The purpose and current product context
This guide shows how to use Perplexity to produce a reusable instruction set with meaningful acceptance checks. Start with a recurring task, accepted example, and failure cases. The central question is: When should this skill run, and how will we verify its output? A useful workflow has a recognizable finish line and a way to distinguish a supported result from an attractive but incomplete response.
Perplexity moved workflows to Skills in September 2026. Skills provide reusable methodologies that guide Computer and its supporting agents. The documentation describes built-in and custom skills; check installation and organizational controls. Check the current official guide.
A worked example
An agency repeatedly prepares research briefs. Its skill defines source preferences, output sections, and unresolved-question handling. A test deliberately includes missing evidence. The accepted output labels the gap instead of creating a plausible fact, proving that the reusable method respects its boundaries.
The example illustrates a review decision, not a reported product test. Before applying it, write down which information in your own situation is confirmed and which is provisional. If you cannot explain the difference, resolve that first. Otherwise, a summary may appear to solve the problem while merely making an uncertain assumption easier to overlook.
Choose the right approach
Specify triggers as carefully as instructions. A skill should match a defined task and preserve a reviewed method. Include example inputs and outputs, but remove private client details. Update the method when repeated mistakes show that an instruction is ambiguous.
Step-by-step workflow
- Define the trigger. Begin by defining the scope of this step and the evidence it needs. Keep the question visible while collecting context. If a detail is missing, mark it rather than substituting a likely value. This makes later review specific and helps you avoid correcting an answer to a different question.
- Write the method. Check the available product controls and the source material before proceeding. Confirm that the intended account and permitted information are being used. Test a small known detail where possible. A setup or access problem should be resolved as such, rather than mistaken for a failure of your research method.
- Test a normal and incomplete input. Request a bounded output and inspect the first result against the requirements. Focus follow-up instructions on a concrete mismatch. Name the omitted condition, unclear relationship, or unsupported statement. Preserve correct evidence during revision so improving presentation does not accidentally alter the factual meaning.
- Review before reuse. Apply the review criteria before using the result elsewhere. Keep consequential claims traceable and unresolved points separate. If the answer needs a decision from another person, record that decision explicitly. Finish with an output someone can understand and review without reconstructing the entire conversation.
A reusable prompt
Use the prompt below as a brief. Replace bracketed fields with approved information and remove requirements that do not apply. Keep source text distinguishable from your instructions. The prompt is useful only when its assumptions match the task; pasting it unchanged does not establish that the resulting answer meets your needs.
Task: When should this skill run, and how will we verify its output?
Inputs: [provide a recurring task, accepted example, and failure cases]
Audience: [intended reader]
Required result: a reusable instruction set with meaningful acceptance checks.
Use approved sources and preserve dates, definitions, conditions, and exceptions. Separate confirmed facts, interpretation, and unresolved questions. Do not invent missing details. Show the evidence needed to review consequential claims. Ask one focused question if a missing requirement blocks the result.
Acceptance checks:
- Trigger matches the intended job
- Outputs follow the accepted structure
- Missing evidence remains visible
- Examples are sanitizedReview checklist and evidence table
- Trigger matches the intended job
- Outputs follow the accepted structure
- Missing evidence remains visible
- Examples are sanitized
Review substance before presentation. Ask whether the output preserves the actual requirement, important conditions, and the status of each statement. Then check readability and navigation. A polished layout can make an unsupported conclusion more persuasive, so visual improvement should follow factual review rather than replace it.
| Check | What to retain | What to do if it fails |
|---|---|---|
| Trigger matches the intended job | Source passage, observed result, or explicit decision | Correct the mismatch and mark remaining uncertainty |
| Outputs follow the accepted structure | Source passage, observed result, or explicit decision | Correct the mismatch and mark remaining uncertainty |
| Missing evidence remains visible | Source passage, observed result, or explicit decision | Correct the mismatch and mark remaining uncertainty |
| Examples are sanitized | Source passage, observed result, or explicit decision | Correct the mismatch and mark remaining uncertainty |
Common mistakes and repairs
A vague description can activate a skill in the wrong context. Narrow its purpose.
Diagnose the failing layer before repeating the entire task. If the source is wrong, repair the input. If the question is ambiguous, narrow it. If the output changes the meaning, correct the specific claim. Keep a record of repeated defects so the reusable instruction or review stage can be improved, rather than patching the same final sentence every time.
Practice the workflow
Write a research-brief skill and test a case with one unavailable source.
Complete the practice once, record the reviewed outcome, and then change one important condition. Explain what should change in the answer before requesting another response. This small variation tests whether your method handles the task rather than simply reproducing a familiar example. Use the checklist to identify the exact error if your expectation and the response differ.
Save a useful result
Save a reusable instruction set with meaningful acceptance checks together with its source references and review date. Preserve trigger matches the intended job as a visible acceptance condition. Include unresolved questions and the person or evidence needed to settle them. Remove private details from reusable prompts so the method can be shared without exposing unnecessary context.
Download the prompt and checklist for your practice record.
Frequently asked questions
Are workflows still the right navigation label?
The current guide explains the September migration to Skills.
What makes a reusable skill useful?
A clear trigger, tested method, and accepted output example.
Should skills contain real client examples?
Use sanitized examples unless the information is necessary and approved.
What should I prepare before starting?
Prepare a recurring task, accepted example, and failure cases. Remove material unrelated to the question. Name the outcome and decide how you will check it. This preparation gives you a basis for judging the answer rather than accepting any polished response as useful.
How do I adapt this workflow to my own situation?
Replace the fictional scenario with approved information and keep the underlying question explicit: When should this skill run, and how will we verify its output? Adjust the audience and output format. Preserve conditions and source details that affect the result instead of copying the example mechanically.
What if the first answer is too broad?
Return to the intended result: a reusable instruction set with meaningful acceptance checks. Name the missing requirement and ask for a focused correction. Adding unrelated instructions often makes the task harder to review. Compare the revision with your original question and retain unresolved details.
How should I handle missing information?
Keep it explicitly unresolved. Explain which source, permission, or decision would settle it. A plausible guess can be harder to detect than an obvious gap. For this workflow, completeness means knowing the limits of the answer as well as the confirmed result.
How can I check whether the workflow helped?
Compare the reviewed output with a baseline you understand. Apply the checklist in this guide and record what needed repair. Judge relevance, accuracy, and usability rather than response length. Keep one accepted example to make later reviews consistent.
What should I keep after finishing?
Retain the reviewed result, source versions, prompt, and unresolved questions in the place where work continues. Remove unnecessary private details from reusable examples. Another reader should understand the result without replaying every chat turn or relying on assumptions you never wrote down.
When should I recheck the product instructions?
The official references on this page were checked on 5 October 2026. Revisit them when your interface, account plan, source behavior, or permissions differ. Product controls can change; the practical review method remains useful, but documented availability should always be checked for your own account.
Official sources and related guides
The sources above support product details. The examples, review decisions, and practice exercises are original editorial guidance. When product documentation and your account differ, check the applicable account and platform guidance before following a step. Keep evidence beside consequential conclusions instead of treating links as decorative proof.
Continue with Perplexity Scheduled Tasks: Monitor Useful Changes, Perplexity Privacy: A Practical Data Review, NotebookLM for Exam Revision: A Study Routine.
