How to use AI effectively in your OKR Setting?

AI can do a lot for your OKR process, but it won’t save you from a bad prompt. The quality of what you get out depends almost entirely on what you put in.

You could ask ChatGPT something like “Create an objective and 3 key results for my product team.” But you shouldn’t. That’s not a well-written prompt, and you’ll get a generic OKR to match. This guide will show you how to do it properly.

One thing AI should never do for you: define your objective. That conversation needs to happen with your team. The real value in OKR setting is in the discussion, the trade-offs, and the shared ownership that comes from deciding together what the one or two most important things are this quarter. Don’t outsource that.

Where AI earns its keep is in the key results. It can help you find the right metrics, sharpen the framing, and sense-check whether you’re measuring outcomes or just activity. That’s where good prompting makes the difference.

How to write a good AI prompt for OKR

Think of your prompt as a set of building blocks. You don’t need all of them every time, but the more relevant context you give, the better the output. Most prompts only need the first three or four.

• Description. Tell AI exactly what you want it to produce. “I’m looking to brainstorm a list of methods for gathering customer feedback” is clearer than “help me with feedback.”
• Role. Ask AI to take on a specific role to frame the response. “Act as a Product Manager launching a new program” will produce a different result than a generic ask.
• Context. Be specific about what you’re working on. “I want this key result to focus on our Q2 sales target. My objective is X” gives AI something real to work with.
• Specific requirements. Include metrics you want to hit, the tone you want, any documents or data worth feeding in.
• Boundaries. Tell it what to avoid. If you don’t want NPS as a metric, say so.
• Reasoning. Explain the thinking behind the criteria you want AI to follow. It produces better output when it understands why, not just what.

AI writes better OKRs when you give it a real problem to solve, not just a team name and a job title. If you have strategy documentation, feed it in.

Give AI your OKR guidelines once

The other thing worth doing is giving AI a clear OKR writing standard. You only have to do it once, and it will significantly improve output consistency every time you use it.

Here’s the guideline set to copy in:

Objective

Write a single inspiring, specific, and time-bound statement that reflects the biggest challenge you need to solve or your main priority this quarter. It should align with your strategy, push the team toward something meaningful, and read like a newspaper headline that grabs attention. Action-oriented, focused on change, not maintenance.

Key Results

Provide 2 to 5 measurable Key Results that show how you’ll track progress toward the Objective. Each one needs a starting metric (baseline) and a target metric (desired outcome by end of quarter). Key Results must be outcome-focused, not tasks or activities. Instead of “launch a new feature”, write “increase daily active users from 10,000 to 15,000.” Use leading indicators your team can influence this quarter.

Initiatives

Suggest 3 to 5 initiatives or projects the team can work on to support the Key Results. These are specific actions, but they should not be measured as Key Results. They’re the work you think will move the metrics.

Clarity and alignment

Make sure the OKR connects directly to your overall strategy or vision. Each Key Result should contribute to achieving the Objective. Aim for ambitious but achievable: around 70% completion if you really stretch.

Three prompt examples

Example 1: Writing key results from scratch

The objective is around increasing customer retention by providing the best experience possible.

Your prompt:

“Act as an expert OKR Coach and help me, a Customer Support Manager, define measurable key results for my Q4 objective of increasing customer retention. Focus on metrics that can be tracked in real time. Avoid suggesting metrics like social media followers or NPS score. Explain why each key result is relevant.”

Example 2: Brainstorming leading indicators

Ask AI to generate 10 well-established leading indicators for your objective, then use that list as a discussion point with your team. It saves time and keeps your team in the decision, which is where they should be.

Example 3: Refining an existing key result

“Act as an OKR Coach and give me feedback on the following key result. The objective focuses on providing an amazing customer experience.

Key Result 1: Increase customer retention rate to 85% by the end of Q4”

You’ll get specific suggestions back: establish a baseline, check achievability, assess the link to the objective. Start simple and refine from there.

Using AI to write better OKRs

AI can help you write sharper OKRs, but only if you give it the right context. A generic prompt gets a generic OKR. The output is only as good as the input.

To make this easier, we built Telos, an AI OKR coaching agent trained on the OKR Quickstart methodology. You give it your team’s situation, the problem you’re trying to solve, and your current metrics. It helps you move from vague objectives and activity-based key results to OKRs that actually measure outcomes. It spots vanity metrics, challenges output thinking, and helps you find the right problem before you commit to a goal.

How to run Telos in Claude, ChatGPT, or Gemini

The Telos prompt works in any capable AI tool. The best way to run it depends on how often you plan to use it.

If you are trying it once, paste the full prompt at the start of a fresh conversation, then start talking. Tell it what you are working on, the team, and the problem you are trying to solve. That is enough for Telos to get going.

If you are going to use it more than once, save it as a reusable instruction set inside your tool. Every capable AI has a way to do this now. The names differ, the idea is the same.

Claude

Set up a Project in Claude. Paste the Telos prompt into the project instructions field. Every new conversation in that project starts with Telos already loaded, so you can jump straight in.

Drop your team’s strategy doc, your current OKRs, or a snapshot of your metrics into the project’s knowledge and Telos will coach you against your actual context, not a generic one.

The Opus, Sonnet, and Fable model families all run Telos well. I reach for Opus when I am doing a substantive review or a grill session, because the reasoning holds up better on the harder judgement calls.

ChatGPT

Make a Custom GPT. Paste the Telos prompt into the instructions field, give it a name and a short description, and save. It sits in your sidebar from then on. Every chat you start with it has the full prompt loaded automatically.

If you do not want a Custom GPT, paste the prompt at the top of a fresh chat and it will hold for that conversation. Start a new chat when you switch topics so the context stays clean.

Gemini

Create a Gem. Paste the Telos prompt into the instructions field. The Gem sits in the sidebar and stays loaded across sessions, the same way a Custom GPT does in ChatGPT.

If Gems are not available in your workspace, the paste-at-the-start-of-a-chat pattern works fine.

Working with Telos day to day

Once it is loaded, the way you talk to it matters more than which tool you are in. A few things worth knowing.

Lead with the problem, not the goal. “Customers are dropping off in month three, they hit an activation cliff and never come back” gives Telos something to work with. “Write me a customer retention OKR” gives it nothing.

Use the modes. Telos has four. Diagnose, when you are not sure something should even be an OKR (“is this an OKR or BAU?”). Build, when you want a draft (“help me write an OKR for X”). Review, when you have one already and want it critiqued (“here is our OKR, tell me what is off”). Grill, when you want it torn apart line by line (“grill me on this OKR”). Grill is opt-in and it will not run unless you ask.

Push back. Telos is direct on purpose. If it flags something you disagree with, tell it why. It will change its mind if you make the case, and hold the line if you do not. That back-and-forth is where the real coaching happens.

Bring the numbers. Telos will not invent metrics. If you have your dashboard open with the current baseline and the target you are aiming for, drop them in. You will get sharper Key Results.

If you want a refresher on what good OKRs look like before you start prompting, see our ultimate guide to OKR with examples.