6.2Recurring issues (playbook)
The essentials in 30 seconds
- Reflex: one symptom = likely causes = one targeted action — diagnose before re-prompting at random.
- Structure: a vague answer is fixed with an RTCF prompt with an example; invention is fixed with verification and sources.
- Reference: any context gap is fixed with the right file (/file) and verified synchronization.
Hicham thread — the playbook taped next to the screen
Hicham keeps this playbook next to his screen. Verdict after one quarter: three lines served him most — the vague prompt that produced off-topic content, the unreferenced file Copilot said was not found, and the wrong Excel formula from a range never converted into a table. For each one, he ended up applying the sequence (symptom, causes, action) instead of re-prompting at random.
Observed symptom, verified causes, actions in order: the reflex holds in three steps.
One reflex per symptom
Most disappointments with Copilot come from classic causes: too vague a prompt, unreferenced file, unverified result. Linking each symptom to its causes and to one action saves precious time. The table below groups the most frequent situations.
| Symptom | Likely causes | Actions |
|---|---|---|
| Vague / off-topic answer | blurry prompt, no context (L0.4) | structure in RTCF: role, task, context, format; give an example |
| Copilot "invents" (hallucination) | it generates the plausible, not the true | always verify facts and figures; ask for sources; reference a file |
| "I cannot find that file" | file not referenced or not synced | use /file (L1.4); check OneDrive/SharePoint |
| Unsuitable tone (too casual/formal) | tone context not specified | specify the tone and the recipient in the prompt; have it reformulated (L1.3) |
| Wrong Excel formula | need poorly worded, data not in a table | convert into a table (L2.1); describe precisely; verify the result (L2.3) |
| Doubt about confidentiality | not knowing the data scope (L0.3) | review L0.3: Copilot 365 respects access rights; do not paste external secrets |
| Copilot unavailable in the app | unsuitable licence or version (L0.2) | check the Copilot 365 licence (L0.2) and the Office update |
| Incomplete meeting summary | recording/transcription missing | enable Teams transcription; reformulate the request (L4.1) |
| Wasting time re-prompting | writing everything all at once | iterate: start from a first result, then refine with small requests |
Ready-to-say script — diagnosing without re-prompting at random (30 s)
"My symptom: [vague / invention / file not found / wrong formula]. I check the playbook's likely cause (prompt, file, table, licence), I apply the targeted action and I check the new result."
Copy-paste template — framed re-prompt after off-topic content
First result judged [vague / off-topic].
New prompt: You are [role]. Do [precise task].
Context: [file + attached /file]. Format: [length + tone].
Counter-example to avoid: [what the first result missed].
Small business / tight budget
No formal table: one symptom, the playbook's likely cause, the action and the owner noted.
Algiers SME
Integrate local realities into the diagnosis: shared files, verified synchronization, known owner.
Remote / hybrid
Add written traces to the actions: at a distance, an unrecorded symptom looks like a tool that works.
Cross-cultural
Spell out findings in writing: what seems obvious (which file, which tone) is not obvious to everyone.
48h challenge — apply tomorrow
Spot 1 current symptom (vague, file not found, wrong formula), apply the playbook line (named symptom + verified cause + targeted action).
- I can link a symptom to its likely causes.
- I structure in RTCF before re-prompting.
- I can lock in with a referenced file and verification.
Copilot answers off-target on Hicham's file. Which reflex applies the playbook?
One same failure that keeps coming back is not ten incidents: it is a missing reflex to fix at the root (prompt, context, referencing, verification).