How Do You Research an Amazon Listing Before You Write It?
What I checked before AI wrote a word of a small manufacturer's first Amazon listing, and what each check changed.
If you're about to write an Amazon listing, answer four questions first: what shoppers type, what page one says, what Amazon allows today and what buyers say after buying. Use Amazon's search estimates, every big seller's keywords, Amazon's rules pages and buyer reviews or forums. Each source finds something the others miss. Titles have been capped at 75 characters since July 27, 2026, except for media. Then ask the owner four questions at most, with the question blocking the draft first.
Key takeaways
- Check whether a search tool returns the whole answer.
- Separate Amazon's rules from agency advice.
- An empty review report doesn't prove there are no reviews.
- Ask whether every version matches what your draft says.
- Check limits with a script before approving the copy.
What do shoppers actually type into Amazon?
Amazon's own monthly search estimates can show you the words shoppers use, including names you haven't heard. I pulled those estimates through DataForSEO, a service that sells keyword data. A keyword is basically a word or phrase someone searches for. The first pull cost about eight cents.
This was for a small two-person manufacturer getting their first product onto Amazon. They had sold the product in person for years. But the listing started with no copy, no reviews and no sales history. They hired me in September 2026, and nothing is live yet.
The search data turned up a second name for the product. Nobody on the project had heard it, including me. That name went into the listing. The name used within the trade had no measurable searches, so I left it out of the title.
In my experience, businesses often name things the way their trade does. Buyers may use something quite different. So I check the search data before deciding which words deserve the limited space at the top of the page. Familiar wording isn't enough by itself.
Claude Code ran the searches, scripts and analysis on September 25, September 30 and October 1. The keyword data across those days cost roughly 64 cents. The hours I spent reading what came back cost a lot more than that.
What does page one already say?
Read the big sellers' listings and keyword lists together, rather than treating the leading seller as the whole market. My first pass covered one competitor's keywords, one full listing and ten search results. That sounded more complete than it was.
I asked whether the AI had checked every competitor on page one, "to make sure we are using all the keywords they are using in their listings." The answer was no. So I went back through the research. I also checked why the search had returned only ten products.
The data connection requested a shortened answer by default. For Amazon results, that meant the first ten products, without a warning. Asking for the full response returned 124 placements and 94 different products. Some products appeared more than once, so those weren't 124 separate competitors.
Then I pulled the full keyword lists for the 18 biggest sellers. That covered every product selling at least 100 a month in the research. I stopped adding sellers when the last five brought no new search terms. The stopping point came from the results, rather than a round number I picked beforehand.
That left 277 different search terms to check against the draft. Amazon matches individual words within a search against listing words, so I checked them word by word across the fields. Every core term was already there. The fuller pass added six small terms I had missed.
Competitor brand names stayed out, even when they had searches. Amazon bans them in every field of a listing. And matching a seller's keywords doesn't give you permission to copy their claims. This product would sell at four to eight times page one's price, like for like, so every line had to earn the difference with something true.
What does Amazon allow in 2026?
Amazon's current title limit is 75 characters, including spaces, for every category except media. Amazon announced the change on June 10, with a July 27, 2026 start date, in its Seller Central announcement. Most categories previously allowed up to 200 characters.
The same announcement added Item Highlights. This is a searchable field showing with the title, with another 125 characters for details that help buyers compare options. So you have another place to put useful information. You don't have to squeeze everything into the title.
Two months after the change, page one still had plenty of long titles. Amazon rewrites old titles gradually. The first keyword report told me to follow those long titles. Only checking Amazon's own announcement showed that the advice was out of date. My finished draft's title is 74 characters.
I also had two AI research reports built in NotebookLM, Google's source-based research tool. Between them, they used 58 cited web sources and nine YouTube tutorials. That sounds like plenty. But they mixed Amazon's rules with agency opinion and some invention.
One report said a brand without Brand Registry, Amazon's brand registration program, must list as "Generic." Amazon's own guidance on generic products says "Generic products are unbranded products that do not belong to any identifiable brand," and this product has a brand. Another report supplied precise example claims that had no factual basis. They read just like the accurate sections around them.
A long source list doesn't tell you which statements to trust, which is why I wrote a separate post on how to tell if AI research is any good.
So I labeled each rule by who said it: Amazon, agencies or the law. Two often-repeated limits came from agency consensus, meaning agreement among agencies, rather than Amazon's own words. I kept that distinction visible:
- The hidden search terms field takes 249 bytes, according to agency consensus. Bytes measure storage size, which can differ from character count.
- Agency consensus says exceeding that limit can cause Amazon to ignore the whole field without warning.
- Agency consensus also says Amazon counts only the first 1,000 bytes across the bullets for search matching.
Then the rules went into a checking script, a small program that tests the draft. It checks field limits, banned characters, banned words and competitor brands. A draft that fails can't quietly move on. I also read the category's rules before writing, because ordinary words can count as regulated claims.
The checked draft used 74 of 75 characters in the title and 121 of 125 in Item Highlights. The description used 1,901 of 2,000 characters. Hidden search terms used 217 of 249 bytes. Those numbers show the draft passed the checks. They say nothing yet about whether it will sell.
What do buyers say once they own the product?
Reviews and discussions show experiences that a keyword list doesn't capture. After the search research, I realized we hadn't looked at what people were talking about after buying. That needed its own pass.
I read several sources, because each had different gaps:
- Sixteen Reddit threads with about 290 top comments.
- About 150 comments under eight YouTube how-to videos and 19 top comments under one TikTok video.
- Amazon's "Customers say" summaries for the two biggest sellers.
- Sixty-nine Amazon reviews from seven competing products.
- 323 Google Shopping reviews from three established competitors, including 29 one-star reviews.
That Google Shopping pull cost three cents. The wider research also used about 40 credits from a forum-and-video data service and two scraper credits. A scraper is software that collects information from web pages. Again, the collection cost wasn't the same as the cost of checking what came back.
For this project, Amazon's "Customers say" box was the best single source. It sits on the public product page and counts positive and negative mentions by topic. Amazon explains in its review highlights article that the summaries use written reviews from verified purchases and require several buyers to share an opinion. I still read individual comments alongside it.
It also settled the category's biggest complaint: the product coming loose. The two best sellers' summaries counted 247 mentions of how well it stays on, and 119 of them were negative. No keyword list showed that. This product attaches a different way, so the line about how it attaches moved up to bullet three.
Four separate buyers said under how-to videos that there were no instructions in the box. That changed the packaging plan to include an instruction card. Gift buyers also appeared repeatedly. Gift words went into the hidden search terms and stay out of the ads.
There were two collection problems I had to catch. The first report said competitors had "no reviews" because an unreliable source returned nothing. A later run returned 69 reviews. Another request reached Amazon's sign-in page instead of the reviews, but received a success code.
So I check the returned page or data before accepting the report. A successful request can still contain nothing useful. An empty answer tells me the collection needs checking before I conclude that buyers have nothing to say.
What can only the owner tell you?
The owner has to confirm facts about the product that search tools and reviews can't establish. I keep those emails to four questions at most. The question I can't continue without goes first. In my experience, a long list is easier to leave unanswered.
On October 1, I sent three questions and got answers within hours. One answer changed three fields. My draft described one version as though it applied to every version. So I now make a point of asking whether every unit is the same.
On October 3, the first of four questions asked whether every part was made in the US. The title depended on the answer. The FTC's Made in USA guidance allows an unqualified claim, meaning one without a limiting explanation, only when all or virtually all of the product is US-made. That includes components.
The owner answered all four on October 4. Each answer changed or confirmed a line in the listing. Questions that would only add another line waited for the sign-off email. I wanted the necessary answers before asking for everything else.
How do you start on your own listing?
Start with the four questions, one source each, before AI writes a word. I put the four questions into the free Before You Write worksheet, so you can keep the sources and unanswered points in one place. It's also the first step of the small business marketing and SEO work I do for clients. So far this project has a checked draft and nothing live, so there are no sales or rankings to report yet.
Frequently asked questions
How long can an Amazon title be in 2026?
Since July 27, 2026, Amazon limits titles to 75 characters, including spaces, in every category except media. Check Amazon's current rules before publishing. Older competing listings may still display longer titles while Amazon updates them.
What is Item Highlights?
Item Highlights is a searchable field that appears with the title. Amazon allows 125 characters for information that helps shoppers compare products. The draft in this project used 121 characters.
Should I copy a top seller's title?
Use it to find wording to investigate, then write from your own facts and current rules. In this project, following the leaders' long-title pattern would have exceeded the new limit. Check other big sellers too, because one listing won't show every useful search term.
Where can I see buyer complaints for free?
Start with public Amazon reviews and the "Customers say" summaries where available. Read relevant Reddit discussions and comments under how-to videos too. Check whether several buyers describe the same experience before changing your copy.
How many questions should I send a manufacturer?
I send four at most, with the question blocking the draft first. Ask about facts you can't confirm elsewhere. Save optional additions for sign-off, rather than making the owner work through a long email before answering the necessary question.
Want a system built for your business?
Everything here comes out of real work, mine and my clients'. If you want something custom, let's talk.
Get in touch