Perfect Listing AI
Draft product listing copy for Amazon, Shopify, Etsy or eBay, then edit it before you publish.
Generate Listing
What this tool does
Perfect Listing writes the text half of a product page. You type a product name, pick a category, a marketplace and a tone, and it asks the model for five title options, three descriptions at different lengths, seven bullet points, ten keyword tags, a meta description, a rough SEO score and three suggestions for improving it. Whatever comes back is rendered as plain text you can select and paste into Shopify, Amazon, Etsy or eBay once you have edited it.
It exists because listing copy is the part of the job that scales worst. One product is fine, twenty products is a weekend. What you get back is a draft, not a finished page, and its value is that the blank page is already filled in when you sit down to edit.
How it works
- You fill in the brief. Product name is the only required field. Category, marketplace and tone change the wording rather than the structure. The optional extra info box is where you put features, materials, measurements and the buyer you have in mind.
- You sign in. The generate endpoint checks for a session and returns a 401 without one, so the sign-in box appears before the request goes out. Once the request is accepted, your email is recorded as a lead and the product name and the marketplace are written to the usage log, before the model is called.
- The fields become one prompt. They are packed into a single request to DeepSeek alongside a system instruction to act as an e-commerce copywriter and to answer with one JSON object using fixed keys. Temperature is set to 0.3, which is low on purpose: the model sticks close to the brief instead of inventing an angle you did not ask for.
- The reply is parsed. The request is attempted up to three times, waiting one second and then two seconds between attempts, so a timeout or a busy server is retried rather than surfaced. The server then pulls the JSON block out of the model's answer and reads the fields. If the call still fails, or the answer comes back unparseable, you get a short placeholder listing with wording like "Premium [your product]" rather than an error. If a result looks unusually thin and generic, that is what happened, and running it again usually fixes it.
- The page renders each field as its own block so you can copy one piece at a time. The generated copy itself is not saved — only your email, the product name and the marketplace go into the log — so take what you want before you leave.
What you get back
The prompt asks for 5 titles, 3 descriptions, 7 bullets, 10 tags, 1 meta description, 1 score and 3 tips, and the page renders whichever of those fields come back. Nothing in the code counts them, so a run can return fewer, and the placeholder listing you get after a failed call is shorter than all of them. Below is a written-out run so you can see the format before you spend a generation on it. This is an illustration of the requested shape, not a real user's result.
The prompt asks for the three descriptions at roughly 100-150, 300-400 and 600-800 characters, so you can pick the one that fits the slot you are filling. The lengths you actually get are whatever the model wrote; nothing measures them.
Where it helps and where it does not
It helps most when you already know the product and just need the words: a supplier page with broken English, a batch of variants that all need copy, or a listing you want to rewrite in a different tone for a different marketplace.
Here is what it cannot do, stated plainly:
- It fetches nothing. No supplier page, no product URL, no reviews, no competitor listings, no search volume. It writes from the words in the form and from what the model already knows about products like yours.
- It will state specifications it cannot know. Battery life, waterproof ratings, certifications, warranty length. If you did not supply the number, treat any number in the output as a placeholder to verify or delete. This is the edit pass that matters.
- The SEO score is an opinion, not a measurement. The same model that wrote the copy assigns the score in the same breath. A high number does not predict rankings.
- The keyword tags are not researched. They are plausible phrases, which is a reasonable starting list, but check them in a real keyword tool before you build a strategy on them.
- Character limits are requested, not enforced. The prompt asks for lengths that fit common marketplace rules; nothing in the code truncates the result. Amazon title limits in particular vary by category, so count before you publish.
- One product per run. There is no CSV import, no bulk queue, no image generation, and no Shopify or Amazon connection. You copy and paste.
In categories where compliance wording matters, such as supplements or anything making a health claim, write the copy yourself and use this only for the boring parts.
What to do before and after this step
Writing copy is the third thing you do, not the first. Score a shortlist of candidates with Product Scout before you commit to one, then run the survivor through the Product Validator to check demand and saturation, because a perfect listing for a dead product is still a dead product. Once you have copy you are happy with, Cross-Border Translation will localize it for markets outside English, and Viral Ad takes the same product brief and turns it into ad hooks rather than store copy.
Frequently asked questions
Is the product listing generator free?
Yes. Generating a listing costs nothing, there is no card field anywhere in the flow, and the code has no daily cap once you are signed in. The running cost is the model call, which we pay for.
Do I need an account to use it?
Yes. The generate request is rejected unless you have a session, so you need a free account with an email and password. Your email is stored together with the product name and the marketplace you picked, which is how the usage log works. The generated text itself is not saved, so copy it before you close the tab.
What data does it use to write the listing?
Only the words you type: product name, category, marketplace, tone and the optional extra info. It does not open a supplier page, read reviews, or look up search volume. The keywords it picks come from the language model's training data, not from a live keyword tool, so treat them as suggestions to check rather than researched terms.
How accurate is the SEO score?
The score is written by the same model that wrote the listing, in the same response. It is a self-assessment on a 0-100 scale, not a measurement against a live index or against the listings you are competing with. It is useful for comparing two drafts of your own copy and useless as a prediction of where you will rank.
Can I paste the output straight into Amazon or Shopify?
Mechanically yes, it is plain text. Practically you should edit first. Check every factual claim against your actual product, because the model will write plausible specifications it has no way of knowing. Also check length against the platform's real limits, since the character counts in the prompt are a request to the model, not something the code enforces.
Will the copy look the same as everyone else's?
If you only type a product name, probably close to it. Two people entering "wireless earbuds" with the same tone give the model the same thin brief and get similar drafts. Filling the extra info box with real specs, materials and the buyer you are targeting is what makes the output specific, and editing the result is what makes it yours.