks internal language from reaching your answers. Drafts that contain system or instruction-style wording are rejected rather than shown.
Known limitations
These are real and expected. None of them are bugs you should report; they are the current boundaries of what the tool can do.
- AI models make mistakes. Language models can misread a document, misattribute a number, drop a qualifier, or state something with confidence that is wrong. Fluent writing is not evidence of correctness.
- Document parsing is imperfect. Complex tables, multi-column layouts, footnotes, and unusual formatting can be misread. Questions or requirements may be missed or merged.
- Scanned documents are not read. Image-only PDFs contain no extractable text and are rejected; optical character recognition is not connected.
- Quality and confidence scores are indicative only. They reflect how much confirmed information supported an answer — not whether the answer is accurate, competitive, or compliant. A high score is not a prediction of funding.
- Word limits are counted approximately. Funders may count differently. Confirm against the official portal.
- Deadlines, eligibility, and requirements shown in CRAA come from your uploaded documents, which may be outdated, amended, or superseded. Always confirm against the funder’s official listing.
Your responsibilities before submitting
- Read every answer in full and correct anything inaccurate or out of date.
- Confirm every number, award amount, date, and statistic against your own records.
- Check eligibility and priority claims against the funder’s official criteria.
- Confirm official word limits, required attachments, formats, and the submission deadline.
- Put the answers into your organization’s own voice.
No guarantee of any outcome
CRAA does not guarantee that an application will be accepted, scored favourably, deemed eligible, or funded. Funding decisions rest entirely with the funder. No part of this service is legal, financial, tax, accounting, or professional advice, and using CRAA does not create a professional relationship of that kind.
Reporting a problem
If CRAA produces something inaccurate, misleading, or unsafe, please report it so it can be corrected. See the Terms of Service for contact details, and the Privacy Policy for what happens to the documents you upload.