The Reversal of Perspective

This needs the least translation of any chapter, since the technique is a conversational move rather than a prompt structure - ask the question in the same chat and the original prompt is still there to be examined. What the platform adds is the next step: once you know which word was ambiguous, the fix belongs in a Gem, so a term you use in a specific way carries its definition into every future chat instead of being redefined each time. For the objective-level error in the third example, grounding adds something real - after the diagnosis, search for what the audience actually responds to rather than relying on anyone's instinct.

• Ask in the same chat; your original prompt is still there.
• Put the fix in a Gem - define your terms once.
• Ask whether search results or your wording misled it.
• For objective errors, ground the corrected brief.
• Keep the diagnosis and the retry as separate turns.

### TURN 1: THE REVERSAL ###

Stop - do not try again yet. I want the diagnosis.

**MY PROMPT:** "Write a social media post about our company's new charity initiative. We need to get this out there and be very professional and buttoned-up about it."

1. What tone and style did you understand I was asking for? Quote the exact words that told you.
2. What other readings of "professional and buttoned-up" were available? Why this one?
3. Given it is a charity initiative on social media, what did I probably actually want - and where does that fight with what I wrote?
4. Is there a contradiction between "get this out there" and "buttoned-up"?

Do not search for this. It is a question about my sentence, not about the world. And do not apologise or tell me the prompt was fine - name the word that did the damage.

### TURN 2: THE CORRECTED PROMPT ###

I wrote "professional" meaning "not flippant". It was read as "corporate". Let us try again.

Write a social media post about our new charity initiative.

* **Tone:** inspiring and warm. Not corporate, no press-release phrasing, no "we are pleased to announce".
* **Goal:** someone scrolling past feels something and shares it. This is not a formal record of the announcement.
* Lead with the people it helps, not with our company.
* Short enough to read without tapping "more".
* One clear line on how to get involved.

Give me three versions at increasing warmth so I can see the range.

**Then search, for one thing only:** find two or three examples of Indian brands posting about charity work that were widely shared, and cite them. I want to see what actually travels rather than guess at the register - and if what you find is that the warm ones did better, that is evidence rather than my opinion.

### TURN 3: MAKE THE FIX PERMANENT ###

Create a Gem called "Our social voice" holding: when I say "professional" I mean accurate and not flippant, never corporate or formal; our social posts lead with people, not with the company; never "we are pleased to announce".

This is the part worth more than the post. The ambiguity was in my own vocabulary, so fixing one prompt fixes nothing - a Gem fixes every prompt I write after this.

### TURN 1: THE REVERSAL ###

That is not what I meant. Do not revise the plan - diagnose my prompt.

**MY PROMPT:** "I need a plan to improve the velocity of my engineering team."

1. What did you take "velocity" to mean? List every reading you considered, not only the one you used.
2. Which did you choose, and what decided it?
3. Could anything in my sentence have told you whether I meant the Agile sprint metric, general delivery speed, or something looser about momentum?
4. What else was underspecified that mattered more than the word itself?

Do not search. And if my prompt simply did not contain enough to choose, say so rather than justifying the reading you picked - "there was no way to know" is the most useful answer available here.

### TURN 2: THE CORRECTED PROMPT ###

**Definition:** velocity means the Agile metric - story points completed per two-week sprint. Not general speed, not morale.

**Context:** [team size] engineers, currently averaging [X] points per sprint, target [Y]. The internal complaints are [the real bottleneck].

Give me a plan in three clearly separated parts:

1. Changes that genuinely increase throughput.
2. Changes that only inflate the number - looser estimation, splitting tickets, counting differently. Name these so I can avoid them; they are the easiest way to hit my target and learn nothing.
3. What to measure alongside velocity so I notice if quality is paying for the speed.

**Use search for part 3 specifically.** What do published engineering sources actually recommend tracking alongside velocity - DORA metrics, change failure rate, cycle time? Cite them with links I can open. This is where sourcing helps and instinct does not, because the counter-metrics are a documented and contested subject rather than a matter of opinion.

If velocity is the wrong thing to optimise given the bottleneck I described, say so before the plan.

Put it in a Google Doc called "Velocity plan" so I can share it with my leads.

### TURN 3: MAKE THE FIX PERMANENT ###

Create a Gem called "My engineering context" holding: team size, our definition of velocity, the tools we use, and our known bottleneck. The whole failure here was a word that means something precise inside my team and three things outside it. A Gem means I define it once instead of rediscovering this exchange every month.

### TURN 1: THE REVERSAL ###

This outline is not persuasive and I want to know why before asking again.

**MY PROMPT:** "I am giving a presentation to a potential client in Hyderabad. They are a big construction company. My company sells advanced drone-based surveying equipment. The presentation needs to include all the technical specs of our top three drone models. Create an outline."

1. What did you understand the single most important goal to be? One sentence.
2. Which part of my prompt put it there?
3. The real question: if you were advising me rather than following instructions, what would you say the goal of a first client meeting is? Where does that differ from what I specified?
4. Why would a more detailed version of my prompt not have fixed this?

No search for this turn. Note that my prompt was specific, not vague - that is what makes it interesting. Tell me what it specified that it should not have.

### TURN 2: THE CORRECTED PROMPT, GROUNDED ###

**Primary goal:** persuade a major construction client in Hyderabad to book a live demonstration. One outcome; nothing else matters as much.

My company sells drone-based surveying equipment. The audience is a large construction company - project managers plus someone from finance.

**First, research this - use Google Search and cite everything:**

1. What is reported about how large Indian construction firms currently handle site surveying, and what it costs them in time and money? I need their present pain in their own terms, with sources.
2. What does the DGCA currently require for commercial drone surveying in India - registration, pilot certification, airspace permissions? Link the official pages. If my audience's first objection is regulatory and I have not prepared for it, the demonstration does not get booked.
3. What documented case studies exist of drone surveying on Indian infrastructure projects, with reported time or cost savings? Cite them with dates. A sourced number from someone else's project is worth more in this room than any claim of mine.

**Then build the outline:**

Persuasive, benefits-led, not features-led. Frame everything as what surveying currently costs them - days of delay, rupees, safety risk - not as what our drones can do. Use the researched figures where they help, cited, so I can verify each before I stand up.

Technical specifications: one appendix slide, referenced only if asked.

Per slide: the single message it must land, and the question it should leave the audience wanting to ask.

Close with a specific ask - a date, a site, a named person.

**Output:** Google Slides called "Drone surveying - client pitch", with my sources in the speaker notes so I can check each figure before the meeting.

And tell me the one thing I still have not given you that would most improve this. I have been wrong once already about what mattered in this brief.