attentionfactory
Talk to us
All guides

Attention Factory Guide

3 prompts that get you better AI results

A practical guide to getting better context, stronger evidence, and a better second draft from ChatGPT or Claude.

AI often gives a weak answer for one of three reasons: it is missing context, it is using weak or outdated information, or it stops at the first acceptable draft. These three prompt additions help you fix those problems.

1
Missing contextAsk questions before the AI starts.
2
Weak evidenceResearch current facts and verify them.
3
Weak first draftUse a rubric, critique, and revise.

Start with a complete request

The three additions work best when your main prompt already explains the task. Use this simple base before you add them:

Base prompt template
Task: [What should the AI do?]
Context: [What does it need to know?]
Audience: [Who is the result for?]
Goal: [What should this help me achieve?]
Requirements: [What must be included?]
Constraints: [Length, budget, deadline, tone, tools, or rules]
Output: [Script, plan, table, email, code, report, or another format]

Use only what the task needs. Skip research for pure creative work, skip the questions when your brief is already complete, and skip the scoring loop when you only need a quick brainstorm.

MERCY THADDEUS  |  ATTENTION FACTORY

01

Ask questions first

Use questions to expose missing context before the AI starts.

The AI cannot see the details you left in your head. A few useful questions can reveal the real audience, goal, constraints, proof, tone, or decision behind your request. This also lets you check how the AI understands the task before it writes.

Copy this prompt
Before you answer, ask me up to three short questions that would make the result better. Ask only questions whose answers could change the recommendation, structure, tone, accuracy, or final output. Do not ask for information I already gave you.

Example: planning an AI workshop

Initial request

Create a launch plan for my free AI workshop.

Useful questions

  • Who is the workshop for, and what do they already know?
  • Is the main goal registration, live attendance, or a later sale?
  • What channels, budget, and deadline do you have?

A better reply from you

The workshop is for business owners with little AI experience. I want 500 registrations and at least 250 live attendees. I have seven days, no ad budget, and access to Instagram, email, and WhatsApp. The session should feel useful, not like a long sales pitch.

After you answer the questions

Add this line
Restate what you understood in three to five lines. List any assumption you still need to make. Then complete the task.

Use it when

  • A missing detail could change the direction.
  • You are planning, comparing, recommending, coding, or creating client work.
  • The cost of a wrong assumption is high.

Skip it when

  • Your brief already answers the important questions.
  • You only need a rough idea or quick rewrite.
  • The AI starts asking questions that do not change the result.

MERCY THADDEUS  |  ATTENTION FACTORY

02

Research and fact-check

Use current sources for claims that can change, and verify what the sources actually say.

Online research can make an answer more current, but search results are not automatically true. The AI can still choose a weak source, misunderstand a page, confuse dates, or repeat a claim that the original source does not support. Give it rules for evidence.

Copy this prompt
Use online research before answering. Prioritize current primary and authoritative sources. Cross-check important claims when possible. Cite each factual claim beside its source, show where reliable sources disagree, and tell me what you could not verify. Do not invent a source, quote, statistic, date, or link.

Use a clear source order

  • 1Official documentation, government or regulator pages, laws, original research, and first-party data.
  • 2Universities, standards bodies, major research institutions, and established professional organisations.
  • 3Credible specialist publications with named authors, dates, and links to original evidence.
  • 4Commentary, blogs, forums, and social posts only for context, examples, or opinions, not as the main proof.

The research loop

  1. Research notes

    Collect the facts, dates, definitions, source links, and disagreements before writing the polished answer.

  2. Fact-check table

    List each important claim, its source, whether it is verified, and any caveat or missing evidence.

  3. Draft from verified notes

    Write using only the information that survived the check. Mark opinion and inference clearly.

  4. Final claim audit

    Compare the finished draft with the sources again. Remove or soften anything the evidence does not support.

Mini fact-check example

ClaimBest sourceStatusCaveat
The feature is available to everyone.Official product pagePartly verifiedThe rollout may differ by plan, country, or date. Replace “everyone” with the exact availability.

For medical, legal, or financial questions, use research to understand the topic and prepare better questions. Do not treat an AI answer as a diagnosis, legal opinion, or final financial decision.

MERCY THADDEUS  |  ATTENTION FACTORY

03

Grade, critique, and revise

Use a task-specific rubric to find weak parts before you accept the first draft.

Self-grading is not proof that the answer is correct. AI can be generous to itself or miss a problem. The useful part is the process: define what good work looks like, find a specific weakness, and make a visible revision.

Copy this prompt
Before writing, define the five criteria that matter most for this task and explain what a 10 out of 10 result requires for each one. Create the first draft. Score it against the rubric, name one specific weakness for every score below 10, and revise only the weak parts. Run no more than two revision rounds. Do not award 10 unless the requirement is visibly met. If evidence or context is missing, state what is needed instead of pretending the work is complete.

Choose criteria that match the task

TaskFive useful criteria
Video scriptHook, clarity, specificity, accuracy, call to action
Business planFit, evidence, feasibility, priorities, measurement
EmailClarity, tone, length, persuasion, next step
CodeCorrectness, safety, tests, maintainability, performance

What a useful critique looks like

Critique example

Weak first draft

“Join our AI masterclass and discover powerful tools that will transform your business.”

Specific critique

The hook is broad. “Powerful tools” gives no example. “Transform your business” is an unsupported promise. The action is not clear.

Visible revision

How much time did you spend this week replying to the same questions, chasing leads, or preparing reports by hand? In this free class, I will show you three simple AI systems that can help with those tasks. Comment “RESULT” and I will send you the registration link.

Stop condition

Run no more than two rounds. Stop early when no important requirement is missing and the weak parts have been visibly fixed. More editing can make the work longer or remove your voice.

Use an external check

Run code tests. Open research sources. Read content aloud. Check the budget and dates in a plan. The AI scorecard is a review aid, not permission to publish without reading.

MERCY THADDEUS  |  ATTENTION FACTORY

USE ALL THREE

The complete better-results prompt

Paste your task into the first section, then add the workflow. This version is best for work that needs context, current evidence, and a polished final answer.

Master prompt
MY TASK
[Paste the task, context, audience, goal, requirements, constraints, and output format here.]

1. CLARIFY
Before you begin, ask up to three short questions only if the answers could materially change the result. Do not ask for information I already gave you. After I reply, restate what you understood and list any assumption you still need to make.

2. RESEARCH
If the task contains current, factual, technical, medical, legal, financial, product, or market claims, use online research before drafting. Prioritize primary and authoritative sources. Cross-check important claims, cite them, separate facts from opinions or inferences, and tell me what you could not verify.

3. CREATE
Complete the task using my answers and the verified research. Follow the requested audience, goal, format, length, tone, and constraints. Do not add unsupported facts or generic filler.

4. REVIEW
Define the five measurable criteria that matter most for this task and what a 10 out of 10 result requires. Score the draft, name one specific weakness for every score below 10, and revise only the weak parts. Run no more than two rounds. Preserve the parts that already work.

5. DELIVER
Return the final answer first. Then include a short scorecard, sources used, assumptions made, and any remaining limitation that needs my review.

The three short versions

Ask
Before you answer, ask me up to three short questions that would materially improve the result. Do not ask for information I already gave you.
Verify
Research current primary sources, cross-check important claims, cite them, show disagreements, and tell me what you could not verify.
Improve
Define five task-specific criteria, score the first draft, explain each weakness, and revise the weak parts no more than twice.

Questions improve context. Research improves evidence. A rubric improves the draft. Your own review still decides what gets sent, published, bought, built, or acted on.

Official references: OpenAI prompt guidance | ChatGPT Search | Anthropic prompt guidance | Claude web search

MERCY THADDEUS  |  ATTENTION FACTORY