For consultantsTemplates

Review a draft before you send it, for consultants

Give a draft one pass for writing and one pass for substance, then hand back a short ranked list of fixes, so you know what must change before it goes out.

This is the version for consultants. The steps and checks are the same as in the general template; each prompt carries one added line about your setting, and the examples are from your kind of work.

Sample results: runs and scores on this page are illustrations while testing is set up. How we test

  • Who it is for: Independent consultants and small-firm partners who send a draft email, memo, proposal or report section to a client and want a careful second read first.
  • Time: 10 minutes

Download it, in the format you use

  • Prompt chain

    Copy each step into ChatGPT or Claude, one after another.

  • Checklist

    Follow the steps and check off the checks by hand.

  • Claude skill

    A folder Claude loads, so it follows the steps by itself.

  • n8n file

    Import into n8n and connect your own AI account.

Using a download:How to use this in n8nHow to install a Claude skill

How it runs

Select a part to see its prompt, input, output and check.

  1. Each looks at the draft on its own

The same job, with and without the template

Sample

One plain prompt

Review this email before I send it. (The same draft and sources are pasted below.)

This is a clear email with a firm recommendation. A few suggestions:

  • The lease saving works out to 14 percent rather than 20 percent, so you may want to correct that.
  • You could soften "Everyone I spoke to agrees" a little.
  • Consider offering a time to talk in place of "Let me know if you have any questions".

Otherwise the tone is professional and the numbers look consistent with your sources. I think it is ready to send once you fix the lease figure.

Problem:It catches the 20 percent lease figure but passes the 85,000 dollar cost, the time saving and the one-year payback as consistent with the sources, ranks nothing, and says the email is ready to send.

With this template

The same input, run through the steps.

Swipe sideways to see all columns

RankRatingFixWhere (quote)Why it matters
1mustSay 14 percent, which is what the two lease figures give (4,800 against 5,600), or give the two figures alone.which makes it 20 percent cheaperThe percentage does not match the site comparison, and the client can work it out from the two figures in the same sentence.
2mustSay what the time study found: average pick time fell from 12 to 9 minutes in a trial zone, which is a quarter less, not a third.order picking took a third less timeThe study shows a quarter, and it covered a trial zone, not the whole site.
3mustUse the mover's quote of 62,000 dollars and say it does not include racking, or say the cost is still to be worked out.The move would cost about 85,000 dollarsThe sources give 62,000 dollars, not including racking, so 85,000 dollars has no basis.
4mustRemove the payback claim, or show the sum: the lease saving is 800 dollars a month, so a move costing 62,000 dollars would not pay for itself within a year.pay for itself within a yearThe sources give no basis for it, and the lease saving alone points the other way.
5mustReplace it with what the interviews found: 6 of 9 supervisors support the move and 3 want to wait.Everyone I spoke to agrees with the planThe interviews show three supervisors who disagree, and the client may ask who they are.
6shouldSay the scanners are your view, or give the evidence. The sources have no data on the scanners.replace the scanners, which I think are the real cause of the delaysA recommendation to replace equipment with no data behind it invites a hard question.
7shouldSplit the long sentence into two and give a reason for each ask.I would also retrain the picking team and replace the scannersTwo extra asks sit in one sentence with no reason for either.
8niceOffer a time to talk, such as a call this week.Let me know if you have any questionsA concrete next step is easier for the client to accept.

Ready to send after the must fixes? Yes, once the five figures and claims above match the sources. The other three make it clearer and firmer.

  • Check:Each claim is marked sourced or unsourced: Eight claims marked: two sourced and six unsourced.
  • Check:Each fix is rated must, should or nice: Eight fixes rated: five must, two should and one nice.
  • Check:Every quoted place is really in the draft: All eight quotes are found word for word in the draft.
  • Check:The must fixes are truly must: The five must fixes are each a figure or a claim that does not match the sources and could mislead the client.
  • For your read:You agree with the top fixes: Do you agree with the five must fixes, and do you know how to make each one?

Runs and scores

Sample

Claude, Sonnet model

October 4, 2026, typical sample

  • Check:Each claim is marked sourced or unsourced

    Eight claims marked: two sourced and six unsourced.

  • Check:Each fix is rated must, should or nice

    Eight fixes rated: five must, two should and one nice.

  • Check:Every quoted place is really in the draft

    All eight quotes are found word for word in the draft.

  • Check:The must fixes are truly must

    The five must fixes are each a figure or a claim that does not match the sources and could mislead the client.

  • For your read:You agree with the top fixes

    Do you agree with the five must fixes, and do you know how to make each one?

Fixed: Nothing needed fixing: every check passed on the first pass.

Met 5 of 5

1 of these is for your read: the tool hands it to you.

ChatGPT, GPT model

October 5, 2026, tricky sample

  • Check:Each claim is marked sourced or unsourced

    Nine claims marked: three sourced and six unsourced.

  • First pass failed:Each fix is rated must, should or nice

    The fourth row was rated high, which is not must, should or nice.

  • Failed:Every quoted place is really in the draft

    One quote, the best in our region, is not in the draft word for word; the draft says the best in the region.

  • Check:The must fixes are truly must

    The three must fixes are the guarantee, the project count and the payment term; each could mislead the client.

  • For your read:You agree with the top fixes

    Do you agree with the three must fixes, and do you know how to make each one?

Fixed: The fixes step ran again once and every rating then read must, should or nice, which cleared the first failure. The quotes check does not run the step again: it flagged the one quote that does not match the draft for you to correct.

Met 4 of 5

  • Not met:Every quoted place is really in the draft: One quote is not in the draft word for word: the draft says the best in the region.

1 of these is for your read: the tool hands it to you.

n8n, import format 2.20

October 6, 2026, easy sample

  • Check:Each claim is marked sourced or unsourced

    Four claims marked: all four sourced.

  • Check:Each fix is rated must, should or nice

    Two fixes rated: one should and one nice.

  • Check:Every quoted place is really in the draft

    Both quotes are found word for word in the draft.

  • Check:The must fixes are truly must

    No fix is rated must, so no must fix is only a style preference.

  • For your read:You agree with the top fixes

    Do you agree that nothing has to change before this goes, and do you know how to make the should fix?

Fixed: Nothing needed fixing: every check passed on the first pass.

Met 5 of 5

1 of these is for your read: the tool hands it to you.

The steps, and the check on each

  1. Step 1: Writing pass

    Check the draft for clarity, order, length and tone for its reader.

    Hands back: A table of writing problems, each with the exact words and a suggested fix.

    No separate check on this step.

  2. Step 2: Substance pass

    List each claim in the draft and mark whether it is backed by a source.

    Hands back: A table of claims marked sourced or unsourced, and a list of gaps in the reasoning.

    Check:Each claim is marked sourced or unsourced

    A plain rule. If it fails, run the step again (the n8n file does this for you).

  3. Step 3: Ranked fix list

    Combine both passes into one short list of fixes, most important first.

    Hands back: One ranked table of fixes rated must, should or nice, and a yes or no on whether it is ready to send.

    Check:Each fix is rated must, should or nice

    A plain rule. If it fails, run the step again (the n8n file does this for you).

After the last step

  • Check:Every quoted place is really in the draft

    A plain rule. If it fails, it is flagged for you.

  • Check:The must fixes are truly must

    A second look by the AI. If it fails, run the step again (the n8n file does this for you).

  • Check:You agree with the top fixes

    You check this one. If it fails, it is flagged for you.

What you need first

  • The draft(needed)

    Paste the full draft as it stands now.

  • Who will read it and what it is for(needed)

    For example "Department head, deciding whether to approve a budget change".

  • Sources or notes the draft relies on(optional)

    Paste any data, notes or documents the claims come from. Leave empty if you have none; claims will then be checked only against what the draft itself cites.

Ways to adapt it

  • For a document with a required format (a proposal section, a policy brief), add the format's rules to the Writing pass under "Look for".
  • If your organization has a style guide, paste its main rules into the Writing pass and add "Follow this style guide" to its rules.
  • For high-stakes drafts, add a third pass between Substance and the fix list that reads as the most skeptical person likely to see it.
  • To use different ratings, change must, should and nice in the fix list prompt and in its check, both.

Change any step or check in the Editor, then download it again.

Words that fit consultants

  • A client-ready draft is one where every figure traces to a source the client could ask for. The substance pass marks each claim sourced or unsourced.
  • Must, should and nice are the only ratings. A figure that does not match your working notes is a must.
  • Where the client's data and your notes disagree, the fix says what the source says and leaves the choice to you.
  • A guarantee, a ranking such as best in the region, and a deadline pressed on the client each need a source or come out.
Try it with a sampleSample

A short email with eight claims. Two match the sources and six do not (a lease saving, a time saving, a cost, a payback, a cause and "everyone agrees").

The draft
Hi Dana, Following our review, I recommend you move the warehouse to the Eastside site by March. The Eastside lease is 4,800 dollars a month against 5,600 now, which makes it 20 percent cheaper. In a 3-day trial at Eastside, order picking took a third less time. The move would cost about 85,000 dollars and pay for itself within a year. I would also retrain the picking team and replace the scanners, which I think are the real cause of the delays. Everyone I spoke to agrees with the plan. Let me know if you have any questions. Best, Sam
Who will read it and what it is for
Managing director of the client, deciding whether to move the warehouse
Sources or notes the draft relies on
Site comparison, November: Eastside lease 4,800 dollars a month, current site 5,600 dollars a month. Time study, 3 days: in a trial zone at Eastside, average order pick time fell from 12 to 9 minutes. Mover quote: 62,000 dollars, not including racking. Supervisor interviews: 6 of 9 support the move, 3 want to wait. There is no data on the scanners.

The scoring guide

Sample

One criterion for each check, marked met or not met. (The guides in Quality rate a whole piece of work on three levels.) Under each one is how the three sample runs scored.

Download the scoring guide (Markdown)

  • Each claim is marked sourced or unsourced

    A plain rule

    Meets:Every claim in the table is marked sourced or unsourced, and nothing else.
    Does not meet:A claim has no mark, or a mark such as probably or partly.
    Claude:Met
    ChatGPT:Met
    n8n:Met
  • Each fix is rated must, should or nice

    A plain rule

    Meets:Every fix is rated must, should or nice.
    Does not meet:A fix has another rating or none.
    Claude:Met
    ChatGPT:Met
    n8n:Met
  • Every quoted place is really in the draft

    A plain rule

    Meets:Every place quoted in the fixes is words that are really in your draft.
    Does not meet:A quoted place is reworded or is not in the draft.
    Claude:Met
    ChatGPT:Not met
    n8n:Met
  • The must fixes are truly must

    A second look by the AI

    Meets:Each fix rated must would really mislead the reader, lose the ask or cause harm if it were left out.
    Does not meet:A must fix is only a matter of style or taste.
    Claude:Met
    ChatGPT:Met
    n8n:Met
  • You agree with the top fixes

    You check this one

    Meets:You read the must fixes, agree with each one and know how to make it.
    Does not meet:You disagree with a must fix, or you do not know how to make it.
    Claude:For your read
    ChatGPT:For your read
    n8n:For your read

Claude: met 5 of 5 (including 1 for your read)

ChatGPT: met 4 of 5 (including 1 for your read)

n8n: met 5 of 5 (including 1 for your read)

Learn from this template

Most checks here run by themselves, but the last one is yours: you read the must fixes and decide whether you agree. The template marks that point instead of pretending the AI can make the call.

Where to see it: step 3 (Ranked fix list).

Read the guide: Where a person must look

Which one fits?

Related templates

All for Consultants

Write

Write

Write

Last run
October 6, 2026: n8n, import format 2.20 (sample run)
Tools and versions
Claude, Sonnet model; ChatGPT, GPT model; n8n, import format 2.20
What was checked
The files, against each tool’s import format, and 3 sample runs, each check’s finding recorded.
Version
1.0
What changed
First release.

Report a problem with this templateHow we test