Tutorial ID | T-25 |
Section | AI & Analysis |
Title | Saving and reusing AI prompts in the Prompt Hub |
Subtitle | Build AI prompts that produce consistent, correctly scoped reports every team member can find and trust. |
Before You Begin
You have access to AI Prompt Hub and at least one saved prompt for a case type you work with regularly.
You know the report structure, length, and content your team expects for that case type.
What You Will Accomplish
By the end of this tutorial, you will know how to write and maintain prompts that produce consistent output across cases, apply cleanly to every case they're intended for, and stay recoverable and accessible to the whole team. You will also know how to catch a prompt that's gone wrong before it affects a report.
Why Does this Tutorial Matter
A well-built prompt is a reusable asset — write it once, refine it against a few real cases, and it produces a consistent, complete report every time it's used on the right case type. That consistency is what lets a team review reports efficiently, checking against a known structure instead of starting from scratch each time.
Left unmanaged, a vague or narrow prompt produces different results from case to case, an unclear name leads to the wrong prompt being applied to the wrong case type, an unnoticed edit changes every report generated afterward, and a prompt that's deleted or restricted to one team has to be rebuilt or duplicated from scratch. Each of these turns a one-time investment into repeated, avoidable rework.
What to Do if Something Goes Wrong
The list below covers the most common problems teams run into with saved prompts. Each entry follows the same pattern: what you'll notice, why it likely happened, a real example, and how to fix it.
Problem: The same prompt produces reports of noticeably different structure, detail, or coverage across cases of the same type.
Likely cause: The prompt contains vague or open-ended instructions, so cases with a different volume or mix of documents get interpreted differently by the AI each time.
Example: A prompt says “summarize the medical history.” On one case that produces a tight chronological summary; on another, a scattered list of unrelated events.
Fix:
Open the prompt in AI Prompt Hub and identify any instruction that sounds vague, open-ended, or could be read more than one way.
Rewrite vague instructions using specific, measurable language — for example, replace “summarize the medical history” with “summarize the medical history in chronological order, covering diagnoses, procedures, and treatment outcomes.”
Add explicit structure instructions specifying the required sections, their order, and the expected length of each.
Run the revised prompt on two or three cases of the same type and compare the outputs side by side.
Refine the prompt further until the output is consistent across those cases.
If the prompt is already precise and the report quality is still low, open Files and check whether relevant records are missing, requesting them if needed.
Problem: A report is missing sections the case type needs, while sections that don't apply to it dominate the report.
Likely cause: A prompt built for one case type was applied to a different case type, often because prompts weren't named clearly enough to tell them apart at a glance.
Example: A prompt built for a burn accident case is run on a workers' compensation case, and the resulting report has no wage-loss section.
Fix:
Open AI Prompt Hub and review the names of all saved prompts to confirm each one clearly states its content.
Before running a prompt, open the Case Summary and Expected Outcome fields and confirm the case type.
Match the case type to the correct prompt, opening each candidate prompt's content if more than one could apply.
If a prompt was already applied to the wrong case type, open the report and identify which sections are misaligned.
Select the correct prompt and rerun the workflow with the appropriate document selection.
Compare the new output against the previous report to confirm the structure now matches the case type.
Once confirmed, delete the previous report to avoid ambiguity about which version is current.
Problem: A prompt produces incomplete output or fills gaps with assumptions on some cases of a type it's meant to cover.
Likely cause: The prompt was built around one specific case and assumes a document, provider, or detail that isn't present in every case of that type.
Example: A prompt built around a single treating physician produces an incoherent report when run on a case involving multiple specialists.
Fix:
Open the prompt in AI Prompt Hub and identify any instruction that assumes a specific document, provider, or case detail.
Rewrite those instructions to describe the category of information rather than a specific instance — for example, “summarize records from a treating physician” instead of naming one.
Add conditional instructions for expected variation, such as including a surgical section only if surgical records are present.
Run the revised prompt on several cases with different document compositions and compare the outputs.
Add a conditional instruction for any section that consistently fails when a particular document type is absent.
Save the revised prompt once it produces reliable output across the range of cases it's meant to cover.
Problem: A team member can't quickly tell which saved prompt is the right one to use.
Likely cause: Prompts are named generically, such as “Report Prompt 1” or “Final Version,” instead of describing what they're for.
Example: Three prompts are all named some variation of “Final,” and a reviewer under deadline pressure picks the wrong one.
Fix:
Open AI Prompt Hub and review the names of all currently saved prompts.
Identify any prompt whose name doesn't clearly convey its content.
Confirm a naming convention with your team that will apply to every renamed prompt.
Rename each unclear prompt using that consistent format — for example, “Bill Overview” or “Accident Description.”
If multiple versions of the same prompt exist, compare their content and archive or delete outdated versions, keeping only the current approved one active.
Problem: A report's structure or content changes unexpectedly, with no one aware of why.
Likely cause: Another team member modified a shared prompt in AI Prompt Hub without telling the rest of the team.
Example: A required back-reference instruction is removed from a shared prompt, and every report generated afterward is missing citations, with no one noticing until several reports have gone out.
Fix:
Before running any workflow, open the prompt in AI Prompt Hub and read through it in full rather than assuming it's unchanged.
Compare it against the expected structure for this report type, checking required sections, length, and back-reference requirements.
If you find an unexpected change, don't run the workflow — note what changed and raise it with the team first.
Restore the prompt to its approved version and confirm with the team that the correction has been made.
Going forward, document any intentional prompt changes — what changed, why, and when — in the prompt name or in logs the whole team can access.
Problem: A prompt the team relied on is gone, and there's no working copy to fall back on.
Likely cause: The prompt was deleted, intentionally or by mistake, without a backup saved anywhere outside AI Prompt Hub.
Example: A carefully refined prompt is deleted during cleanup, and the team has to rebuild it from memory before the next case of that type can be processed.
Fix:
Check whether a report previously generated with the deleted prompt is still available in the case, and use its structure and content as a reference to rebuild the prompt.
If the report alone isn't enough, open AI Chat, paste in a sample section from the report, and ask it to suggest the instruction that would have produced that output.
Refine the suggested instruction based on your knowledge of the case type.
Review the reconstructed prompt against a recently generated report to confirm it produces the expected output, then save it in AI Prompt Hub.
Before intentionally deleting any prompt, confirm with at least one other team member that no active case relies on it.
Before completing a deletion, copy the full prompt text into a shared document outside the platform so it can be recovered later.
Problem: A team can't see or use a prompt that another team already built for the same case type.
Likely cause: The prompt's Team field restricts access to specific teams, and the team that needs it wasn't included.
Example: A well-refined burn accident prompt exists, but a second team handling the same case type can't find it and builds a duplicate from scratch.
Fix:
Open AI Prompt Hub and locate the prompt the team can't access.
Open the prompt and find the Team field, which controls which teams can view and use it.
Confirm whether the field is empty, meaning all teams can access it, or lists specific teams only.
If the needed team isn't listed and the field isn't empty, add the team from the available list.
Ask a member of that team to refresh AI Prompt Hub and confirm the prompt is now visible.
Notify the team that a new shared prompt is available, including its name and a brief description.
Problem: The same prompt produces inconsistent output from case to case with no clear pattern.
Likely cause: The prompt contains two instructions that contradict each other, often added gradually by different team members without anyone reviewing the prompt as a whole.
Example: A prompt asks the AI to “be concise” and, separately, to “include all available detail from the source documents,” and the AI resolves the conflict differently each time.
Fix:
Open the prompt in AI Prompt Hub and read through every instruction from start to finish as a whole, not one at a time.
Identify any pair of instructions that pull in opposite directions, such as conflicting length, tone, section, or focus instructions.
For each conflict, decide which instruction should take priority for this case type and rewrite or remove the other.
Read the revised prompt through once more and confirm every instruction points toward the same coherent output.
Update the prompt in AI Prompt Hub and notify the team of the change.
Run the revised prompt on two or three cases of the same type to confirm the output is now consistent.
Problem: A report's length, section order, or referencing style varies unpredictably from one case to the next.
Likely cause: The prompt doesn't specify the required report structure, length, or back-reference format, so the AI makes a different choice each time.
Example: One report runs six pages with citations after every paragraph, and the next runs fourteen pages with a single citation at the end — both from the same prompt.
Fix:
Open the prompt in AI Prompt Hub and check whether it includes explicit instructions for structure, length, and back-reference placement.
If structure isn't specified, add a section list in the required order.
If length isn't specified, add a page count or range instruction.
If back-reference requirements aren't specified, add a placement instruction, such as requiring a source citation with document name and page number.
Run the updated prompt and confirm the output matches the structure, length, and back-reference instructions.
Refine and rerun if any requirement still isn't met.
Tip Treat every prompt as shared infrastructure: name it clearly, keep it accessible to every team that needs it, and back up major changes in a shared document outside the platform. A prompt that breaks one of these habits usually breaks the others too. |
Summary
Every prompt is now specific enough to produce consistent output, matched to the right case type, clearly named, backed up, accessible to the teams that need it, and free of internal contradictions — so a saved prompt keeps paying off case after case instead of creating rework.