How we surface the hidden inferences, defaults, and interpretations that shape AI’s choices

_{—Part of CSS Primer for the AI Era — CSS Communication Toolkit}

Revealing assumptions is how
we expose the invisible structure
behind the output.

AI does not wait for 'perfect clarity'.
When something is missing, ambiguous, or underspecified,
AI fills the gap with an assumption:

• about intention
• about structure
• about tone
• about hierarchy
• about behavior
• about constraints

These assumptions are rarely visible
unless we ask for them.

So, the next communication skill is simple:

We ask AI to reveal the assumptions behind its choices
so we can correct the reasoning, not just the output.

Assumptions are the quiet source of drift.
Revealing them is how we restore alignment.

Mistakes are easy to fix.
Assumptions are harder
— because they are invisible.

When AI misreads us,
it is usually because it assumed:

• a different intention
• a different layout pattern
• a different hierarchy
• a different constraint
• a different tone
• a different risk tolerance

If we do not reveal the assumption,
the mistake will repeat.

So, we bring the assumption into the open.

Every assumption falls into one of three
categories:

1. Assumptions about intention
   What AI thinks we are trying to achieve.
2. Assumptions about structure
   How AI interprets the architecture or layout.
3. Assumptions about constraints
   What AI believes it is allowed to change.

When we name the category,
AI knows what to reveal.

Assumptions About Intention.

Examples of what to say:

“Explain the intention you assumed 
when making this change.”

“Describe what you thought the goal was.”

“Explain the purpose you inferred from the request.”

Examples of what to avoid saying:

“Why did you do this.”

We ask about the assumed intention,
not the action.

Assumptions About Structure.

Examples of what to say:

“Explain the structural assumptions 
behind this layout.”

“Describe how you interpreted the 
parent/child relationships.”

“Explain what you assumed about the 
container or flow.”

“Describe the hierarchy you inferred.”

Examples of what to avoid saying:

“Why is the layout wrong.”

We ask about the model of the structure,
not the correctness.

Assumptions About Constraints.

Examples of what to say:

“Explain what you assumed you were allowed to 
change.”

“Describe the boundaries you inferred.”

“Explain what you believed was flexible 
and what was fixed.”

Examples of what to avoid saying:

“Why did you change that.”

We ask about the assumed permissions,
not the decision.

We define the scope:

“Reveal only the assumptions about spacing.”

“Reveal only the assumptions about tone.”

“Reveal only the assumptions about structure.”

“Reveal only the assumptions about responsiveness.”

A simple principle:

Assumptions become useful
when they are isolated.

AI sometimes tries to defend its choices.
We prevent this by saying:

“Reveal the assumptions, 
not the justification.”

“Describe what you inferred, 
not why it is correct.”

“Explain the assumptions without 
recommending an option.”

This keeps the collaboration analytical,
not defensive.

Once AI reveals its assumptions,
we can respond with clarity:

“The assumed intention is incorrect 
— here is the real one.”

“The structural assumption is wrong 
— here is the actual hierarchy.”

“The constraint assumption is incorrect 
— this part must remain unchanged.”

“The tone assumption is off 
— we want quiet, not expressive.”

We correct the assumption,
and the output corrects itself.

A complete request looks like this:

“Explain the assumptions you made about the layout.
Do not change anything yet.
Reveal only the structural inferences.”

Or:

“Describe the assumptions you made about tone and hierarchy.
Do not recommend an option.
Reveal the reasoning only.”

Or:

“Explain what you assumed you were allowed to change.
Focus only on constraints.
Do not modify the code.”

This is how we expose the invisible logic.

When we ask AI to reveal its assumptions:

• the reasoning becomes visible
• the drift becomes correctable
• the constraints become respected
• the intention becomes aligned
• and the collaboration becomes more predictable

A simple closing thought:

Assumptions shape the work.
We reveal them,
and the collaboration becomes clear.

• Suggested Next Reading: How to Ask AI to Identify Missing Information

_{Tony de Araujo —New York | Lisbon}

• Top↑
• Back to Topic ↔
• Home←