Giving good context · for customer support professionals

Giving good context in customer support

A model knows how support replies usually sound. It does not know your refund policy, your plan names, or the outage your team posted about this morning. It has not read the three earlier messages in this thread. Without that material it writes a reply for an imaginary company.

Giving good context means pasting in what a new agent would need before answering. That includes the relevant part of the thread, the current policy text, the product facts, and the team’s tone guide. It also means leaving out material that would mislead, such as last year’s pricing.

Where it comes up

  • A reply to a customer on their fourth message

    Paste the earlier messages in the thread, with personal details removed. Without them, the model may repeat a fix the customer already tried. Repeating a failed fix makes an upset customer more upset.

  • A reply that depends on policy

    Paste the current policy text from your internal knowledge base, with its last-updated date. Do not describe the policy from memory. The model fills gaps with policies that sound plausible and are wrong for your company.

  • A help-center article for a new feature

    Provide the release notes, the exact names of menus and buttons, and the plans that include the feature. Steps written from the model’s general knowledge will name screens your product does not have.

What it looks like at each band

  1. E1 · Foundational

    Pastes the customer’s latest message and asks for a reply. Output is generic and sometimes contradicts the earlier thread or the policy. Does not yet see the missing material as the cause.

  2. E2 · Proficient

    Includes the relevant thread, the current policy text, and the product facts the reply depends on. Output matches the account and the policy. Pastes source material rather than paraphrasing it from memory.

  3. E3 · Distinguished

    Chooses context for each ticket on purpose. Leaves out outdated macros and old pricing that would confuse the model. Marks which facts are confirmed and which are the customer’s claims. Keeps a current set of policy excerpts ready to paste.

  4. E4 · Exceptional

    Builds and maintains the shared context the team uses: policy excerpts written for AI use, a tone guide with approved examples, and current product facts. Notices when a policy or feature changes and updates that material before agents send wrong answers.

Common mistakes

  • Paraphrasing the policy

    A policy described from memory loses its exceptions and its time limits. Paste the actual text, and note the date it was last updated.

  • Pasting the whole account history

    Two years of tickets bury the three messages that matter for this reply. Choose the part of the history that bears on the current problem.

  • Treating the customer’s account of events as fact

    The customer may be right, or may be describing a different plan or feature. Tell the model which statements the account records confirm.

A worked example

A reply to a customer who says a feature stopped working after an update.

A weaker attempt
The customer says exporting to CSV is broken since the update. Write a reply.

The model does not know whether this is a known issue, whether a workaround exists, or what the customer already tried. It will suggest clearing the cache and restarting, which may waste the customer’s time.

A stronger attempt
Write a chat reply. Context: engineering confirmed this morning that CSV export fails for files over 10,000 rows since version 4.2. A fix is planned but has no date. The workaround is to filter the report to fewer rows and export in parts. The customer already tried a different browser. Here is the status page wording we are using: [text]. Do not give a fix date. Under 80 words.

The model has the confirmed cause, the workaround, what the customer tried, the approved wording, and a limit. The reply can help the customer today instead of sending them through steps that will not work.

The full study guide for this skill, for every role, is Giving good context. The other five skills for customer support professionals are on the AI skills for customer support professionals page. A one-page summary is in the AI context checklist.

See where you stand

The free check grades giving good context and the other five skills in about fifteen minutes. Nobody sees the grade but you.

Then build the skills and prove it. The three steps for customer support professionals list the courses and templates that fit the work.