Decisions prompt

Pre-Mortem Prompt — Find Why It’ll Fail Before You Start

Post-mortems are autopsies — useful but too late. Pre-mortems are simulated autopsies before the work starts: assume the project failed, then ask what went wrong and what you could’ve prevented. This prompt walks through the exercise in 5 minutes and ends with one specific thing to do this week.

✓ Works in ChatGPT, Claude, Gemini ✓ Tested template ✓ Copy + paste ready

The template

Replace the [bracketed] fields with your specifics. Send to any AI chat.

I'm about to [decision or project — describe in 2-3 sentences].

Context I'm working with:
- My situation: [budget, team size, deadline, anything material]
- Why I'm doing this: [the real reason, not the post-rationalized one]
- What I'm hoping success looks like: [specific outcome]

Imagine it's 6 months from now and it failed badly. Write the post-mortem:

1. What went wrong (top 5 most-likely failure modes, ranked by probability)
2. What were the early warning signs (one per failure mode)
3. What could I have done differently in week 1 to prevent each
4. What's the ONE specific thing I should do THIS WEEK to reduce the biggest risk

Be willing to point out things I probably don't want to hear.

How to use it

Be honest about the ‘why’ — including the unflattering version. ‘I’m doing this because everyone else is’ is a useful pre-mortem input even if you wouldn’t say it out loud.
Read each failure mode and ask ‘do I disagree?’ If you can’t articulate a counter-argument in 30 seconds, take the failure seriously.
Actually do the one this-week action. The whole exercise is wasted if you read the pre-mortem and don’t change behavior.

Example output

Here’s an example of what a quality AI returns when you run this template:

**The post-mortem (6 months from now):** heyLooAI failed to break 5,000 monthly visitors. Here’s why: 1. **(40%) Asmat couldn’t sustain the publishing rhythm.** Solo-operator content sites consistently die at the 6-week mark when motivation flags and SEO hasn’t kicked in yet. Early warning: a week with zero posts published. Could’ve prevented: built a 2-week buffer of drafts before launch. 2. **(25%) The articles ranked but for low-intent queries.** Articles got first-page rankings for “comparison” queries where the searchers were doing research, not buying. Affiliate conversion stayed near zero. Early warning: high impressions, low click-through, low time-on-page. Could’ve prevented: prioritized “best X for Y” queries (high commercial intent) over “X vs Y” queries (research intent) in week 1. 3. **(15%) The prompt generator drove traffic but no signups.** People used the tool and bounced. The newsletter form was buried below the fold and the value prop wasn’t clear. Early warning: traffic to /tools/ way exceeds newsletter signups for 4 weeks running. Could’ve prevented: A/B test the form placement in week 2. 4. **(15%) Google’s next core update down-ranked AI-niche sites broadly.** Site got caught in a topical update that hit similar sites. Hard to prevent — the only insurance is content depth and original perspective. Could’ve prevented: more original screenshots and proprietary data per article. 5. **(5%) A bigger competitor entered the niche and outranked you.** Quillbot or similar launched a directly competing tool. Early warning: their pages start appearing in your tracked-keyword results. Could’ve prevented: nothing structural — only execution speed. **THIS WEEK: build the 2-week content buffer.** The top failure is the one with both highest probability AND highest agency. Draft 4 articles now, before launch energy fades.

Why this prompt works

The ‘imagine it failed’ frame breaks optimism bias.

Asking ‘what could go wrong’ produces shallow lists. Asking ‘imagine it DID fail and write the post-mortem’ forces the model into specific failure scenarios that pattern-match real failure modes.

Probability-ranked failure modes prevent paranoia paralysis.

Without ranking, you get 10 equally-weighted risks and no priority. Probabilities force a judgment about which to actually address first.

Early-warning signs make the failures detectable.

A failure mode you can’t detect is one you can’t prevent. The early-warning section converts each risk into a metric you can watch for.

One specific this-week action prevents ‘I’ll do a project plan’ inertia.

The exercise’s value is the behavior change. Asking for one action in one timeframe makes the change concrete.

Which AI to use

**Claude** for the qualitative depth — its pre-mortems tend to surface specific failure modes other models miss. **ChatGPT** for project-management heavy plans with timelines. **Gemini** for plans involving data you have in Sheets you’d want it to reference.

Read the full comparison in ChatGPT vs Claude vs Gemini in 2026 →

Frequently asked questions

Why pre-mortem and not just risk planning?

Risk planning produces lists of risks. Pre-mortems produce stories — and stories are more actionable because you can spot the early-warning signs in your own behavior. Stanford’s Hal Hershfield has done research on this; the imaginative frame matters.

How early should I run a pre-mortem?

Before the plan is committed. Once you’ve publicly committed to a direction, motivated reasoning kicks in and you’ll dismiss legitimate failure modes. Pre-mortem in the deciding window.

What if the failure modes seem obvious?

Often the obvious failures are the ones that actually kill projects. Smart people predict the obvious failures and then don’t do anything about them. The ‘one this-week action’ is the prevention.

Should I share the pre-mortem with my team?

Yes — written pre-mortems shared with the team produce better team accountability than the same risks discussed verbally. Make the early-warning signs the team’s KPIs.

How often should I redo it for the same project?

Once at the start, again at the 30-day mark, again at the halfway point. Each one will surface different failure modes as the project evolves.

Build any prompt in 30 seconds

Free tool, no signup, runs in your browser. Pick a category, fill in the blanks, copy your prompt.

Try the prompt generator → Read the full guide