Vague client feedback used to cost me the most hours of any part of freelancing. Not the actual work — decoding what "make it pop more" or "just needs something" actually meant, then guessing wrong, then doing a second round because I guessed wrong.

I built one prompt that turns mush into a checklist I can act on in one pass. It's saved me at minimum an extra revision round on most projects since.

The prompt:

You are helping me interpret vague client feedback so I only need one revision pass, not three. Here is the feedback, verbatim: [paste exactly what they said, don't clean it up] Here is context on the deliverable: [what it is, who it's for, what stage it's at] For each piece of feedback: 1. Restate it as a specific, actionable change (not "make it pop" but your best guess at what visual/structural change would satisfy that comment) 2. Rate your confidence in that interpretation: high / medium / low 3. For anything medium or low confidence, write ONE clarifying question I could send back instead of guessing Do not guess on anything you'd rate low confidence. Flag it instead. I'd rather ask one smart question than redo work twice.

Why the confidence rating matters more than the interpretation itself

Early on I skipped that part and just asked for "translate this feedback into action items." The output was confident-sounding nonsense for the ambiguous stuff — clean, specific action items that were completely wrong guesses. I'd implement them, send it back, and get more vague feedback because I'd solved the wrong problem.

Rating confidence forces the real distinction: stuff that's obviously fixable (they said "the button color feels off," clearly a color problem) versus stuff that's genuinely unclear (they said "it's not quite there yet," which could mean anything). The low-confidence items become questions, not guesses.

Real example from three weeks ago

Client feedback on a landing page draft: "Overall good direction but the top section needs something. Also can we make the pricing feel less salesy."

Ran it through the prompt:

• "Top section needs something" → LOW confidence. Clarifying question generated: "When you say the top section needs something — is it the headline message, the visual layout, or does it feel incomplete because something's missing (like a photo or a stat)?"

• "Pricing feels salesy" → MEDIUM confidence. Interpretation: probably referring to urgency language ("limited time," "act now") or too many exclamation points. Question offered as backup: "Want me to remove the urgency language specifically, or restructure the whole pricing section's tone?"

I sent both questions in one email instead of guessing. Client replied: the top section was missing a customer photo (I never would have guessed that), and yes, remove the urgency language specifically. One round of actual fixes. Done.

Compare that to my old process: I would have redesigned the entire top section based on my own guess about what "needs something" meant, sent it back, gotten "closer but not quite," and burned another round.

The pattern behind this

Clients aren't bad at giving feedback because they're difficult. They're bad at it because they can feel something's off without having the vocabulary to name it precisely. Your job isn't to guess harder. It's to ask the right narrow question so THEY can name it, since they're the only one who actually knows.

What I never do with this prompt: I never send the AI's interpretation straight to the client as if it's my own read. I use it to organize my thinking and draft smart questions, then I reword everything in my own voice before it goes out. A robotic-sounding clarifying question undoes the whole point.

This is one of maybe a dozen prompts I keep in my client-ops toolkit — feedback translation, meeting recaps, proposal drafts, all built to cut redundant revision rounds. I go through the complete set inside Module 3 of the AI Mastery Hub course.

Next time you get feedback that makes you sigh, run it through this before you touch the actual work. One clarifying question beats one wasted revision round every time.

Inbound leads are not a personality trait. They show up when a stranger can explain you in one sentence after 15 seconds on your profile.

Most people reverse this. They post for 90 days hoping the market will figure out what they do. The market will not. The market is tired.

I rebuilt my inbound around three artifacts: a sentence, a proof stack, and one repeating content bet. AI helped me draft them. I still had to pick. Here's the exact pass.

1. The sentence (20 min, once)

I dumped 10 real client problems I'd actually solved (not job titles) into a chat:

"Here are problems I have actually been paid to solve: [list]. Write 10 positioning sentences in this shape: I help [specific person] get [specific result] without [the thing they fear doing]. Rules: no word 'passionate.' No 'I help businesses grow.' If the person isn't recognizable in a coffee shop, rewrite it. Then tell me which 3 are narrow enough that the wrong people would self-select out."

I picked one that made me slightly uncomfortable because it excluded work I sometimes take:

"I help solo consultants turn delivery docs into products, without building a giant audience first."

If your sentence still works if you swap in any freelancer, it's a bio, not a brand. Keep going.

2. The proof stack on the profile (30 min)

Profile is the landing page. Posts are ads for the landing page. I asked:

"Using this sentence: [paste]. Rewrite my headline, about section, and featured-link blurb. Headline: result + who, max 12 words. About: 6 lines. Line 1 is the sentence. Lines 2-4 are 3 proof bullets with real constraints (time, money, or mess), no fake stats. Line 5 is who should not follow me. Line 6 is how to reach me. Flag any line that sounds like a course ad."

Then I replaced every invented-sounding proof with something I could show: a screenshot, a quote, a before/after of a doc. AI will try to polish you into a guru. Delete that.

3. One content bet you can repeat for 8 weeks (45 min to set, 25 min/week after)

Inbound dies when your topics bounce around. Pick one bet. Mine for that stretch: "productize the thing you already deliver."

Setup prompt:

"Give me 8 weekly post angles for this bet: [sentence]. Each angle must be a different objection (too early, too niche, clients will be mad, it's not a real product, etc.). For each: a hook that names the objection, a 150-word body that uses one scene from real work (I'll fill the scene), and a last line that is a question only my ideal client would bother answering. Ban: motivation, 'comment YES if,' and lists of tools."

Every Monday I pick the next objection, drop in a real scene (a client asking for the same deck again, a Sunday spent rebuilding a spreadsheet), and post. Same bet. Different objection. People start finishing the sentence for me in DMs: "you're the productize guy, right?"

That's the job. Not going viral. Becoming obvious.

What inbound actually looked like

Week 3 of that bet, a consultant I'd never spoken to DMed: "I sell workshops. I rebuild the workbook every time. Is that the thing?" We got on a call. She didn't need to be pitched on who I was. The profile sentence plus three posts had already done the filter.

That's a better lead than 50 cold emails, and it started with copy on a profile, not a funnel.

What I will not do

• Post 5 times a day in a different niche each time and call it a brand.

• Let AI invent clients, metrics, or a persona I don't serve.

• Hide the offer so I seem mysterious. Mysterious does not get DMs. Clear does.

If you want inbound this quarter, do not start with a content calendar. Write the sentence until it excludes people. Put proof under it. Repeat one bet until strangers can introduce you.

I walk through positioning, proof stacks, and the 8-week content bet inside the AI Mastery Hub course. For this weekend: write 10 sentences, pick the one that loses you the wrong work, and put it at the top of your profile before you publish anything else.

Obvious scales. Clever doesn't.

If you type "give me 10 content ideas for [niche]," you get the same list everyone else got. "5 mistakes," "what I wish I knew," "hot take on productivity." That's not a brainstorm. That's the model's greatest hits.

I stopped asking for ideas in the abstract. I run a 3-pass technique I call kill list, collision, then one-line tests. It takes about 15 minutes. The output is weirder, which is the point. Unique is usually just "not the first thing the model wanted to say."

Pass 1: Build the kill list (3 min)

Before any ideas, I make it confess the obvious.

"I need ideas for [project: e.g. a LinkedIn series for freelance designers who hate selling]. First, list the 15 ideas you would normally give. Be honest. These are banned. Do not reuse them, even rephrased, in later passes."

I skim that list. If my "original" idea is already on it, I kill it myself. This pass is uncomfortable the first time because you realize how often you were about to publish item #4.

Pass 2: Force a collision (7 min)

Now I don't ask for ideas. I smash two things that don't belong together, plus a hard constraint from real life.

"Using none of the banned ideas, generate 12 concepts by colliding: - Domain A: [the actual topic, e.g. freelance design sales] - Domain B: [unrelated field I actually know, e.g. restaurant service, gym coaching, or how airports sign people] Constraint: every concept must work as a 4-slide carousel or a 90-second video. No ebooks, no 'ultimate guides.' For each concept give: a title that could only exist because of the collision, the mechanic (what the reader does), and why a designer would care this week. If a concept could exist without Domain B, delete it and replace it."

Domain B has to be something I can talk about without research. If I pick "quantum physics" for flavor, the ideas get cute and unusable. I pick restaurant service because I spent years noticing how servers pitch specials without sounding desperate. That is a real analogy I can finish.

Pass 3: One-line tests (5 min)

Twelve concepts is too many. I paste them back:

"For each concept, write one sentence I could post as a hook. Then mark FAIL if the hook still works after you strip out Domain B. Only PASS hooks survive. Rank the PASS list by how annoyed a smart designer might be (annoyance = they think they already know, then they don't). Give me the top 3 only."

I pick one. Sometimes I pick none and rerun pass 2 with a different Domain B. Rerunning is cheaper than publishing a mid idea.

What this looked like last month

I needed a week of posts for a designer who sells retainers and kept defaulting to "how to price your work."

Kill list included: pricing posts, "stop undercharging," portfolio tips, "raise your rates" carousels. All banned.

Collision: design retainers x how a neighborhood restaurant runs regulars (not tourists). Constraint: carousel or short video only.

One that survived the FAIL test:

"Tourists ask 'what's good.' Regulars get 'the usual?' Your retainers are stuck in tourist mode if every month starts with a new briefing form."

That became a 5-slide post: tourist vs regular clients, the one question that signals a regular, how to build a "usual" into the contract, a screenshot of a one-line monthly kickoff. It was the best-performing post of her month, and it did not mention pricing once.

Two rules that keep this from becoming a gimmick

1. If I cannot explain the collision in a voice note without looking at the prompt, it's too clever. Kill it.

2. I never publish the raw 12. The raw 12 is compost. The one-line test is the crop.

Asking AI to "be creative" does nothing. Creativity here is subtraction (kill list) plus an awkward pairing (collision) plus a hook that dies if you remove the pairing.

I teach this ideation method, including how to bank Domain B lists so you're never starting from a blank chat, inside the AI Mastery Hub course. For today, take the topic you're stuck on, generate the 15 banned ideas, and collide it with a job you've actually had. Send me the one hook that survives pass 3 if you want a second pair of eyes.

The model already gave the internet its first draft. Your job is to make that draft illegal.

Most creators have a content problem that looks like a production problem. They think they need more posts. Usually they need to stop ignoring the 40 pieces already sitting on their site, half of them outdated, a few of them quietly doing all the work.

I used to publish through the pile. New week, new idea, old posts rotting. Then I ran an audit on a client's blog (and later on my own). We didn't add a single new article for two weeks. Traffic and inquiries both went up, because the good pages finally made sense.

Here's the audit I actually run.

Step 1: Make a stupid-simple inventory (20 min)

Export or copy titles, URLs, publish dates, and any stats you have (views, time on page, replies, inbound emails that mentioned a post). Paste into a sheet. If you have no stats, use 10 titles you remember people mentioning. Perfect data is how this task never starts.

Step 2: Sort into four buckets (15 min, with AI)

I paste the list in:

"Here is my content inventory: [paste titles + one-line topic + any stats]. Sort each piece into exactly one bucket: 1. WINNER — people already find it or reply to it 2. FIXABLE — good topic, weak execution (vague title, outdated, no CTA, buried point) 3. THIN — could be a section inside a winner, not its own page 4. RETIRE — duplicate, wrong audience, or I wouldn't stand behind it today For FIXABLE items, name the single highest-leverage fix (title, intro, examples, CTA, or update facts). Don't suggest rewriting everything. If stats are missing, say you're guessing and why."

I argue with the list. AI will call a vanity post a winner because the title sounds clever. I override those.

Step 3: Only rewrite the FIXABLE bucket, and only one layer (30-45 min per piece)

For each fixable post I paste the full draft:

"Audit this piece as a skeptical [ideal reader] who has 90 seconds. Give me: - What they think the post is about after the first 3 sentences (vs what I meant) - 3 places I went generic - 1 claim that needs a stronger example or a source - Whether the CTA matches the promise in the title Then rewrite ONLY the headline, the first 80 words, and the CTA. Leave the rest. If the bones are bad, say so and don't polish junk. Post: [paste]"

I apply those three rewrites, then add one thing AI cannot know: a sentence from a real client conversation or a number from my own work. That's the difference between "updated" and "worth ranking / sharing again."

Step 4: Merge thin posts, don't "refresh" them

Thin posts get a 301 or a note at the top: "This is now part of [winner]." Combining three mediocre listicles into one solid guide is an audit. Publishing a fourth listicle is procrastination.

Real pass I ran in April

Inventory: 37 posts. Buckets: 6 winners, 9 fixable, 14 thin, 8 retire.

One "fixable" was a 2024 post called "AI tools for freelancers." Title was a magnet for tire-kickers. Intro took 200 words to say nothing. CTA was "follow me."

After the audit:

• Title became "The 5 AI tasks I actually delegate (and the 3 I still do myself)"

• Intro started with the 3 I won't delegate, which is what people argued with in the comments

• CTA asked them to reply with the task they still do manually

Same URL. Same skeleton. Replies that week beat the previous month of new posts combined. We never needed a new article on "AI tools." We needed that one to tell the truth faster.

What I don't do

• Ask AI to "make it more engaging." That produces adjectives.

• Refresh dates without changing the substance. That's a lie to yourself and to Google.

• Audit 100 URLs in one sitting. Six fixable pieces is a good week.

If you have more than 15 published pieces and you haven't looked at them in 6 months, skip this week's new content. Run the four buckets. Fix two posts. Merge one. You'll feel the difference in your inbox before you feel it in analytics.

I teach the full content audit (inventory, buckets, rewrite-the-right-layer, and when to kill a URL) inside the AI Mastery Hub course. For this weekend, pick your 10 most-visited pieces and run step 2. Don't write anything new until the FIXABLE list is shorter.

Your archive is a product. Treat it like one.

I still send cold emails. Not 200 a day. About 8 to 10 a week, to people I actually want to work with. Reply rate sits around 30%. Meetings from those replies: enough that I don't have to live on inbound alone.

The trick is not a "high-converting template." Templates are why cold outreach feels cold. The trick is a research pass so the first two lines could only have been written to that person.

Here's the 8-minute method.

Rule I use before I even open a chat

If I cannot name a specific thing they did in the last 90 days (a post, a hire, a launch, a page on their site that's clearly broken or ambitious), I don't email them. AI cannot invent a reason to write. It can only help me say the reason clearly.

Minute 1-4: Collect three artifacts

I grab:

1. Their latest post or newsletter

2. The homepage headline + the main CTA

3. One extra: LinkedIn about, a job listing, a podcast appearance, whatever is recent

I paste those three things in. Not their entire website. Entire websites make the model summarize. Summaries sound like every other freelancer.

Minute 4-6: The observation, not the pitch

"From these three artifacts, give me 5 observations that are specific to this person or company. Each observation must reference a detail I could only know from this material (a phrase they used, a product name, a mismatch between the headline and the CTA, a post topic). Then pick the strongest one and write 2 opening lines I could send. Constraints: - No 'I came across your profile' - No 'I help companies like yours' - No compliment that could apply to anyone in their industry - Do not pitch me or my services yet If you don't have enough to be specific, say so and stop."

If it says it doesn't have enough, I don't send. That's the system working.

Minute 6-8: Attach a small, useful note

Only after the opening is real do I add a pitch, and even then it's small:

"Using this opening: [paste]. Write a 90-word email. - Sentence 3: one concrete observation about a problem implied by their artifacts (not a lecture) - Sentence 4: one relevant thing I've done, in plain language, no case-study theater - Close: a single question that's easy to answer with a yes/no or a one-liner - Sign-off like a person, not a brand Do not mention 'synergy,' 'value-add,' or 'I'd love to hop on a call.'"

I rewrite the last 20% every time. Usually I cut a sentence and add one fact only I know.

A real email I sent in June (anonymized)

They had just posted about killing their blog because "it never led to demos," and their homepage CTA was still "Read the blog."

Opening I used:

"Your note about killing the blog made sense. The homepage still sends people there as the main action, which is a weird leftover if demos are the actual goal."

Then one line about a similar fix I'd done for another B2B team, and a question: "Are you replacing that CTA, or is the blog still doing some other job I'm missing?"

They replied in 40 minutes. "Ha. Yeah. We forgot." Call booked for Thursday. Not because I was clever. Because the email could not have been a mail merge.

What I never let AI do

• Generate 50 variants for a list. Personalization theater (first name + company name) is still cold.

• Scrape and spray. If you need a list of 2,000, you don't have an offer, you have a numbers problem.

• Write the compliment. "Loved your recent insights on leadership" is spam with nicer grammar.

Volume that actually works for me

10 researched emails beat 100 "quick notes." I block 90 minutes twice a week. That's the whole outreach practice. The rest of the week I do the work I already closed.

If your current sequence starts with "I help X do Y," delete it. Spend 4 minutes on their last post and 4 minutes on the draft. Send fewer. Sound like you read something.

I go through research-led outreach, follow-ups that aren't "just bumping this," and when to stop emailing inside the AI Mastery Hub course. For this week, write 5 emails with the 8-minute pass and compare replies to your last batch of templates.

Cold is not the channel. Cold is the feeling that you didn't look.

Creators love the idea of "passive products" and then spend three weeks designing a Notion dashboard nobody asked for.

The templates that actually sell are not invented in a vacuum. They are the thing you already make for clients, cleaned up, with the client's name ripped out, and sold as a fill-in-the-blank version of your real process.

I shipped my first one on a Thursday afternoon. It still sells. Here's the workflow.

Step 1: Steal from your own delivery folder (15 min)

Open the last 5 client projects. Find the document you rebuild every time with slight tweaks. For me it was a "content operating system": editorial calendar, briefing sheet, and a QA checklist. I was rewriting it from scratch for every client like an idiot.

If you don't have client work yet, skip template-selling until you do. Guessing what strangers need is how you end up with a pretty file and zero buyers.

Step 2: Strip it and turn it into blanks (25 min)

I paste the real doc into AI with:

"This is an internal deliverable I use with clients. Turn it into a reusable template someone else could fill in without me on a call. - Replace every client-specific name, date, metric, and example with [BRACKETS] and a one-line instruction under each - Keep my structure and section order. Do not add cute extra sections I didn't use - Add a 1-page 'how to use this' at the top, written like I brief a new hire, not like a course - Flag any leftover details that still look like they belong to one company Here's the doc: [paste]"

Then I do the unglamorous part: I open the output and delete anything that still smells like Client X. AI misses leftover brand names. You will too if you skim.

Step 3: Make a sample filled version (20 min)

People don't buy empty templates. They buy proof it works. I pick a fake-but-realistic business (not a real client) and prompt:

"Fill this template for a fictional [type of business] so a buyer can see what 'done' looks like. Invent plausible numbers. Label the file clearly as SAMPLE so nobody thinks it's their copy. Keep my section headings identical to the blank version."

I sell both files together: blank + sample. The sample does more selling than my product description.

Step 4: Write the listing like a sales page for one job (15 min)

"Write a product listing for this template. Buyer is a [role] who has this problem: [problem]. Structure: - Headline that names the job the template does, not 'ultimate pack' - Who it's for / who it's not for (2 bullets each) - What's inside, as a file list - What they still have to do themselves (be honest) Do not claim income results. Do not say 'passive income.'"

I cut the hype, add the exact file names, and put checkout on Whop so payment, access, and delivery stay in one place. I am not emailing Google Drive links at 11pm.

Real example:

The content OS template. Price: $29. Time to build the productized version: about 2 hours including the listing. I mentioned it twice in posts I was already writing. First week: 11 sales. Not a life-changing launch. Enough to prove the file was a product, not a homework assignment.

The useful part: two buyers emailed asking if I also had a version for agencies. That told me what to productize next. Your customers will brief your second template if you actually talk to them.

What I refuse to do

• Generate 50 templates with AI and dump them in a mega-pack. Buyers can smell a pile of unused docs.

• Put fake screenshots or fake revenue on the listing.

• Sell a client's confidential work with the names changed poorly. If it isn't yours to productize, it isn't a product.

Minimum viable template

If you want to ship this week: one document you already use, one sample fill, one honest listing, one checkout. Do not wait for a brand, a mini-course, and a webinar.

I go deeper on productizing your delivery (what to productize, how to package it, how to price a first digital product) inside the AI Mastery Hub course. For today, pick the doc you've rebuilt three times and run it through steps 2-4.

You don't need 100 SKUs. You need one file that does a job people already pay you to do.