You know the competitor outsells you.
You cannot say which part of their page does it.
So the usual move is to copy the parts that look good, the title structure, the fifth photo, the way they word the guarantee, and hope something sticks. It rarely does, because the thing actually working is often not on the page at all. It is in the reviews: the complaint they never fixed, spelled out by forty buyers in the buyer's own words, which is exactly the sentence that belongs in your bullets and is sitting in public where anyone can read it.
From their reviews to your bullet, without a slide in between
Give it the raw pages, not the question
Asking a model what it thinks of a competitor gets you a market summary that would describe anybody. Saving their listing pages, ad copy, price table and review export into a folder and pointing at it gets you something about that one competitor. Ten pages at once is a normal read here, and five competitors run as five agents in parallel rather than one long conversation where the first one has fallen out of memory by the time you reach the last.
Rank the gaps by how many people said it
One loud reviewer and a structural weakness read identically until you count them. Each complaint comes back next to its share of their negatives, so the thing worth rebuilding your page around separates from the thing that happened to one person on a bad day. Their five-stars get the same treatment, because what buyers keep praising tells you which claims are already saturated and not worth fighting over.
Write it into your page while you still remember why
The gap hands off to the listing skill, which drafts the bullet, the FAQ answer that pre-empts the doubt, and what belongs in the information slot on photos three to five. Same client, same folder, so the evidence is still attached when you review the draft and you can see which review the line came from.
Not a data subscription, and not a way around the rules
Both are places where a competitor tool is easy to oversell, so it is worth being blunt.
It does not know their sales or their ad spend
Helium 10, Jungle Scout and Keepa pay for marketplace data that exists nowhere else, and none of it is in here. If your question is how many units they moved last month, that is a subscription. If it is what their buyers keep complaining about and what your page should therefore say, that is this. They are not substitutes for each other.
Their homepage number is their marketing
Prices change without notice, and a figure lifted from a competitor's own copy stays labelled as coming from their copy rather than being promoted into a fact you repeat on a comparison page. If you are going to publish a number about another company, it has to come from third-party testing or their own documentation.
Nothing is fetched from behind a login
Logins, captchas, anti-bot measures and paid API limits are not worked around, so the material is whatever the platform lets you export or view. In practice that is enough, since public review pages and your own category exports carry most of the signal, but if you cannot obtain something cleanly the honest answer is that the analysis runs without it.
Before you replace your saved prompt
I have a competitor prompt that works. Why package it as a skill?
Because in March you will want to rerun it and compare. A prompt you retype drifts a little each time, so two reports come back in different shapes and cannot be put side by side, which is exactly what you need when the question is whether they changed anything. A skill fixes the structure and the requirement to cite a source, and it can run on a schedule.
I already use Claude Code. Does this replace it?
No, it drives it. Claude Code, Codex, OpenClaw, OpenCode and Hermes can each be run from inside the same conversation, and anything already written in the Claude skill format keeps working unchanged. What you gain is five readers on five competitors at once, a project folder that survives between sessions, and the seller connectors in the same window.
Can it write the comparison page too?
It drafts one and marks which figures came from where, so you can see what still needs verifying before it goes live. That labelling is the useful part: a comparison page built on numbers taken from a competitor's marketing copy is the kind of thing that earns you a complaint rather than a conversion.
Does this only work for Amazon?
No. It reads whatever you save, so Shopify storefronts, Etsy listings, eBay pages, TikTok Shop and independent sites all work the same way. The seller connectors cover thirty-odd platforms if you want the material pulled instead of saved by hand, each through an application you register yourself.
Their reviews are public. That is the whole trick.
Free, macOS and Windows. Bring your own model key or use an Orkas plan.