Fixing Incomplete or Shallow Responses

The Model Covered the Topic. It Just Didn't Go Deep Enough. Here's Why.

There is a specific frustration that comes from AI output that addresses everything you asked about but leaves you no better informed than when you started. The response touched the right subjects, used the right terms, and arrived at plausible-sounding conclusions — but all of it could have been produced from a basic understanding of the topic, with no real analysis, no engagement with the specific challenge you described, and no depth that would actually help you make a decision or take action.

This is shallow output, and it is different from incomplete output. Incomplete output left something out. Shallow output covered everything you asked about but at a surface level. The root causes differ, and so do the fixes.

What This Failure Looks Like

Incomplete and shallow outputs have distinct symptoms that are worth separating:

Incomplete output looks like: a list that stops at three items when you needed five, a response that covered the first two topics in your brief but ignored the third, an analysis that addressed the opportunity but said nothing about the risks, a summary that excluded a key theme from the source material.

Shallow output looks like: a response where every section is one paragraph when you needed two, where each point is stated and then not developed, where tradeoffs are named but not explained, where the conclusion follows from the topic but does not follow from any actual analysis. The right subjects are present; they are just not developed to the level of usefulness.

The distinction matters because the fix for incomplete output is scope specification (name what was missing), while the fix for shallow output is depth instruction (tell the model how deep to go).

What Caused It

Both failure modes trace back to missing instructions, but different ones.

Incomplete output is caused by scope ambiguity. The model addressed what seemed like the central request and either did not see or did not prioritize the additional elements you needed. This is most common in prompts where multiple requirements are listed without being signaled as equally important — the model satisfies the first requirement clearly and attenuates on the later ones.

Shallow output is caused by missing depth instructions. The model's default optimization when asked to cover a topic is breadth — address all the relevant aspects at a reasonable level of generality. This is the appropriate default for many tasks. When you need analytical depth rather than topical breadth, the prompt needs to signal that explicitly. Without a depth instruction, the model does not know whether you want an overview or an analysis, a list of factors or an explanation of how they interact.

The Fix

The fix for incompleteness is explicit scope: name every element you need covered and signal that all of them are required, not optional.

Scope-ambiguous prompt: "Analyze the risks of launching a new product in a saturated market."

Scope-explicit prompt: "Analyze the risks of launching a new product in a saturated market. Cover all four of these areas: competitive pricing pressure, customer acquisition cost, differentiation sustainability, and channel access. Do not summarize — analyze each one specifically for a bootstrapped company with no marketing budget."

The fix for shallow output is a depth instruction that tells the model what level of analysis you need:

No depth instruction: "Explain the tradeoffs between building in-house versus buying off-the-shelf software for a 200-person company."

With depth instruction: "Explain the tradeoffs between building in-house versus buying off-the-shelf software for a 200-person company. For each tradeoff, explain the mechanism — not just what the tradeoff is but why it plays out the way it does, and what specific conditions at this company size make it more or less significant. I need enough depth to present a recommendation to a non-technical executive."

Depth instructions often include a purpose clause: "enough depth to present a recommendation," "detailed enough to build a decision framework," "specific enough that a new team member could act on it." Purpose clauses give the model a calibration target for depth rather than an abstract instruction like "be thorough," which is interpreted as loosely as any other abstract adjective.

When This Fix Doesn't Work

Depth and scope instructions do not resolve all incompleteness or shallowness. Three cases where additional approaches are needed:

When the model lacks sufficient knowledge to go deep: Depth instructions tell the model to produce deeper analysis — they do not give it knowledge it does not have. In specialized domains or for recent events, shallow output may reflect genuine knowledge limits rather than a missing instruction. Check whether the shallowness is uniform across the topic or concentrated in specific areas that might fall outside reliable training data.

When depth conflicts with length: A depth instruction and a length constraint can create tension. Asking for deep analysis of five factors within 300 words forces the model to choose. Resolve this explicitly: either remove the length constraint, reduce the scope (fewer factors, deeper on each), or specify which constraint takes priority.

When the task needs iteration, not just depth: Some analytical tasks produce genuinely better output through iteration — a first pass that generates the framework, a second that applies it, a third that addresses gaps. For complex analytical tasks, a single prompt with a depth instruction may not be enough. Breaking the task into turns and using each output as input to the next often produces more rigorous analysis than trying to get everything in one shot.