Last updated: September 2026
Two hundred forty-nine bytes. That is the entire budget Amazon gives your backend search-terms field: one line, roughly 35–40 short keywords, no overtime. Most listings fill it with words already in the title.
The waste starts upstream. Amazon keyword research is the discipline of finding the exact terms shoppers type, judging which ones carry buying intent, and assigning each term to the field where it can actually rank. Done well, it compounds for years. Done with guesses, it quietly caps every listing in your catalog. Once keywords are settled, the X vs Y comparison template shows how to turn that research into the buying-stage pages shoppers search next.
AI changed the economics of that discipline. Clustering two hundred raw phrases, classifying intent, and packing a byte-perfect backend line used to eat a weekend; it now takes an afternoon — and the free data sources are better than most sellers realize.
This guide covers the stack: official search data from Brand Analytics, a manual reverse ASIN method that costs nothing, prompts for every research step, a keyword-to-field assignment matrix, and the bilingual math for sellers running Amazon.com and Amazon.sa together.
Short answer: Start with Brand Analytics search data (free with Brand Registry), expand and cluster terms with AI, mine competitor ASINs manually, classify intent, then assign each keyword to title, bullets, description, or backend — keeping the backend line under 249 deduplicated bytes. Verify with an indexing test before spending on ads.
The 249-byte rule, decoded
The backend search-terms field allows one line of up to 249 bytes — not characters, bytes. Helium 10's definitive explainer breaks down what counts: letters count, including stop words; spaces and punctuation do not. Anything outside the limit is simply not read by Amazon's system.
Three consequences follow. First, duplication is pure waste: repeating words already in your title or bullets burns budget on terms you already rank for. Second, the field is for what shoppers type but you cannot say on the page — misspellings, synonyms, Spanish or Arabic phrasings on US and Saudi traffic. Third, bytes punish non-Latin scripts: in UTF-8, an Arabic character costs 2 bytes where an English letter costs 1, so an Arabic backend term eats double.
The field used to be generous — 5,000 characters until 2020, when Amazon cut it to roughly 250 to force relevancy over stuffing. Jungle Scout's optimization guide still finds sellers debating the old number; the operative one is 249 bytes. A realistic packed line — synonyms, misspellings, regional variants, deduplicated against your visible copy — lands in the 230–249 range with room to spare.
What is Amazon keyword research?
Amazon keyword research is the process of discovering the search terms customers actually type, estimating demand and competition for each, and placing them in the fields Amazon indexes: title, bullets, description, backend terms, and PPC campaigns. It is the difference between writing what you would say and writing what shoppers search.
It is not Google SEO. Jungle Scout puts the intent split plainly: people go to Google to search how to do something, while people go to Amazon to search for something to buy. An Amazon SEO keyword strategy filters for commercial intent — "stainless steel dog bowl" beats "how to clean a dog bowl" even if the informational phrase has more volume.
Volume itself has two flavors worth separating. Exact match volume counts how often a phrase was searched as-is in the past month; broad match volume folds in close variants and synonyms. A keyword tool showing one but not the other flatters weak terms. As a worked Amazon search terms example, "insulated water bottle" might show strong exact volume while "flask for cold drinks" only registers under broad match — one is a proven query, the other a maybe.
Finally, research assigns placement. Long-tail keywords — three-plus words, lower volume, sharper intent — are where new listings win first, because head terms are locked by incumbents with years of reviews. The matrix later in this guide maps every keyword type to its home field.
Why the sellers who rank treat keywords as inventory
Keywords are stock, not decoration: each one is a placement you own or concede. The commercial context makes ownership expensive to skip. In Jungle Scout's 2025 Consumer Trends report, 63% of shoppers ranked price and discounts as the most influential Amazon purchase factor, and 76% remain worried about inflation. Price-sensitive shoppers compare harder, which means the visible keyword match — title and bullets — decides who enters the consideration set.
The competitive pool is deepening. Amazon reports independent sellers have generated more than $2.5 trillion in cumulative sales, with a record number surpassing $1 million in annual sales in 2025. Professionalized competitors mean the free data edge matters more: Brand Analytics — including Search Query Performance and Top Search Terms — is free for brand-registered sellers and shows real impressions, clicks, and cart adds per query.
And the bilingual edge is still cheap. Amazon.sa launched June 17, 2020, and Saudi Arabia's e-commerce market is projected to grow from USD 251.3 billion in 2025 to USD 732.9 billion by 2034 (IMARC Group). Most US and UK sellers never build an Arabic keyword map; the ones who do face thin competition on high-intent terms.
How to do Amazon keyword research with AI: 6 steps
1. Pull the official data first
Brand Registry is free with an active or pending trademark, and it unlocks Brand Analytics: Search Query Performance and the Search Terms report show what customers typed, with impressions, clicks, and cart adds per query. This is Amazon's own demand data — start here, not with a tool's estimates.
2. Expand with AI seed generation
Feed your product and five seed terms into the expansion prompt below. The output buckets — exact intent, problem-aware, comparison, gift — map to different listing fields and PPC campaign types. Mark every three-plus-word phrase as long-tail candidate.
3. Mine competitor ASINs for free
Reverse ASIN lookup does not require a subscription. Copy the titles and bullets of the top five competitors in your niche, paste them into the competitor-mining prompt, and get the most repeated meaningful phrases back, deduplicated. Paid tools like Helium 10 Cerebro or Jungle Scout Keyword Scout automate this and add volume — the manual version still finds the phrases.
4. Classify intent and competition
Run the classifier prompt across your merged list: transactional, informational, or navigational; low, medium, or high competition. Ten minutes here prevents the classic mistake of spending a title slot on a research phrase nobody buys from.
5. Assign every keyword to a field
Use the matrix below. Primary to title, secondaries to bullets, long-tail to description, synonyms and misspellings to backend, competitor brands to PPC only. Then hand the assignments to the listing-copy step — our Amazon listing writing guide turns the map into copy.
6. Verify indexing before you spend
Amazon indexes a term only after it processes your listing. Test it: search your keyword plus a nonsense word on the marketplace — if your ASIN appears in results, the term is indexed. Diagnose non-indexed terms (placement too deep, brand-new edits, policy flags), then feed proven winners into exact-match PPC.
AI prompts for every research step
Each prompt works with any capable model. In ArWriter they are saved in the prompt library and run bilingually — English output for Amazon.com, Arabic and English for Amazon.sa — from $4.99/month.
Seed expansion:
Given product [PRODUCT] and seed keywords [LIST], generate 40 Amazon search terms real shoppers would use, split into: exact-intent, problem-aware, comparison, and gift-intent buckets. Mark which are likely long-tail (3+ words). Output as CSV.
Competitor mining:
Here are titles and bullets from 5 competitor listings: [PASTE]. Extract the 25 most repeated meaningful phrases (ignore brand names), estimate relative frequency, and flag which my product [PRODUCT] could also target.
Backend builder:
Pack these keywords [LIST] into ONE backend search-terms line under 249 bytes. No repetition of my title keywords [PASTE], no stop words, no competitor brands. Show the final byte count. Note that Arabic words count double bytes.
Intent classifier:
Classify each keyword [LIST] as transactional, informational, or navigational, and estimate expected Amazon competition (low/medium/high). Output a table with a recommended placement: title, bullet, description, backend, or PPC exact.
Cluster-to-fields:
Group [KEYWORD LIST] into 5 semantic clusters, name each cluster, and assign: 1 primary keyword to the title, 3 secondary to bullets, the rest to description or backend. Explain each assignment in one line.
Bilingual keyword mirror:
Build a bilingual keyword map for one ASIN sold on Amazon.com and Amazon.sa. Input: [ENGLISH KEYWORD LIST]. For each term, output the English term, the Arabic phrase real Saudi shoppers would type (search intent, not literal translation), and any transliterations. Mark terms that only exist in one language.
The keyword-to-field assignment matrix
Placement is where most research dies — sellers find great terms then spray them everywhere. The matrix keeps every Amazon SEO keyword in exactly one home:
| Keyword type | Example | Field | Why there |
|---|---|---|---|
| Primary keyword | stainless steel dog bowl | Title, opening bytes | Heaviest ranking weight on the page |
| Secondary keyword | slow feeder bowl | Bullets, one each | Context without stuffing the title |
| Long-tail keyword | non slip dog bowl for large breeds | Description | Catches late-stage, high-intent searches |
| Synonyms and misspellings | doggy bowl, dog bolw | Backend (249 bytes) | Invisible to shoppers, still indexed |
| Seasonal and gift terms | gifts for dog owners | Backend plus PPC | Time-boxed demand, swap seasonally |
| Competitor brand names | rival brand + product | PPC only, exact and negative | Banned from organic fields by policy |
Two rules keep the matrix honest: never repeat a backend term that already sits in visible copy, and never spend a title slot on a phrase the classifier flagged informational. The assignment prompt above automates the mapping; you approve it.
Reverse ASIN lookup without a paid tool
A reverse ASIN search answers the only question that matters in competitor research: which terms does a ranking listing actually hold? Paid tools reverse-engineer it from databases; the manual method reconstructs it from the page itself, because competitors put their ranking keywords in their own titles and bullets.
The procedure takes about an hour per niche. Pick five competitors: same category, similar price, page-one regular results (ignore sponsored slots). Copy each title and bullet block into a document. Run the competitor-mining prompt to extract repeated phrases with rough frequency, then deduplicate against your own list.
What the free version gives you: the phrase set your niche considers important, and relative emphasis. What it cannot give you: search volume and rank tracking over time — that is the paid layer, worth buying only after the free method stops surfacing new terms.
Cross-check frequency against Brand Analytics where you can, and eyeball Amazon's own autocomplete: type a seed letter-by-letter and note the suggestions, which reflect real query volume. Between the Search Terms report, autocomplete, and competitor extraction, the zero-cost stack covers a launch.
What the research stack costs
| Layer | Option | Cost | Best for |
|---|---|---|---|
| Official demand data | Brand Analytics Search Query Performance | Free with Brand Registry | Real impressions, clicks, cart adds |
| Fresh phrasing | Amazon search-bar autocomplete | Free | Long-tail and seasonal wording |
| Competitor intel | Manual reverse ASIN plus AI extraction | Free (AI plan) | Zero-budget niche mapping |
| Volume and tracking | Helium 10 Magnet or Cerebro, Jungle Scout Keyword Scout | Paid subscription | Scale, volume data, rank tracking |
| Writing and clustering | ArWriter bilingual AI | From $4.99/month | Clustering, intent labels, Arabic and English copy |
The honest sequence is top-down: exhaust the free rows first, buy tooling when catalog size makes hours expensive, and keep the writing layer constant. For a deeper look at using frontier models for search research, see our Grok SEO keyword research walkthrough.
One ASIN, two keyword languages
Amazon.sa shoppers search in Arabic, in English, and in transliteration — often in the same session. An English-only keyword map concedes every Arabic query; a literal-translation map concedes every transliterated one. The fix is mirroring, not translating.
Build the bilingual map with the mirror prompt above: for each English term, capture the Arabic phrase shoppers actually type and its transliterated cousin. Arabic-first terms belong in the Arabic marketplace's visible fields; transliterations and English terms fill the backend, where the 2-bytes-per-Arabic-character rule makes budgeting matter.
A concrete pattern: the English term "bluetooth headphones" might be searched on Amazon.sa as the English phrase, as the Arabic phrase for wireless headphones, and as an Arabic-English hybrid typed one way by locals and another by expats. That is three distinct query streams for one product feature — and three indexable placements instead of one.
The same logic extends beyond Saudi Arabia — Amazon.ae and Amazon.eg carry the same bilingual behavior, and our ecommerce localization guide covers the wider playbook for adapting copy market by market.
How an Austin kitchen brand rebuilt its keyword map in one afternoon
Priya Raman co-owns a modular-kitchenware brand in Austin, Texas — cutting boards, prep stations, cabinet inserts. Her ads were bleeding: an ACOS hovering around 41%, concentrated on head terms her organic rank had no chance to support.
She ran the free method on a Saturday. Manual reverse ASIN extraction across six competitors produced 180 raw phrases; the clustering prompt collapsed them into 61 unique terms across five semantic clusters. The classifier moved three research-intent phrases out of her title plan, and the backend builder rebuilt her search-terms line from a half-duplicated mess into a full deduplicated line packed near the byte ceiling.
Two months later the numbers told the story: ACOS down to about 33%, page-one organic placement for nine long-tail terms, and the research files reusable for every new SKU launch. Total tooling cost for the project: her existing AI plan — the paid keyword suites never entered the budget.
Her process note for anyone copying the approach: the research only paid off because placement followed. The same 61 terms, sprayed across every field, would have produced the same flat pages she started with. The matrix, not the list, was the asset.
Frequently asked questions
What is Amazon keyword research?
It is the process of finding the search terms shoppers actually type, judging demand and competition for each, and placing them in the fields Amazon indexes — title, bullets, description, backend terms, and PPC. The goal is owning the terms that match buying intent, not the terms with the biggest raw volume.
What is the Amazon search terms character limit?
Technically it is a byte limit, not characters: 249 bytes, one line. Spaces and punctuation do not count; letters do, including stop words. Arabic characters cost 2 bytes each in UTF-8, so Arabic backend terms burn budget twice as fast as English ones.
What are Amazon backend keywords and do they still work?
Backend search terms are the hidden field behind your listing, and they still feed indexing. They work best for misspellings, synonyms, and regional phrasing you cannot put in visible copy. They fail when sellers repeat title keywords, stuff brands, or exceed the byte limit — Amazon ignores the overflow.
How do I do a reverse ASIN lookup for free?
Copy the titles and bullets of five page-one competitors, then have AI extract the most repeated meaningful phrases with rough frequency. Deduplicate against your own list and cross-check against Brand Analytics. Paid tools add volume data and tracking; the phrase discovery itself costs nothing.
How many keywords should I put in backend search terms?
As many distinct, relevant terms as fit under 249 bytes after deduplication — typically 35–40 short English keywords, fewer with Arabic terms at double bytes. Exclude anything already in your title or bullets, stop words, and competitor brand names, which violate the search terms guidelines.
Do plural and singular keywords count separately on Amazon?
Amazon's public documentation does not confirm a single behavior, and results differ by category. The practical play: check which form appears in Brand Analytics queries, prioritize that one, and add the other to the backend if byte budget allows rather than spending visible field space on both.
How do I know if my keyword is indexed on Amazon?
Search your keyword plus a nonsense word — for example your keyword with a random syllable appended — on the marketplace. If your ASIN appears, the term is indexed. If not, check placement depth, give Amazon time to reprocess recent edits, and retest before changing the listing.
What to do next
- Enroll in Brand Registry, then export the Search Terms report for your category.
- Run the seed-expansion prompt and the free reverse ASIN extraction on five competitors.
- Merge, deduplicate, and classify the full list with the intent-classifier prompt.
- Assign every keyword with the matrix — one home per term, no spraying.
- Pack and byte-check the backend line; verify indexing with the nonsense-word test.
- Feed proven winners into exact-match PPC and re-rank weekly with a keyword tracker.
- Mirror the whole map into Arabic for Amazon.sa while the terms are fresh.
ArWriter runs the clustering, classification, and bilingual mirroring in one workspace from $4.99/month — start at app.arwriterai.com or see pricing. When the map is done, turn it into copy with the listing guide and then upgrade the page with A+ Content.
Sources
- Jungle Scout: Amazon Keyword Research — exact vs broad match volume definitions and the Google-vs-Amazon intent split.
- Helium 10: Use Amazon Search Terms Effectively — the 249-byte rule, what counts, and the 5,000-character history.
- Amazon: How to Create Product Listings — Brand Analytics and Search Query Performance as free official data.
- Helium 10: Reverse ASIN Searches — how reverse ASIN lookup works in paid tools.
- About Amazon: Independent Sellers — seller ecosystem scale, $2.5 trillion cumulative sales.
