How to Write Effective Guidelines
Last updated: September 24, 2026
Guidelines are plain-text rules that Kota follows every time it runs a workflow. Each tab in the Knowledge Base — Connectors, AI Roleplay Bots, Scorecards — has its own Guidelines box.
Rules you save under AI Roleplay Bots shape how every bot Kota creates or edits behaves in a roleplay.
Rules under Scorecards shape how Kota writes and structures scorecards — the scoring model that grades calls never reads them, so they steer what gets drafted, not how a finished scorecard is scored.
Think of guidelines as the standing instructions you would give a new teammate who builds bots and scorecards for you. Kota reads them every time, so anything you find yourself correcting by hand more than once belongs here as a rule.
What makes a good guideline
Most weak guidelines fail the same way: they describe a quality, such as “realistic,” “challenging,” or “fair,” instead of a behavior. Kota cannot act on an adjective. Four principles fix that:
Specific. Name the situation, the behavior, and the threshold. “Push back once” is clearer than “don’t give in too easily.”
Testable. Someone should be able to check compliance with a yes or no — against a roleplay transcript for bot rules, or against the drafted scorecard for scorecard rules.
One behavior per rule. A rule that bundles three behaviors may only get partially followed. Split it into separate rules.
Phrased as observable behavior. Write what the bot says or does, or what a drafted criterion must contain — not what anyone should “understand” or “feel.”
Writing guidelines for AI roleplay bots
Bot guidelines control how the prospect persona behaves: how hard it pushes, what it volunteers, and when it gives ground.
Example 1: Objection behavior
Weak guideline:
Bots should be challenging.
Strong guideline:
Every discovery bot must raise at least one budget objection before agreeing to any next step.
“Challenging” could mean anything from interrupting to stonewalling. The strong version names the objection type, the minimum, and the timing, so every bot Kota builds gives reps guaranteed practice on the objection that matters.
Example 2: Handling the brush-off
Weak guideline:
Bots should not give in too easily.
Strong guideline:
When the rep’s opener is generic, bots should give a “just send me some info” brush-off and push back once before conceding — only engaging if the rep responds with a specific, relevant value statement.
The weak version does not say what “giving in” looks like or how much resistance is enough. The strong version defines the trigger, the bot’s move, and the exit condition. Reps get consistent practice at the moment most cold calls die.
Example 3: Persona realism
Weak guideline:
The prospect should sound realistic.
Strong guidelines:
Prospect bots answer in 1–3 sentences; they do not monologue.
Bots only volunteer a pain point after the rep asks a follow-up discovery question, never in response to the first question.
Bots never state their budget unless the rep asks about it directly.
“Realistic” is the kind of vague adjective that leaves Kota guessing. Broken into three observable rules, it produces prospects that behave like guarded real buyers: short answers, earned disclosures, and no free information.
Writing guidelines for scorecards
Scorecard guidelines shape how Kota writes scorecards. The scoring model that grades calls never reads them, so a guideline cannot directly change how a call is scored. It changes the criteria Kota drafts. If you want a rule to matter at scoring time, the guideline’s job is to make sure it gets written into the criterion itself.
Two kinds of guidelines are especially useful here:
Business or domain context Kota cannot know on its own
Rules about how every criterion should be written
Example 1: Business and domain-specific feedback
Weak guideline:
Feedback should be relevant to our business.
Strong guideline:
Our sales calls run 45–60 minutes and we never book 15-minute follow-ups. Criteria and feedback text must never recommend a short follow-up call.
Kota has no way to know your sales motion unless you tell it. The strong version states the domain fact and its consequence for the scorecard text, so every scorecard Kota drafts gives feedback that fits how your team actually sells instead of generic sales advice.
Example 2: What every criterion must contain
Weak guideline:
Criteria should be clear.
Strong guideline:
Every criterion must include at least one example of what good looks like — either a sample quote or a described behavior a rep could copy.
The weak version leaves “clear” up to interpretation. The strong version enforces a consistent structure: every scorecard Kota drafts comes out with examples baked in, so reps reading their scores always see what to do differently, not just a number.
Example 3: How criteria are phrased
Weak guideline:
Criteria should be fair.
Strong guideline:
Criteria must be phrased as observable rep behaviors with a countable bar, such as “asked at least three open-ended questions before pitching,” never as traits like “was engaging” or “showed curiosity.”
Since the scoring model reads only the criterion text, a criterion phrased as a trait gets scored as a vibe. This guideline makes Kota write the measurable bar into every criterion, which is how a rule ends up influencing scoring.
Already have a playbook? Upload it instead
If your rules already live in a sales playbook, call-review rubric, or objection-handling document, you do not have to retype them:
Click Add files under Guidelines on the relevant tab.
Upload up to 5 documents. Supported formats include PDF, DOCX, TXT, and MD.
Kota reads them and drafts a set of guidelines.
Review the draft, edit anything that is off, then click Save.
Nothing is stored until you save, and the uploaded files stay listed so you can always see what the guidelines were built from. The best inputs are documents that already describe behaviors: playbooks, QA rubrics, persona or ICP one-pagers, and objection-handling cheat sheets.
Still need help? Reach out to support@hyperbound.ai