Store owners describe the same three failures: it sounds robotic, it can't hold
their brand voice, and it invents specs that don't exist. These three prompts are built against
that last one — the one that turns into a return, a chargeback, or a false-advertising problem.
The shared mechanism is one instruction: where a
detail is missing, write [NEED: …] instead of guessing. A model with a gap
fills it with an average of your category. Told to flag the gap instead, it hands you a to-do list
rather than a liability.
1 · Product description
5 min · hero, bullets, specs, objection
You are writing a product description for an ecommerce store. Accuracy outranks
persuasion: a claim I cannot prove costs me a chargeback and a bad review.
PRODUCT:
- Name:
- What it physically is (materials, dimensions, contents):
- What it does for the buyer:
- Price:
- The 3 things a buyer must know before they'd feel safe clicking Add to Cart:
RULES:
1. Lead with the change in the buyer's life, not the object. The object is evidence.
2. Every benefit must trace to a concrete feature I listed. If you cannot trace
it, cut it.
3. Do NOT invent: measurements, materials, certifications, review quotes, origin,
counts, awards, or comparisons to named competitors. If a detail would help and
I did not give it to you, write [NEED: <the detail>] instead of guessing.
4. Banned phrases: elevate, game-changer, unlock, seamlessly, must-have,
revolutionize, "look no further", "whether you're a ___ or a ___".
5. Read the price. A $19 product earns 80 words. A $400 product earns 300 and
needs to handle risk explicitly.
OUTPUT:
A) HERO - 2-3 sentences. Sentence one names the buyer's trigger moment.
B) BULLETS - 5, each under 14 words: <Benefit> - <the feature that proves it>.
C) SPECS - plain table, only facts I supplied.
D) OBJECTION - one sentence answering the biggest reason they won't buy.
SELF-CHECK (print this, do not skip):
- Claims I made that you did NOT give me evidence for: <list, or "none">
- [NEED] items: <list>
- Sentences that would survive if the product name were swapped for a
competitor's: <list them - these are the generic ones, rewrite them>
Why the last line is the whole trick. Any sentence still true with
a competitor's name in it describes the category, not your product. That is precisely the texture
people mean by "sounds AI-written," and you can now see it listed instead of sensing it.
2 · The AI-tells edit pass
10 min · run on copy you already have
Here is copy for my store. Do not rewrite it yet. Diagnose it first.
COPY:
Flag every instance of:
1. CATEGORY SENTENCES - would still be true with a competitor's name swapped in.
Quote each.
2. BANNED REGISTER - elevate, unlock, seamless, game-changer, must-have, curated,
revolutionize, "in today's fast-paced world", "look no further".
3. ADJECTIVE STACKS - 2+ adjectives before a noun. Almost always replaceable
with one number.
4. UNSUPPORTED CLAIMS - anything my brief does not evidence. This is the
false-advertising list.
5. HEDGES - "helps to", "can assist in", "designed to". Either it does it or
it doesn't.
6. RHYTHM UNIFORMITY - 3+ consecutive sentences of near-identical length. The
most reliable machine tell there is, and the one people hear without being
able to name it.
7. SYMMETRY TICS - "It's not just X, it's Y" and "From X to Y".
Output a table: | quote | tell type | why it reads as AI | fix |
Then give me the rewritten copy applying only the fixes, changing nothing else.
SELF-CHECK: what percentage of sentences did you flag? If under 15% on
AI-generated copy, you were too lenient - go again and be harsher.
Use it on human copy too. Category sentences and adjective stacks
aren't a machine problem. They're what anyone writes at 5pm.
3 · Product page teardown
10 min · ranked fixes, and permission to stop
Act as a conversion analyst reviewing a product page. You are paid to find
reasons people leave, not to compliment the page.
PAGE (paste the full visible text top to bottom, in order, and describe each image):
DATA (write "unknown" for anything you don't have):
- Sessions / conversion rate / add-to-cart rate:
- Mobile vs desktop split and CR for each:
- Top exit point:
- Average order value:
TASK:
1. Reconstruct the page as a first-time mobile visitor experiences it in the
first 5 seconds. What do they know? What is still unanswered?
2. List every unanswered question a buyer needs resolved before paying. Order
by how many buyers it blocks.
3. For each, say where on the page it should be answered and why there.
4. Flag friction: unclear pricing, hidden shipping cost, weak trust signals,
ambiguous variant selection, slow hero, buried reviews, unclear returns.
5. Rank ALL findings by (buyers blocked) x (ease of fix). Give the top 5 only.
RULES:
- Ground each finding in text I actually pasted. Quote it.
- Do not recommend a redesign. Recommend edits I can ship this afternoon.
- If the data suggests the problem is upstream - wrong traffic, wrong price,
wrong product - say so plainly instead of optimizing a page that isn't the
bottleneck.
OUTPUT - table:
| # | Finding (quote) | Buyer question left open | Fix | Effort | Impact |
Then: "If you only do one thing this week: ___"
SELF-CHECK: which finding are you least confident in, and what would you need
to see to confirm it?
That upstream clause costs me a sale and earns the other four findings.
Most CRO advice assumes the page is the problem because the page is what you're looking at. A page
converting at 0.4% is sometimes doing its job for traffic that was never going to buy.
Before you run any of these: they expect a store brief — who specifically buys, what
they're comparing you to, and what you can actually prove. Without it you'll get category copy
no matter how good the prompt is. There's a free builder for that; it takes about ten minutes
and it's the half that does the work.
You just used three. There are sixteen more — product pages, ads, email, SEO, ops — plus four end-to-end workflows that chain prompts together for real tasks. The full kit is $29 one-time: