News

Copilot is capable. But it is not a mind-reader.

When people say Copilot is not that good, what they often mean is this: the output did not match the real need. Usually, that is not a model problem. It is a workflow problem.

Copilot outputs go wrong in predictable ways. Once you can spot the patterns, you can fix them in seconds, not minutes. This article covers the 7 most common failure patterns I see in real organisations, plus fast fixes and reusable prompt lines you can keep on hand.

A quick rule that prevents most issues

If you want reliable results, always specify:

• the audience
• the purpose
• the format
• the constraints

If one of those is missing, expect rework.

1. The brief is vague, so the output is generic

What it looks like

You get bland, safe, surface-level content. It sounds plausible but says very little.

Why it happens

Copilot is filling in missing context with best-guess defaults. That produces generic writing.

Fix it fast

Add the decision, the audience, and what good looks like.

Prompt lines to reuse:

• Audience is [role], they care most about [2 priorities].
• The goal is to [decide / approve / inform / persuade].
• Write it as [email / table / 6 bullets / 1 page brief].
• Success looks like [specific outcome].

2. It answers the wrong question

What it looks like

You asked for an email and got a policy. You asked for recommendations and got a summary. Or it focuses on the wrong part of the problem.

Why it happens

Your request describes the topic, not the deliverable.

Fix it fast

Ask for the output first, then the content.

Prompt lines to reuse:

• First, confirm what you will produce in one sentence.
• Then produce it, following this structure: [structure].
• If my request is ambiguous, ask 2 clarifying questions before writing.

3. Hallucinated facts, invented numbers, or confident guesses

What it looks like

Made-up dates, fabricated metrics, false product capabilities, or claims that sound authoritative but are not grounded.

Why it happens

Copilot optimises for fluent completion. If you do not instruct it to verify, it will attempt to be helpful by filling gaps.

Fix it fast

Force evidence, constrain sources, and require uncertainty where needed.

Prompt lines to reuse:

• Only use information from [this document / this email thread / this spreadsheet].
• If you cannot verify a detail, write unknown rather than guessing.
• Separate facts from assumptions in two lists.
• Add citations or references to where each fact came from in the content I provided.

Practical habit:
If accuracy matters, use Copilot to draft structure and wording, then validate facts yourself.

4. It misses the internal context that matters

What it looks like

The output ignores your organisation’s tone, policies, stakeholder sensitivities, or history. It is technically correct but politically wrong.

Why it happens

Copilot cannot infer your unwritten rules unless you tell it.

Fix it fast

Add a simple context block once, then reuse it.

Prompt lines to reuse:

• Our organisation tone is [direct / warm / formal]. Avoid [phrases].
• Constraints: do not mention [sensitive detail]. Do not blame.
• Stakeholders: [names/roles] care about [priorities].
• Background: previous decision was [x]. We must align to [policy].

5. The output is too long, too short, or the wrong shape

What it looks like

A wall of text when you need a skim-friendly brief. Or a short answer when you need a complete plan.

Why it happens

You did not define format, length, or level of detail.

Fix it fast

Give a template and set a length.

Prompt lines to reuse:

• Keep it under [120 words / 1 page / 6 bullets].
• Use headings: Situation, Options, Recommendation, Next steps.
• Use a table with columns: Action, Owner, Due date, Risk.
• Write for scanning: short paragraphs and bullets only.

6. It produces polished text that is not actionable

What it looks like

It reads well but does not move work forward. No owners. No dates. No decisions.

Why it happens

Copilot will happily write content. It will not automatically convert content into execution unless asked.

Fix it fast

Force actions and accountability.

Prompt lines to reuse:

• Turn this into an action plan with owners and deadlines.
• Extract decisions made and decisions required.
• List blockers and what is needed to unblock them.
• Give me the next 3 steps I should do today.

7. It gets the tone wrong (too harsh, too fluffy, too salesy)

What it looks like

A reply that sounds irritated, defensive, overly enthusiastic, or corporate.

Why it happens

Tone is subjective and Copilot chooses a default based on general business writing.

Fix it fast

Specify the emotional intent, not just the style.

Prompt lines to reuse:

• Tone: calm, firm, helpful. No passive aggression.
• Aim to protect the relationship while being clear on boundaries.
• Write as a senior professional: concise, direct, respectful.
• Remove anything that could sound like blame or judgement.

Fast technique:
Ask Copilot to produce two tone options so you can choose quickly.

A simple troubleshooting playbook you can teach your team

When an output is wrong, do not start over. Run this sequence:

• Re-state the goal in one sentence
  What decision or outcome does this need to enable?
• Lock the format
  Email, brief, table, slides, action list.
• Add constraints
  Length, tone, must-include, must-avoid, what not to assume.
• Force verification
  Facts vs assumptions, do not guess, cite sources where possible.

Here is a reusable fix prompt:
Rewrite this with the same content, but:

• audience: [role]
• goal: [outcome]
• format: [structure]
• length: [limit]
• constraints: [must include / must avoid]
• accuracy: do not guess; flag unknowns

Why this matters for Copilot adoption

The skill gap is not knowing where the Copilot button is.

The skill gap is knowing how to:

• brief it like a professional
• verify it like a manager
• shape outputs into decisions and action

That is the difference between occasional novelty and consistent productivity.