Troubleshooting AI Behavior

If Grantable’s AI isn’t responding the way you expect — producing weak, off-topic, or repetitive results — a few simple checks can usually resolve the issue.

Written By Philip Deng

Last updated 4 months ago

Common Symptoms

You may notice:

  • The AI’s responses are too general or miss the point of your question.

  • The AI seems to ignore your files or context.

  • Responses repeat language or restate the same ideas.

  • The AI fails to cite when Research (Pro) is enabled.

  • The AI stops mid-response or times out.

💡 Most issues come down to configuration or missing context — not system errors.


Step 1: Check Your AI Settings

Open the ⚙️ Settings icon in the chat box to review your configuration.

  • Use Organization Profile — Enable this to ensure the AI references your factual, unchanging data (e.g., name, EIN, mission).

  • Follow Style Guide — Enable this to restore your organization’s tone and phrasing.

  • Search Workspace — Enable this to let the AI access your internal files and documents for deeper context.

  • Research & Citations (Pro) — Enable this only when you need the AI to perform web lookups and provide sources.

💡 If you’re getting inconsistent tone or missing details, it usually means one or more of these toggles are off.


Step 2: Check Attached Sources

If your prompt refers to a specific topic or document:

  • Confirm you’ve attached the correct file or record using the @ or paperclip icons.

  • The @ icon attaches items already in Grantable.

  • The paperclip uploads new files from your computer.

  • Check that the source appears as a card above the chat box — if not, the AI can’t see it.

💡 The AI only references what’s visible and attached — it doesn’t automatically search your entire workspace unless Search Workspace is toggled on.


Step 3: Refine Your Prompt

The AI performs best when you give clear, specific instructions. Try:

  • Asking for a particular format:

    “Write a one-paragraph summary highlighting the program’s measurable outcomes.”

  • Including context cues:

    “Using the attached file, summarize the eligibility criteria in plain language.”

  • Using verbs like summarize, rewrite, expand, draft, or compare.

💡 Avoid vague prompts like “write something better” — the AI needs direction on what “better” means.


Step 4: Review Workspace Content

If the AI pulls incorrect or outdated information:

  • Check your Organization Profile for outdated facts or contacts.

  • Update Files and Programs with the latest data.

  • Remove old or irrelevant attachments from the chat.

💡 The AI mirrors what it’s given — updating your workspace improves accuracy instantly.


Step 5: Check for Plan or Access Limits

  • If Research & Citations doesn’t appear active, you may not be on the Pro plan.

  • If AI responses pause or don’t generate, you may have hit your usage throttle — wait briefly or upgrade your plan.

  • Confirm your internet connection is stable (for web lookups).

💡 Plan limits don’t delete data — they only pause generation until your next reset period.


Step 6: Regenerate or Reset

If the AI gets stuck or outputs something strange:

  • Click Regenerate response to try again.

  • Rephrase your prompt or simplify it.

  • Clear attachments and reattach the most relevant ones.

💡 A small change in wording or context can completely change the quality of the result.


When to Contact Support

If the AI continues to behave unpredictably:

  • Responses appear empty or cut off every time

  • Citations consistently fail to load on Pro

  • Your workspace settings seem unresponsive

Contact [email protected] with a short description of what you were doing and include screenshots if possible.

💡 The support team can often diagnose whether it’s a configuration issue, system update, or rate limit.


Key takeaway: Most AI issues can be fixed by checking your settings, attachments, and prompts. Grantable’s AI works best when given clear instructions and relevant context — treat it as a collaborator that depends on what you share, not a mind reader.