A practical system for writing a PIP with AI that a real employee can actually follow and a real HR team can actually stand behind, built from watching where these plans succeed and where they quietly fall apart.
Somewhere between 41% and half of employees on a performance improvement plan go on to meet its goals and keep their job, depending on whose research you read and how the plan was actually run. The largest factor in that outcome, based on what I’ve seen training managers through this process, is whether the manager stayed genuinely engaged through the plan rather than treating it as paperwork on the way to a decision already made. This guide covers using AI to draft specific, measurable PIP goals instead of vague ones, structuring a realistic 30 to 90 day timeline, building in real support rather than just consequences, and the parts of the process AI cannot do for you no matter how good the prompt is.
In the manager training sessions I run, I ask a version of the same question every time a PIP comes up: how many of the scheduled check-ins actually happened? The honest answer is almost never all of them. That’s the pattern I see constantly: the document was never the hard part. Most PIPs that fail have perfectly clear goals on paper. The manager who wrote it just stopped showing up somewhere around week two.
The numbers on this are genuinely mixed depending on which research you check, and I think that’s worth being upfront about rather than picking whichever stat sounds cleanest. Some HR sources put successful PIP completion around 41%. Others, looking specifically at well-designed plans with real manager engagement, put recovery rates closer to 47 to 74%. That gap, in my view, reflects the difference between a PIP that was actually run and one that existed mostly to create a paper trail.
AI changes the drafting part of this equation completely. It can turn a rough sense of “this isn’t working” into a specific, measurable, professionally worded plan in minutes instead of an afternoon. What it can’t change is the part that research keeps pointing to as the real predictor of outcome: whether the manager stays engaged, coaches through the plan, and actually wants the employee to succeed rather than just documenting that they didn’t.
Before you open ChatGPT to draft anything, answer this honestly: has a decision to let this person go already been made? If yes, say so internally and loop in HR now. Writing a PIP for someone you’ve already decided to fire isn’t a performance plan, it’s a liability, and no amount of AI polish changes that.
A generic prompt gets you a generic PIP, and a generic PIP is close to useless. “Write a performance improvement plan for an underperforming employee” produces something that could apply to almost anyone, which means it applies meaningfully to no one. The fix is giving AI the specific, sometimes uncomfortable, detail of what’s actually gone wrong.
Try something closer to this: “Write a 60-day performance improvement plan for a [role] whose specific performance gaps are: missed three client deadlines in the last quarter, average email response time of 3+ days versus a team standard of 24 hours, and two instances of incomplete deliverables flagged by the client directly. Include 3 to 4 measurable goals tied to these specific gaps, a weekly check-in structure, and available support resources. Tone should be direct and specific, not vague or overly soft.”
Notice what that prompt does: it hands AI the actual facts instead of asking it to invent generic ones. The output will name the real behaviors and set goals against them, like “respond to all client emails within 24 business hours, measured weekly” instead of something unmeasurable like “improve communication.” Vague PIPs are effectively unenforceable, because neither the employee nor the manager can point to a clear yes or no at the end of it.
Read the goals out loud and ask: could I, as the manager, say definitively in 60 days whether this was met? If the answer requires a judgment call rather than a clear yes or no, rewrite it before the plan goes anywhere near the employee.
The single most common mistake I see is accepting the first AI draft wholesale, simply because it reads professionally, not because the prompt was bad. AI-generated PIPs tend to sound polished by default, and polish is easy to mistake for correctness. A well-formatted plan with the wrong tone, unrealistic goals, or language that doesn’t match how your company actually talks can do more damage than a rough one, because it reads as though it was carefully considered when it wasn’t.
Run every AI draft past someone who wasn’t in the room when you wrote the prompt, ideally an HR partner, before it goes to the employee. They’ll catch things you won’t: a goal that’s technically measurable but practically unfair given the employee’s actual workload, a tone that reads harsher in writing than you intended, or a support commitment that sounds generous on paper but nobody on the team actually has the bandwidth to deliver.
Most PIP guidance converges on a 30 to 90 day range, but picking the number inside that range is where a lot of managers default to whatever their HR template says instead of thinking about the actual issue.
Ask AI to build the check-in cadence around whichever timeline you pick, not the other way around. A 30-day plan needs weekly check-ins at minimum. A 90-day plan can run biweekly early on and weekly toward the end, once you’re closer to the actual decision point.
A narrow, specific, fast-to-verify issue: missed deadlines on one task type, a skipped compliance step, a clear attendance pattern.
Most behavior or quality gaps: communication habits, client-facing work quality, collaboration issues needing a few full cycles to show change.
A genuinely new skill, a role transition, or a gap tangled up with a recent change like a new manager or new tools.
Timeline ranges as discussed above, matched to the nature of the performance gap.
Don’t let a shorter timeline function as a way to fail someone faster. If you’re picking 30 days because you privately expect this to end in termination regardless, you’ve already answered the good-faith question from the last section, and the honest move is to stop calling it a PIP.
A PIP that’s all consequences and no support reads exactly like what it usually is: a formality on the way to a decision that’s already been made. AI is actually useful here, because it’s good at generating a concrete list of support options once you tell it what’s realistic for your team and budget.
If this PIP followed a formal review cycle, How to Use AI for Performance Reviews covers the step before this one: getting the underlying feedback specific and fair in the first place, which makes everything downstream easier to write.
This is the section most guides skip, and it’s the one that matters most. A PIP written in bad faith, meaning the outcome was decided before the plan started, creates real legal and cultural risk regardless of how well-written the document is. AI has no way to know whether you’re using it in good faith. Only you know that.
If you’re using ChatGPT or Claude directly rather than an HR platform, strip identifying details you’re not comfortable leaving a general AI tool’s servers, the same caution that applies to any HR data. How to Use AI Without Leaking Company Data is worth a read before you paste anything sensitive in.
Every manager I’ve coached through a PIP has underestimated the same thing. Writing the document takes an hour with AI helping. Sitting across from someone and actually saying the hard parts out loud takes a completely different kind of preparation, and most people skip it because the document feels like the finished task.
Ask AI to draft talking points for the delivery conversation itself: how to open it, how to state the performance gaps factually without piling on, how to leave room for the employee to respond, and how to close with a clear next step. A prompt like “Draft talking points for delivering this PIP to the employee in person. Include how to open the conversation, how to state each gap factually and respectfully, how to invite their perspective, and how to end with clear next steps” produces something genuinely useful to rehearse against.
ChatGPT Prompts for Managers has a broader set of prompts for exactly this kind of difficult conversation prep, if this is a skill you’re building generally rather than just for this one situation.
Here’s the honest ending to this guide. Everything above gets you a strong document in a fraction of the time it used to take. None of it determines whether the PIP actually works. That comes down to whether the scheduled check-ins happen, on time, with real substance, every single week the plan is active.
Manager engagement predicts whether a PIP works better than anything about the employee does, in my experience running these programs. If you already know you won’t have the bandwidth for six weekly check-ins over the next two months, solve that problem before you send the plan. Solving it after the plan has already failed is too late to help anyone.
You can generate a strong first draft that way, but you shouldn’t send it without editing. AI doesn’t know your specific documented incidents unless you provide them, doesn’t know what support is realistically available on your team, and can’t judge whether the plan is being issued in good faith. Treat the AI output as a fast first draft that a human, ideally with HR input, reviews and personalizes before it goes anywhere near the employee.
Most effective PIPs run 30 to 90 days depending on the nature of the gap: 30 days for a narrow, fast-to-verify issue, 60 for most behavior or quality gaps, and 90 for something involving a genuinely new skill or a role transition. Ask AI to build the specific check-in cadence around whichever length actually fits the problem, rather than defaulting to whatever a generic template suggests.
Give the AI an explicit tone instruction in your prompt: something like “direct and specific, not vague or overly soft, but respectful and focused on support alongside expectations.” Then read the output yourself and check that every goal is stated as a fact (missed X deadlines, average response time of Y) rather than a judgment (“unreliable,” “unprofessional”). Facts hold up under scrutiny. Judgments invite pushback and rarely change behavior.
The AI drafting itself isn’t the legal risk. The risk is the same one that exists with human-written PIPs: a plan issued in bad faith, with vague or unmeasurable goals, or without real support, can create exposure regardless of who wrote the words. Loop in HR before sending any PIP, keep separate documentation of the specific incidents behind it, and avoid pasting identifying employee details into a general-purpose AI tool that isn’t part of your company’s approved HR systems.
Make the decision the plan was actually built around, clearly and on schedule. If the goals were specific and measurable, as they should have been from the drafting stage, you should be able to state plainly whether each one was met. Loop in HR before the final conversation, document the outcome against each specific goal, and avoid extending the plan indefinitely just because the decision feels difficult. An extension without a real reason is its own signal to the employee that the goals were never truly fixed.
I researched current PIP completion and success-rate data from multiple HR research sources, including PerformYard, AIHR, and HiBob’s published guidance, and cross-checked the finding on manager engagement as the leading predictor of outcome against several independent HR practitioner sources rather than relying on a single figure. The prompt structures below reflect current guidance from HR-focused prompt libraries including Lattice and PeopleManagingPeople.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.