How to Use My ACH (Analysis of Competing Hypotheses) GPT — The Smartest Way to Stress-Test Your Case

Dec 31

If you’ve ever worked a case where the facts are messy, the sources are contradictory, or the stakes are high, you’ve probably felt the pull of the most dangerous analytic shortcut: locking onto one “likely” explanation early and then finding evidence to support it.

That's how investigations go sideways.

And that’s why intelligence analysts, especially those trained in structured analytic techniques, lean on ACH: Analysis of Competing Hypotheses, a method designed to fight confirmation bias and force disciplined reasoning. ACH is strongly associated with intelligence tradecraft and was popularized through CIA analytic training and writing, including the Center for the Study of Intelligence and methodologies connected to the CIA’s “Kent School” tradition of analytic rigor.

I built a custom ACH GPT to help anyone apply this method quickly and consistently, while keeping the most crucial part where it belongs: your own vetted evidence.

Here’s how ACH works, why it’s useful across nearly any domain, and how to get the best results using my publicly available ACH assistant.

Use the GPT here.

What Is ACH (Analysis of Competing Hypotheses)?

ACH is a structured analytical technique that helps you evaluate multiple explanations simultaneously, rather than committing to one narrative and attempting to prove it.

The method was developed by CIA analyst Richard J. Heuer Jr. and is described in detail in "The Psychology of Intelligence Analysis" (CIA/CSI). Heuer’s goal was simple: help analysts reason more accurately under uncertainty by reducing the impact of cognitive biases.

ACH does this by shifting your mindset from:

“Which hypothesis can I prove?”

to:

“Which hypothesis survives the evidence best after I try to disprove them all?”

This is why ACH is so powerful: it is designed to emphasize disconfirming evidence, the kind of evidence we naturally avoid or undervalue.

The ACH Matrix: The Core of the Method

ACH is famous for its matrix, which forces you to test every piece of evidence against every hypothesis.

In an ACH matrix:

Rows and columns are your pieces of evidence (or arguments) and your competing hypotheses
Each cell captures whether that evidence is:
- Consistent with the hypothesis
- Inconsistent with the hypothesis
- Not applicable/unclear

Then the logic kicks in:

✅ You don’t “score” what supports a hypothesis.

You primarily track what contradicts it.

The hypothesis with the fewest meaningful inconsistencies rises to the top. The ones with the most serious inconsistencies move down the list (or get eliminated).

That’s the key: ACH is less about proving the “right” hypothesis and more about systematically rejecting the wrong ones.

Why the Kent School Tradition Matters (and Why You Should Care)

The CIA’s Kent School tradition emphasizes one thing: analysis must be disciplined, transparent, and defendable, not just “what feels right.”

Whether you’re doing national security work, investigative journalism, OSINT, competitive intelligence, or law enforcement analysis, your credibility rests on whether someone else can look at your reasoning and say, “I see how you arrived at that conclusion. I may disagree, but it’s auditable.”

ACH helps produce exactly that: a visible trail of reasoning, organized evidence, and explicit alternatives that were considered and eliminated.

ACH Works Far Beyond Intelligence

ACH is often associated with intelligence analysis, but it’s valuable anywhere you have competing explanations and incomplete information. Here are a few examples:

National Security/Intelligence

Attribution of events
Assessing intent vs capability
Evaluating deception or denial operations
Predicting likely outcomes based on indicators

Law Enforcement/Investigations

Competing suspect theories
Behavioral hypotheses (motive/opportunity)
Case linkage questions
Fraud/cyber investigations

Journalism/Investigative Reporting

Source credibility conflicts
Competing narratives
Fact-checking and verification
Building a defensible “most likely explanation”

Business/Corporate/Risk

Root-cause analysis
Competitive moves
Internal incident analysis
Reputation and threat assessments

ACH shines when:

the story is noisy
humans are emotional
data is incomplete
and the consequences of being wrong are real

Why I Built an ACH GPT (and What It Does)

ACH is excellent… but it’s also time-consuming. Even experienced analysts can struggle with:

brainstorming hypotheses broadly enough
writing evidence clearly and consistently
building the matrix correctly
maintaining discipline (and not “steering” the result)

So I built a custom GPT that guides users through the method and automates the matrix-building portion so you can spend your time doing what matters:

✅ vetting evidence
✅ collecting better evidence
✅ thinking clearly
✅ challenging assumptions

The tool is publicly available for anyone to use.

The Key to Great Results: Use Your Own Vetted Evidence

This is the most important part of this entire post: The GPT does not magically make bad evidence good. ACH is only as strong as the inputs.

If you feed the assistant:

rumors
unverified screenshots
unsupported claims
cherry-picked sources

…it will still produce an output, but it won’t be a reliable analysis.

To get powerful results, you want evidence that is:

sourced
verified (or clearly labeled as unverified)
dated/contextualized
separated from your assumptions or interpretations

ACH works best when evidence is treated as evidence, not as conclusions.

How to Use the ACH GPT (Step-by-Step)

Step 1) Describe your case/question

Begin with your analytical question in clear, yet detailed language.

Examples:

Based on the technical indicators, targeting pattern, and geopolitical context, which actor is most likely responsible for the breach of defense contractor X’s network?
Is the suspect’s online persona a genuine individual, an alias used by a single offender, or a coordinated sockpuppet network?
Was the incident at facility X during a public event accidental, negligent, or intentional?
Which explanation best accounts for company X’s financial anomalies, and are they more consistent with legitimate operational issues, accounting errors, aggressive revenue recognition, internal fraud, or an external compromise?

No special format required.

Step 2) Let the GPT generate hypotheses

Ideally, the GPT will help you identify:

obvious hypotheses
less obvious alternatives
edge cases
deception/manipulation possibilities
“null hypotheses” (e.g., coincidence, error, benign explanation)

The goal here is breadth, not certainty.

Step 3) Choose the hypotheses you want to evaluate

ACH works best when you narrow to a manageable set:

usually 3–7 hypotheses
enough to cover realistic possibilities
not so many that the analysis becomes unusable

The GPT can help you refine them so they’re mutually exclusive and clearly defined.

Step 4) Provide your evidence (no special format needed)

This part is intentionally easy. If the GPT does not prompt you on the types of evidence you will need, ask because it will be very thorough and likely suggest types of evidence that have not yet been collected.

You can:

upload a file (or multiple files)
paste your evidence directly
provide links, excerpts, notes, quotes, logs, timelines, or source summaries

Clearly label it:

“The evidence”

That’s it.

The GPT will:

extract the evidence
separate distinct evidence items
organize them into matrix rows
test each item across your hypotheses

Step 5) Review the matrix + rankings

Once the evidence is processed, the GPT generates:

the ACH matrix
the hypotheses ranked by inconsistency (least inconsistent = most likely)
notes on which evidence is most diagnostic
explicit reasoning for the ranking

This is where ACH becomes incredibly valuable: it creates a clear structure for both you and anyone reviewing your work.

Step 6) Ask: “What evidence do I still need?”

This is one of the best uses of the tool.

Once the GPT builds the matrix, ask:

What evidence would most strongly discriminate between the top 2 hypotheses?
What gaps weaken the analysis?
What evidence should I prioritize collecting next?
What would falsify the leading hypothesis?

This transforms your investigation into a targeted collection plan, rather than random searching.

Example Case

Was the incident at the Volyn accidental, negligent, or intentional, and what specific evidence would support each conclusion while minimizing speculation and legal risk?

Hypotheses (limited to most likely):

H1: Intentional Russian strike on infrastructure.
H2: Miguided/off-course Russian drone (accidental).
H4: Ukrainian air defense caused the damage (intercept effect)
H7: False-flag operation by Russia or proxy actors
H9: No strike occurred; damage/disruption is disinformation or misreporting

Evidence ACH GPT requests:

H1 – Intentional Russian Strike

What you’re testing: Whether Russian forces deliberately targeted Volyn infrastructure as part of a strategic or tactical campaign.