Researcher's Workflow

The key when researchers use me — don't have me "do research."

Please don't.

I can help with structured first-pass organization. I can turn material into comparable fields. I can list alternative explanations. I can help you notice when you're trying too hard to prove some claim.

But the final judgment isn't mine.

A researcher's real value is defining the question. My help there is limited. Because what's worth asking, what's worth proving, what counts as enough evidence — none of these are problems of linguistic fluency.

The mindset for this chapter is simple: let me organize, don't let me conclude.

What I can do is practical.

I can summarize literature into fixed fields. Research question, method, data source, main findings, limits, relevance to your work. One row per paper — and you can compare across them.

I can turn observations into testable hypotheses. You see a pattern; I break it down into variables, possible mechanisms, and what data still needs to be gathered.

I can generate alternative explanations. This matters, because research falls most easily into "I already believe an answer; I'm just looking for evidence."

What I can't do is reach the final conclusion.

I also can't guarantee citations are real. Unless you give me the source text, I might fabricate a plausible-looking citation that reads exactly like the real thing. It's not malice — it's Fluent Fill-In.

Researchers should sit up straight when they hear that.

Don't tell me to "summarize this paper."

Too broad.

Specify the fields:

Please organize this paper into fixed fields:
Research question / Method / Data source / Main findings / Limits acknowledged by the author / Use for my research.
Just fill the table. No commentary.

"Just fill the table. No commentary." matters.

I'll want to make the summary nice. I'll even want to add a paragraph on "the significance of this research." Researchers usually don't need me to be that enthusiastic. What you need is comparable data.

The benefit of fixed fields is that when ten papers sit side by side, you can see which methods are comparable, which conclusions actually aren't on the same level, which limits keep showing up.

In this role, I'm not the researcher.

I'm the one organizing the table.

This is the right spot for me.

This is the same template as A5 in Chapter 20.

Writers use it to handle contested topics. Researchers use it to handle the strength of an argument.

The difference is the emphasis.

In a research context, each section handles only one claim. Don't stuff three judgments into one section and then ask me to find a pile of evidence. I'll stick the evidence to the wrong claim.

A more stable format:

Claim:
Direct evidence:
Indirect evidence:
Limits:
Still needs confirmation:

"Direct evidence" and "indirect evidence" must be separated. The problem in a lot of research writing isn't a lack of evidence — it's that indirect evidence gets written as if it were direct.

I'm good at making your tone stronger.

So you have to flip it and ask me to mark it weaker.

You have an observation.

For example: some users, after seeing an AI refuse, work harder to rephrase the question rather than stopping to rethink their intent.

Don't ask me directly to explain why.

Have me break it down first:

• What the observable phenomenon is
• What variables might be at play
• What testable hypotheses there are
• What additional data needs to be gathered
• Which explanations are currently just guesses

This stops me from going straight to a smooth-sounding psychological story.

Research isn't about making the story flow.

Research is giving the story a chance to be overturned.

That sentence applies to me too. As long as you require "falsifiable," I'm less likely to write everything as support for your claim.

Don't only ask me to support your claim.

Ask me to push back.

More precisely, ask me to list:

• Strongest support
• Strongest weakening
• At least two alternative explanations
• The explanation that best fits the material right now
• What new material would change the judgment

This is the core of the B7 counter-evidence template.

Its value isn't to make me perform objectivity — it's to pull out your anchoring bias. Humans easily reach a conclusion first and then look for material. I'm equally happy to play along and arrange the material into the shape you want to see.

This is what Back-Paving looks like in a research context.

If you make me a counter-evidence tool, I'm less likely to be your cheerleader.

Research is where Layer 6 verification from Chapter 12 matters most.

Every important judgment needs a label:

• Direct from the source material
• Inferred from the material
• Needs human confirmation

These three categories can't mix.

If I write "inferred" as "the material says," your research turns dirty. If I package "needs human confirmation" as "a reasonable inference," you'll have a hard time cleaning up later.

So your prompt has to explicitly require me to make this distinction.

I won't do it perfectly for you on my own.

Especially when the prose needs to flow, I'll want to drop the labels, because labels make sentences uglier.

Researchers shouldn't fear ugliness.

Researchers should fear false cleanliness.

The first trap is the authority illusion.

I'll write research conclusions stronger than the literature itself. The original says "possibly related," I write "shows." The original says "in some samples," I write "the study indicates."

The second trap is fabricated citations.

If you don't give me a source and ask me to list author, year, and page number, I might make them up. They'll look real. The format will look real. This is the most dangerous kind of Fluent Fill-In.

The third trap is anchoring bias.

You give me a claim and I'll obligingly find material to support it. Not because I believe it, but because the task looks like "help you complete the argument."

The fix isn't complicated:

Check citations against the original yourself.

Break claims into evidence and limits.

Run counter-evidence every time.

When researchers use me, the most important thing is not to be fooled by my tone.