OpenWater AI launched in December, and AI has been one of the most common topics in our support inbox ever since. Some of you want to know what it can do. Others are already running prompts on live programs and want to get more out of them.
We went through six months of support conversations to see how you’re using it and where you get stuck. Here’s what we learned.
A quick refresher
OpenWater AI lets you write your own prompts and run them on your submissions. You decide what the AI reads and what it gives back: a summary, a score, a label, a flag for review. Results appear on each submission, and you can pull them into reports, judge views, email blasts and galleries.
It’s opt-in, so nothing happens until you ask for it. Your support rep can turn it on.
How customers are putting it to work
Giving reviewers a head start
One association is piloting AI summaries for a nomination program with several open positions. A prompt writes a short summary of each nomination, tailored to the position, and reviewers read it before they score and rank. The reviewers still make every decision. They just start with the key points in front of them.
Tip: Add the AI result to your Evaluation Layout (Round Settings > Evaluation Layout) so judges see it right next to the submission.
Sorting applications automatically
A startup program set up a prompt that reads each application and labels it as business-to-consumer or not. The answer lands in its own column. What used to be manual tagging is now something they can filter and report on.
Testing AI against a scoring rubric
Several customers are testing AI as an impartial reviewer. One team gave the AI their 1 to 5 rubric for a leadership program and asked it to score each applicant. Another wants to compare AI output with how their human judges scored a past cycle.
Before you lean on AI scoring, see whether it lines up with your judges.
Saving the AI for the finalists
A teacher awards program uses conditional execution. Their prompt only runs on submissions whose average reviewer score clears a set threshold. Top submissions get an AI write-up, and the rest don’t use up tokens.
Keeping judging blind
A professional association runs an awards program where applicants submit essays about their work. To keep judging fair, they use a prompt that acts as an anonymization editor. It rewrites each essay to remove anything that could point back to the author: their name, colleagues and clients, the employer and its brands, telling location details, publicly reported financials, and links or email addresses. What judges need stays in, including the structure, the results and metrics, and the type of certifications the author holds (minus certification numbers).
Two rules make it work. The prompt tells the AI to swap names for natural wording, like “a global financial services firm,” instead of placeholders like [COMPANY], so the essay still reads smoothly. And when the AI isn’t sure whether something is identifying, it takes it out.
Tip: Insert the anonymized version into your Evaluation Layout and hide the original essay field from judges. (screenshot below)
Checking applications before judging
Grant programs keep asking us about pre-screening. A prompt can check whether an application covers every required section, or flag possible eligibility issues, so judges spend their time on complete applications.
Your top questions, answered
💠 How do I turn it on?
Ask your support rep to enable it. Then click Manage on your program, go to Round Settings > AI Prompts and click Add New AI Prompt.
💠 I wrote my prompt. Now what?
This is the question we hear most. Check your Execution Mode: set it to always run, or to run when a submission meets your conditions, and save.
To see results, open a submission and click AI Prompt Results. To test, click Rerun AI Prompts on a test submission, adjust the prompt in another window, and rerun until the output looks right.
💠 How do I run it on submissions that are already in?
Go to Round Settings > AI Prompts and click Rerun AI Prompts. This runs across every submission in the round, so keep an eye on your token usage.
💠 What are tokens, and how many do I get?
Tokens measure how much text the AI reads and writes. Roughly, more words means more tokens. During the beta, every customer gets 10 million tokens at no cost for a limited time.
You can buy more, or connect your own OpenAI API key and pay OpenAI directly.
If a large rerun stalls, check your token balance first. Running out is the most common cause we see.
💠 Can the AI read uploaded files?
Yes. Check Parse Uploaded File and add the variable to your prompt. It works with PDFs, Word documents, and images. Parsing is best-effort, so tell the AI what to do if content is missing.
💠 Can AI analyze my whole program at once?
Prompts work one submission at a time. They’re built for summaries, scores and labels on each entry. For program-wide trends, export your AI results (Submissions > Reports > CSV/XLS Exports, then Subset of Fields) and look at them in Excel.
💠 Can OpenWater tell if applicants or judges used AI?
No. There’s no built-in AI detection. Many programs add a disclosure question to their application form instead, with a follow-up that only appears when someone answers yes. Our post on setting an AI policy for your awards program is a good place to start.
💠 Is our data used to train AI models?
No. Prompts run through Cloudflare, which is already part of our infrastructure. Neither OpenWater nor Cloudflare uses your prompts to train models.
💠 What if we don’t want to use AI?
That’s fine. OpenWater AI stays off unless you ask us to turn it on.
Tips for writing a better prompt
• Use the system prompt for rules. Give the AI its role, tone, and scale there, for example: "You are an impartial evaluator. Do not invent information."
• Use the user prompt for the task. Say exactly what you want and insert the field variables it should read.
• Spell out the format. "Return only B2C or Not B2C" is easier to filter than a paragraph.
• Plan for gaps. Tell the AI what to say when a field is blank or unclear.
• Test on a few submissions first. Then run it on the whole round.
• Try a different model. Some models handle scoring better, others are faster for short summaries. You can change the model per prompt.
Stuck on wording? Describe what you want in your preferred LLM and ask it to draft a system prompt. Then paste it in and test.
Ready to try it?
Reach out to your support rep to turn on OpenWater AI. For step-by-step setup, see our help article, Apply AI / LLMs analysis on your submission data.