In this guide
Start here: one thing to know
When the output isn't quite right
A few habits that sharpen any result
Power phrases that reliably improve results
What not to do
Still stuck? Here is how to reach us
Start here: one thing to know
Here is the good news: when the output misses, it is almost always something small and quick to fix. Most of the time it comes down to what the AI had to work with, and that is something you are fully in control of. The more it knows about your company and your goal, the sharper and more targeted its responses will be. A helpful rule of thumb: the better the input, the better the output.
Two things are worth remembering. First, Awarded AI is a non-training model, so every new chat starts as a blank slate with none of your company context carried over. Second, it helps to think of the AI like a brand new intern. It is capable and eager, but if you are not explicit, it will make its own assumptions to fill the gaps. Give it clear direction and the right materials, and you will see returns on the level of quality that it outputs.
📌 The big idea: You almost never need to scrap a session and start over. A weak first answer is a starting point. Refine it, and you will usually get where you want in a turn or two.
Common output issues (and how to fix them)
1. The output feels generic
This almost always means the AI does not have enough of your information to work from. Remember, it starts every session as a blank slate. Double-check that you have actually given it what it needs, in all three ways:
Direct injection. Paste key text, like a SOW, PWS, or Sections L and M, straight into the chat.
Attached files. Upload and attach the specific files you want it to use. Uploading alone is not enough; you have to attach the file to the session for the AI to be able to see and reference it.
Sources Panel. Toggle on the correct knowledge categories so that your company profile, past performance, and preferences are active for the session.
Add whatever was missing and run it again. More of your specifics in means less generic content out.
2. It states something that is not true (it hallucinates)
When the AI presents made-up facts or data as if they were real, you have two strong fixes, and you can use both together.
1. Turn on the built-in factual-accuracy rule. In your Sources Panel, toggle on this rule:
Output Rules & Styles > Rules & Instructions > Configure > General Rules > Use Factual Information Only
That rule tells the AI to "avoid making up facts or data. Stick to verified information and information provided."
2. Build your own rule. Create a "My Preferences" or "[Company Name] Preferences" knowledge category (see Building Your Knowledge Base in Awarded AI) and add a rule that tells the AI to never make up information, to always cite its source, and to reference only factual information from verifiable sources.
And any time you are unsure about a claim, ask the AI directly where it got that information.
3. It used real information, but I don't think I gave it that info??
Sometimes the detail is actually correct, you just are not sure where it came from. The easiest way to check is to ask the AI directly: "Where did you get this information? Be specific." Then follow its answer to verify the source yourself. If it turns out the AI was actually making it up, treat it as a hallucination and use the fixes in the section above.
4. It's using way too many bullet points 🙄
Add a formatting rule to your "My Preferences" knowledge category (see Building Your Knowledge Base in Awarded AI) that tells the AI to limit or avoid bullet points. Once that rule is toggled on in your Sources Panel, it applies across your sessions, so you are not fighting the same formatting battle every time.
5. It's forgetting or summarizing things I told it earlier
This usually means the chat has run out of context. Picture the model's context window like a pitcher of water: every message you add is another drop of water, and once the pitcher is full, the earliest content starts to overflow and gets lost. This usually doesn't happen with the default model for all customer environments (Google Gemini 3.1 Pro, which is a reasoning model with a 1M token context window). If you do experience this, however, two things can help:
Start a fresh chat. To carry the conversation forward, copy and paste the important sections of your old chat into the new one. Just double-check that your file attachments and Sources Panel selections match what you had before, so the new session is working from the same materials.
Check your model. Use one with a large context window, ideally one million tokens, such as Gemini 3.1 Pro. A bigger pitcher holds more of your conversation before anything spills over.
A few habits that sharpen any result
Beyond the specific fixes above, these habits improve almost any output.
Run your prompt through the Prompt Perfector
If your prompt was short or a little vague, this is the fastest win. Type your basic instruction into the Prompt Perfector, click Perfect My Prompt, and it rewrites your request into a far more detailed, contextually appropriate version. For example, a quick "Shred this RFI" comes back fully built out. You can perfect it again, refine it, then copy it into your chat.
Be specific, and go one step at a time
Vague prompts invite the AI to guess. Compare "Write an executive summary" with "Draft a DoD cybersecurity proposal executive summary highlighting our AI threat detection solutions and past performance." When a task is big, break it into smaller pieces and follow your existing, proven proposal process. Ask for the outline first, then work on your response section by section.
Capture your winning prompt
Once you land on great output, make "future you" more efficient by reverse-engineering the win: "Now write me a single prompt that would get me to this result in one shot." Then save that prompt to your Prompt Library as a template, so next time you start ahead instead of from scratch.
Power phrases that reliably improve results
Small wording cues can nudge the AI toward better answers. Sprinkle these into your prompts when a result needs sharpening:
Phrase | What it does |
"Be specific" | Pushes the AI toward precision and away from vague generalizations. |
"Take a deep breath and solve this step by step" | Encourages a methodical approach that breaks a complex task into manageable steps. |
"Important" / "Please prioritize..." | Signals what matters most so the AI focuses there first. |
"Clarify..." | Prompts a more thorough, detailed explanation. |
"Verbatim" | Tells the AI to extract exactly what is there rather than generate something new. |
"Cite your source(s)" / "Use data to support your answer" | Asks for evidence-based responses, which reduces made-up detail. |
"Summarize in bullet points" / "Provide step-by-step instructions" | Controls the format so the answer comes back the way you need it. |
"Consider potential challenges" / "What are the potential outcomes?" | Encourages foresight, so the AI anticipates obstacles and results. |
What not to do
A few habits quietly work against you when using AI. Steering clear of these will save you a lot of time in the long run:
Don't expect 100 percent perfection on the first try. Awarded AI is built to get you to a strong first draft fast, not to replace your expertise. Keep a human in the loop: review, refine, and run everything through your proven proposal process before it goes out the door.
Don't ask the AI to draft an entire proposal in one go. Big, all-at-once requests tend to produce shallow or uneven results. Work section by section instead, following your proven proposal process, and you will get stronger, more focused content.
Don't be vague. Ambiguous prompts invite the AI to guess and take liberties. Specificity is your friend.
Don't lean on exclamation points or ALL CAPS for emphasis. The AI does not read them as extra importance. Use words like "Important" or "Prioritize" instead.
Don't assume it remembers. Each new session is a clean slate. Re-supply your context every time, or better yet, set up your default configuration in the Sources Panel.
Don't expect it to read files you only uploaded. Uploading is step one. You still have to attach the file to the session, or the AI cannot see it.
Do not expect live information by default. Without Web Search or Deep Research turned on, the AI cannot access the Internet, meaning it can't pull recent or real-time facts.
(Typically) Don't scrap everything and start over. Refining and iterating on your existing prompt almost always beats restarting from zero.
Don't give up! Reach out to your CSM if you need help - that's what we are here for! We may be able to help you with a custom prompt, assistant, or knowledge category that vastly improves your results.
Still stuck? Here's how to reach us
If you have worked through these steps and the output still is not landing, please reach out! Your CSM is in Awarded AI every day and is an expert at prompting and troubleshooting, so they can often spot the fix quickly.
Live chat with support inside the app, in the bottom left of your screen.
Email your dedicated CSM
⚠️ Think you hit a bug? If a response errors out or behaves strangely, let your CSM know. They can submit it as a ticket so it gets looked at and resolved, rather than leaving you to work around it.
Procurement Sciences · Save Time. Deliver Faster. Win More.