Iterate: when the first answer is wrong

Samira’s seventh ChatGPT draft of the Northline Logistics email still opened with “I hope this message finds you well.” Finance’s recon sheet, sitting in the other tab, said the Q2 overage was $18,640. The draft still used $14,200, a line from a 2025 rate PDF she had dropped in the first prompt “for context.” She had typed “fix the tone” seven times. The greeting got warmer. The dollar amount never moved. She almost hit send because the seventh version sounded like a person.

This is Part 14 of Phase P, and Part 3 of Prompting for everyone. Part 1 was talk like a clear human. Part 2 packed the first prompt: context, constraints, examples. This part is the follow-up. First drafts are often directionally useful and specifically wrong. You name the error, keep what worked, add one constraint, and try a smaller ask. Next is study, work email, plans, and shopping research. If you still need an account, use AI setup from zero.

What you’ll learn

How to tell a useful first draft from a specifically wrong one (fluency is not a send)
Four kinds of wrong (facts, tone, format, missing constraint) and the next prompt for each
How to keep the paragraph that already works and add only one new rule
A five-step iterate script you can paste into ChatGPT, Claude, Gemini, or Grok
When you are arguing with the model, and when to stop and edit the last 10 percent yourself

First answers are useful and specifically wrong

A first reply can look finished. Complete sentences. A subject line. A polite closer. That is the trap. Models are good at the shape of an email, a study outline, or a shopping comparison. They are sloppy with the one fact you will get in trouble for: the dollar amount, the date, the name of the person who should not be copied.

Directionally useful means the job is already in the draft. Samira’s first pass had “please issue a credit memo” and a Friday ask. Those two lines were keepers. Specifically wrong means a detail that would be false if you sent it. $14,200 was false. The stock greeting was the wrong voice for vendor AP.

OpenAI’s prompting page (as of writing) says the first prompt does not need to be perfect. Review the result, then ask for the specific change you want. Anthropic and Gemini say the same: you iterate. None of them tell you to type “make it more professional” seven times. That names a vibe, not an error. Read numbers first. Names second. Tone last. A warm email with the wrong money is still a bad email. Samira’s source of truth was Q2_vendor_recon_v3.xlsx, cell E14. The chat was not.

Name the error before you type

Four-step iterate loop: read the draft, name one error, keep what worked plus one rule, then ask for a smaller local edit

Before you send the follow-up, say the failure in one sentence you could read aloud. “The opener is a stock greeting I never use.” “The overage is $18,640, not $14,200.” If you cannot quote the bad line or the bad number, you are not iterating. You are rolling the dice.

Four kinds of wrong cover almost every first draft that feels off. Match the kind to the next prompt. Do not mix them in one turn if you can help it. Mixing is how a tone fix rewrites a number you had already corrected.

Kind of wrong	What it looks like	Next prompt move
Facts	$14,200 in the body when E14 says $18,640; a 2025 PDF treated as this quarter	Paste the source. “Use only these figures. If a number is missing, flag it. Do not guess.”
Tone	The stock “finds you well” opener; “we reserve our rights”; a joke in a vendor AP email	Quote the bad line. “Delete this opener. Start with the invoice number. No legal language.”
Format	An essay when you needed 4 bullets; 280 words; a table smashed into a paragraph	“Rewrite as 4 bullets, 90 words max, same facts. Keep the Friday deadline.”
Missing constraint	CC’d the salesperson; mentioned legal; used a nickname the vendor does not have	Add one rule. “Send-to is AP only. Do not mention legal. Use Northline Logistics, not Northline.”

Samira’s seven turns were all tone. The fact error sat there because she never pointed at it. The model had no reason to drop $14,200. She had given that number a home when she attached the old PDF.

Rule of thumb: If you cannot quote the bad line or the bad number, you are not iterating yet. “Fix the tone” is a shrug.

Keep what worked, then ask smaller

A follow-up that says “rewrite this” throws away the good half. Models do not know which paragraph you liked unless you say so. Quote the keeper. “Keep the credit-memo ask and the Friday deadline. Do not touch those two sentences.” That is a fence around work that already passed.

Add one new rule per turn. One. If you ban the greeting, lock $18,640, cap length at 90 words, and forbid legal language all at once, the next draft will obey two and invent a fifth problem. You will not know which instruction landed. Samira’s manager Slack at 4:12pm was “ping Northline about the Q2 overage before they close.” One job: the right invoice and the right money.

Iteration uses the Part 2 kit on a draft you can see. Context is now the draft plus the sheet. The constraint is the one new fence. The example is two lines you actually send: “Hi Northline AP, Invoice 8841 shows an overage of $18,640 against our Q2 recon. Please issue a credit memo by Friday.” Paste that. Do not describe “professional.” Regenerate is a coin flip. A named follow-up is a steer. Ask for a local edit: change only the opener, or replace $14,200 with $18,640 everywhere. If the same error comes back, shrink again: return the email with this opener deleted and nothing else changed. A thread that keeps restoring $14,200 may still have the 2025 PDF in view. Start a new chat with the recon numbers only.

A five-step iterate script you can paste

Copy this into the chat after a draft lands. Fill the brackets. Send it as one message. Then read the reply the way you read the first draft: numbers first.

Iterate script (paste, then fill the brackets)

1. NAME THE ERROR
The last draft is wrong in this specific way:
"[quote the bad sentence or the bad number]"
Do not rewrite the whole piece.

2. KEEP WHAT WORKED
Keep these parts unchanged:
- [subject / paragraph 2 / the Friday ask / the table]
Do not drop them to improve the rest.

3. ADD ONE CONSTRAINT
Add this single new rule: [ban the stock greeting / use $18,640 only / 90 words / no legal language]
Do not add a second rule in this turn.

4. SMALLER ASK
Change only: [the opener / the dollar figure / the sign-off]
Return the full draft with that local edit, not a new essay.

5. STOP CHECK
Count my named fixes on this error: [1 / 2 / 3]
If this is the third time the same error remains, say so in one line and stop.
I will edit it myself or start a new chat with a tighter brief.

The script forces you to quote, protect, and count. Samira skipped the stop check. Turn seven felt like progress. The fact-error count was zero. The greeting count was seven, all unnamed.

Worked example: seven tone prompts, one stale number

Invoice 8841. True overage $18,640 from cell E14. Stale $14,200 from the 2025 PDF. Audience: Northline AP, not the salesperson. Goal: credit memo by Friday, 90 words, no legal. Seven tone turns did this:

Turn	What she typed	What still failed
1	Write a professional email about the Q2 overage. (PDF attached.)	Stock greeting, $14,200, 280 words, a legal closer
2	Make the tone more professional	Greeting stays, $14,200 stays
3	Less stiff, more human	Greeting plus a soft joke, $14,200
4	Fix the tone	Greeting back, still $14,200
5	Don’t sound like a robot	New opener, cousin of the greeting, $14,200
6	Shorter and warmer	About 90 words of warmth, $14,200
7	Fix the tone again	Greeting returns, $14,200, she almost sends

The turn that would have ended it is not another vibe. Keep the credit-memo ask and the Friday deadline. Delete the opener. Start with invoice 8841. Use $18,640 from the recon sheet (paste E14: 18640). Do not mention legal action. 90 words. Send-to is AP only. If you want one rule per turn, lock $18,640 first, then ban the greeting. Two short turns beat seven long ones.

Arguing with the model versus fixing the brief

Arguing looks like a stack of adjectives with no new source. Warmer. Less stiff. More human. Not that human. More like how I write. The model keeps guessing because you have not shown a sample or banned a line. Taste fights burn twenty minutes and still ship the stock greeting under a new costume.

Fixing the brief looks like new information entering the chat. You paste E14. You paste two lines you sent last month. You name the audience (AP, not the AE). You say which file is stale: “Ignore the 2025 PDF. It is last year’s rates.” A follow-up is allowed to correct a thin first prompt or a wrong file. Test: did this turn add a fact, a quote, a ban, or a format cap? If it only added a feeling word, you are arguing. Paste a sample or edit the opener yourself. If the model is stuck on a number it saw earlier, start a new chat with the recon figures and without the 2025 PDF.

When to stop iterating

Four stop conditions: same error three times, a tone war with no new fact, an unsourced number, and a remaining fix you can edit faster y…

Stop when the same named error is still there on the third try. A fourth “fix the tone” will not discover cell E14. Stop when you are in a tone war and no new fact has entered the chat. Stop when a dollar amount, date, or name cannot be pointed at in a sheet you trust. Do not send a figure you did not check. Stop when the remaining work is a delete or a paste you already have on screen.

Twenty minutes is a ceiling for one short email. Samira’s seven turns ran longer, and the $14,200 never got a glance. Polishing tone around an unchecked number is decoration. Part 5 maps when to close the chat. If the script’s step 5 trips, you are done prompting on that error.

Common mistakes

Typing “fix the tone” or “make it more professional” without quoting the bad line.
Leaving a stale PDF in the thread, then wondering why the old number keeps coming back.
Rewriting the whole draft and losing the paragraph that already worked.
Stacking five new rules in one follow-up, then being unable to tell which one landed.
Using Regenerate instead of a named local edit.
Checking tone for seven turns and never checking the money.

How to practice this week

Take one real draft: a vendor email, a study outline, or a short plan. Run the five-step script once. Quote one error. Keep one keeper. Add one constraint. If the same error survives three named tries, stop and edit it yourself. Next: study, work email, plans, and shopping research. The first-prompt kit is context, constraints, and examples. More writing practice: AI for writing and questions and Learn.

Quick recap

First drafts are often useful in shape and wrong in the fact you will get blamed for.
Name the error (facts, tone, format, missing constraint). Quote the bad line.
Keep what worked. Add one constraint. Ask for a smaller edit, not a new essay.
If the same error remains on try three, stop. Paste a better brief or edit it yourself.
Samira’s $14,200 survived seven tone prompts because nobody checked the sheet.

Sources

OpenAI: Prompting (ChatGPT Learn) (first prompt need not be perfect; follow-ups name the specific change)
OpenAI: Prompt engineering (iterate on prompts; review after each change)
Anthropic: Prompt engineering overview (clear instructions, then refine)
Google: Gemini prompting strategies (design, test, iterate)
AMS: Prompting for everyone
AMS: Give context, constraints, and examples (the first-prompt kit this part iterates on)

Keep going

Same lessons in your feed

Short diagrams and hooks on Instagram, X, and Facebook.

@analyticsmadesimple Follow @AnalyticsSimply Follow Analytics Made Simple Follow