The 75/125 title system

On 27 July 2026 one long title field became two with two different jobs. The 75 answers what the product is. The 125 answers why it fits. And an over-limit title forfeits the second field entirely.

Fact-checked 22 August 2026·26 min read·6,032 words·By Hymie Zebede
Amazon confirmed Amazon’s own documentation
Amazon research Published Amazon research
Observed Reproducible practitioner testing
Model This framework’s inference
Contents — 13 sections
  1. What is actually confirmed
  2. The Highlights lock, and why it leads every finding
  3. The division of labour: identity in 75, relevance in 125
  4. The placement test for any contested term
  5. The full placement hierarchy
  6. Field spec: the Item name
  7. Field spec: Item Highlights
  8. Field specs: bullets, description, backend, attributes
  9. Noun-phrase construction and the five questions
  10. The phrase taxonomy and the six copy disciplines
  11. Category formulas and worked examples
  12. The migration sequence
  13. What this means in practice

On 27 July 2026 Amazon replaced one long title field with two. The Item name is capped at 75 characters including spaces, in every category except media. Underneath it sits a second field, Item Highlights, capped at 125 characters. Both are search inputs. Amazon confirmed Neither is prioritised over the other, and Amazon's own clarification says search and discoverability are not changed by the migration itself. What changes is architecture, and the architectural change is permanent.

The rumour cycle around this was worse than the change. Two things needed killing early: that the 125 field is a demotion zone, and that the title cap ends Amazon SEO. Content moved into Highlights is not demoted, because Highlights indexes on equal footing. And the cap does not remove lexical relevance, it removes title dependency. Titles stop being storage containers for a category's entire vocabulary. Everything they used to store still has to live somewhere, and now it lives in fields that were always indexed and mostly ignored.

There is one operational fact that decides whether a migration goes well or badly, and it is barely published anywhere. Item Highlights is disabled until the Item name is at or below 75 characters. Amazon confirmed That is Seller Central's own interface message rather than an inference, and the error code is quoted below. An over-limit title does not simply sit there looking long. It forfeits an entire second indexed field of 125 characters. Every over-75 ASIN in a catalog is therefore two problems, not one: a title that will eventually be rewritten by a machine, and a relevance field that cannot be filled at all until the title is fixed.

This page is the full construction specification: what is confirmed, what the lock means, how the two fields divide labour, where every displaced term goes, field-by-field rules for the whole listing, the phrase system that produces the copy, and the migration sequence that stops a rebuild from turning into a visibility cliff. Counting characters by eye is where migrations break, so verify every field in code. The title checker on this site does the count for you, including spaces, and flags the fields that will trip the lock.

What is actually confirmed

Separate the published mechanics from the operating consequences before doing anything else. The left column here is Amazon's; the right column is what it forces you to do. Amazon confirmed on every mechanic listed.

MechanicDetailOperating consequence
Item name limit75 characters including spaces, all categories except media, effective 27 July 2026 (announced June 2026)Roughly 125 characters of legacy title has to be re-homed, not deleted
Item Highlights125 characters, rendered under the title, indexed for searchA second identity-free field for relevance language
Indexation parityBoth fields index; no stated priority between themCopy strategy becomes allocation, not sacrifice
The lockItem Highlights is disabled until the Item name is 75 characters or fewer (Seller Central UI)An over-limit title costs you the whole 125-character field
Listing statusListings remain active, editable and searchable throughoutThere is no emergency. There is a deadline and a rewrite risk, which is a different thing
Review windowBrand owners get 14 days in Review Listing Changes to review, modify or approve AI-generated updatesA weekly sweep of that queue is mandatory; unreviewed drafts publish themselves
EnforcementAmazon can gradually rewrite over-limit listings through the listing enhancerThe rewriter optimises for readability, not for the terms that drive your purchase share
Bulk pathCategory Listings report, pre-populated; uploaded changes apply within roughly eight hoursReview the sheet before upload, never after Amazon has applied it
Brand placementAmazon recommends keeping the brand name inside the Item nameRoughly a fifth of the 75-character budget is committed before product language is written

One observed behaviour belongs alongside those. The automated rewrite is not neutral. Observed It has been seen dropping phrases that carried real purchase share, because readability and purchase-weighted relevance are not the same objective. If you let the enhancer write your title, you have outsourced your identity root to a system that has never seen your Search Query Performance data.

The shape of the change

One 200-character storage container became two fields with two jobs. The 75 answers what is this? The 125 answers why is this right for me? Everything that fits neither goes to a field that was already indexed and already underused.

The Highlights lock, and why it leads every finding

The message, verbatim

Sellers attempting to populate Item Highlights on an over-limit listing report Seller Central returning error code 100476:

“Provide an Item Name that is 75 characters or less to use Item Highlights.”

The field sits under Manage All Inventory → Edit listing → Product Details, and some sellers first encountered it through the AI “View Enhancements” panel, where the item name and item highlights appear together.

Rollout is uneven, and one credible source disagrees

Two things to hold alongside the lock. The field is behaving inconsistently: sellers report entering Highlights, returning 24 hours later and finding the original recommended text restored, and an Amazon forum moderator has acknowledged that “some sellers are currently experiencing system errors or changes not saving.” Observed

And the dependency is disputed. SellerSprite argues Item Highlights is an independent attribute that should be populated regardless of title length, describing the lock as widely reported but not confirmed by Amazon in writing. The error code above is the strongest evidence for the lock and we treat it as decisive — but the disagreement is real, and if you are about to make this claim to a client, verify it in that account first. Contested

76 charactersis the point at which you lose 125 characters of indexed relevance copy, not just one character of title

Take this seriously because the arithmetic is brutal and almost nobody has published it. A listing at 74 characters has 199 characters of indexed, seller-controlled, top-of-page text. A listing at 76 characters has 76. The penalty for being two characters over is not proportional, it is a cliff, and it is enforced in the interface rather than by the ranking engine. Amazon confirmed

This changes how you report an over-limit ASIN. The wrong framing is "this title is 41 characters too long". The right framing is "this ASIN is currently forfeiting its entire Item Highlights field and cannot fill it until the name comes down". Lead every over-75 finding with the lock. It reorders priorities inside a catalog immediately: the ASINs to fix first are not the ones with the worst copy, they are the ones closest to the limit, because they are the cheapest to unlock.

Three consequences follow that are easy to miss.

Sequencing is forced
You cannot write Highlights first and shorten the name afterwards. The field is unavailable. Name first, Highlights second, always, and that ordering propagates through the whole migration plan.
Audits that count characters lie
A catalog audit that reports "average title 118 characters" has understated the loss by a factor of two. Report the number of ASINs with a locked Highlights field and the total indexed characters forfeited across the catalog. That is the number that moves a decision.
The rewrite risk and the lock compound
An over-limit ASIN is exposed to automatic rewriting and is denied the field where the displaced language should have gone. If the enhancer trims the title for you, the terms it removed are simply gone, because you never had the 125 open to catch them.
Do not count by eye

Spaces count. Commas count. Trailing whitespace counts, and pasted cells from a spreadsheet carry it. Non-ASCII characters behave differently again in byte-limited fields further down the listing. Every count in a migration should be produced in code and checked against the field limit before upload. Use the title checker or your own script, but do not trust a character counter that ignores spaces, and do not trust a visual estimate.

The division of labour: identity in 75, relevance in 125

The two fields are not a long title split at a comma. They have different jobs, and mixing the jobs is the most common failure in migrated listings. Model

Item name, 75 characters: identity

Answers "what is this?" in one breath. Brand, defining adjectives, head noun, form, defining spec, variant token last. It must fit a phone screen and parse as speech when read aloud by a voice device. If a listener cannot say what the product is after hearing it once, the field has failed regardless of its character count.

Item Highlights, 125 characters: relevance

Answers "why is this right for me?" as one readable line. Use case, material or spec consequence, the differentiator a shopper compares on, compatibility, audience or context. It is also the string an assistant is most likely to quote when recommending the product, so it should read as language, not as a keyword pile.

The single hardest rule in the Item name is one identity root. That means the purchase-dominant product-type phrase, in the purchase-dominant word order, taken from your own Search Query Performance data rather than from a keyword tool's volume column. Never two product-type synonyms in the same name. "Wall Sculpture" or "Metal Wall Art", not both. Stacking type synonyms produces semantic identity blur: the retrieval layer is being told the product is two different classes of thing, and the era of embedding-based matching punishes that far more than the lexical era did. Amazon research

Word order matters inside the 75 in a way it does not elsewhere. Query tokens match across fields jointly, so a term in a bullet can complete a phrase begun in the name. But exact-phrase adjacency in the name appears to carry additional relevance weight on head terms. Observed That is the argument for placing the purchase-dominant word order in the 75 and nowhere else, rather than distributing the words of the head phrase across two fields and hoping the join is made.

The displaced synonym, meanwhile, has a legitimate home. The second product-type phrase goes into Item Highlights, or into the description if the Highlights are already committed. It is not deleted, and it is not smuggled back into the name.

The placement test for any contested term

Every migration produces arguments about a term that somebody insists must be in the title. There is a single test that resolves them, and it takes ten seconds.

The placement test

Remove the phrase from the name. Does the shopper still know what the product IS?
If no, the phrase belongs in the 75. It is identity.
If yes, but they have lost a reason it fits them, the phrase belongs in the 125. It is relevance.
If yes, and nothing is lost, it belongs further down the hierarchy, or it is a truth failure and belongs nowhere.

Run it on the arguments you will actually have. "Organic" on a food product: remove it and the shopper still knows it is a coffee bean, so it is relevance, unless the category's buyers use organic to separate the product class, in which case it is identity and it stays. "For Living Room": remove it and identity is intact, so it is context and goes to Highlights or bullets. A fibre-weight grade, a fitment standard, a capacity: these are defining specs, and they earn a place in the name only where the category's buyers use them to separate one product class from another. Identity is not the same thing as benefit, and a spec that is merely impressive is a benefit.

The test also settles what to do with the terms that fail it entirely. Distinguish a placement failure from a truth failure. A placement failure is a real, honest term in the wrong field, and it gets re-homed. A truth failure is a term the product cannot support, and it gets deleted with the reason recorded. Deleting an honest term without an address is how a migration loses queries.

The full placement hierarchy

This is the table the whole migration runs on. Terms move down the hierarchy, never sideways into deletion. Model, built on the confirmed field indexation map.

RankDestinationJobWhat it carriesIndexed for search
1Structured attributesEligibility, filters, agentic scoringMaterial, size, colour, audience, features, use, origin, care, certifications, weight, dimensions, package quantityPowers filters, browse and agentic scoring; some fields feed search Observed
2Item name (75)IdentityBrand, one identity root, form, defining spec, variant token lastYes Amazon confirmed
3Item Highlights (125)RelevanceUse case, spec consequence, differentiator, compatibility, audience or context, the displaced type synonymYes, equal input with the name Amazon confirmed
4BulletsEvidence and objectionsExtractable claims with number and unit, audience and use-case modifiers, care, compatibility, guarantee, pack contentsYes, through the last bullet
5Product descriptionLong-tail and synonym reservoirHonest type synonyms, long-tail modifier phrases, secondary use cases and audiences, gift and occasion languageYes, fully, despite being visually displaced by A+
6Backend generic keywordsSilent recallOnly what has no on-page home: misspellings, other-language terms with domestic volume, trade synonyms, complementsYes; byte-limited — anything past ~249 bytes is not indexed
7A+ and image textPersuasion and comprehensionReinforcement of claims already made in textA+ partial and category-dependent; image text is not indexed at all

Two rules govern movement through it. Attributes ship first and verify before any copy changes, because that layer is where eligibility gates read and where agentic systems score. And anything stated only in an image does not exist to retrieval, so a term pushed out of the title and into a graphic has been deleted, whatever the design deck says.

Subject matter, target audience, intended use and special features sit outside the ranked list. The fields exist Amazon confirmed but per-field search indexing is not guaranteed by category. Observed Fill them anyway. They are cheap and they carry real intent. Never promise a client that they index.

Field spec: the Item name

Construction order, in one line: [Brand] [defining adjective only if genuinely defining] [head noun or product type] [form] [defining spec] [variant token last].

  • 75 characters including spaces. All categories except media. Verified in code, never by eye.
  • Brand stays in the name. Amazon confirmed recommendation. Budget for it before writing anything else.
  • One identity root, in the purchase-dominant word order from your own query data.
  • Defining spec only where the category separates classes by it. A momme weight in silk, a fitment standard in auto parts, a capacity in storage. Not a benefit dressed as a spec.
  • Variant token last, and structurally identical across every child in the family. Children contribute terms to a shared family footprint while child-specific tokens index per child. Observed
  • Character discipline. Numerals with units. Two-digit year ranges (09-18, not 2009-2018). In compatibility categories, makes in the name and models in the Highlights.

Banned outright: benefit language, audience language, promotional language, synonym stacking, ALL-CAPS words, competitor brand names, and any claim you cannot substantiate. In regulated categories the compliance pass runs before the copy ships, not after a takedown.

How exposed is your catalogue on this?

Twenty checks across the five layers, scored 0–100, returning a ranked issue list by severity — the same audit we run on client accounts. Free, no signup to see your result, and it runs entirely in your browser.

Score my listings →

Nothing you answer is transmitted or stored. The written report and the six working templates are the optional email step afterwards.

Field spec: Item Highlights

125 characters, and the priority order inside them is: use case, then material or spec consequence, then the key differentiator, then compatibility, then audience or context. Most products will fit two to three real noun phrases and no more.

  • One readable sentence or clause chain. Never a comma-separated keyword pile. This is the string an assistant quotes, and a pile does not quote well.
  • Every phrase mirrored by an attribute or a bullet that proves it. A relevance claim with nothing behind it is a contradiction waiting to be found.
  • The legitimate home for the second product-type synonym displaced from the name.
  • Do not repeat the identity already in the name. Repetition buys nothing. A term is indexed or it is not; the second occurrence adds zero at the lexical layer and burns surface that could have carried a different query.
  • Promote the listing's single best number here if it survives substantiation review.

Field specs: bullets, description, backend, attributes

Bullets: evidence and objections

Indexed through the last bullet. Five bullets, each opening with a distinct evidence category: material or spec, primary outcome, secondary outcome or audience, care and durability and compatibility, guarantee or pack contents. The per-bullet shape is a lead label, then an extractable claim running fact to consequence to audience or use, with a number and a unit, handling an objection where it falls naturally.

The standard every bullet must pass is the read-aloud test: spoken alone with no surrounding context, does it answer a shopper's question completely? Failures are predictable. A referent that lives outside the sentence. A fact split across two bullets. An implication where a statement was needed. A number without a unit. A fact stated only in an image. What an assistant cannot quote from you, it rebuilds from reviews, competitors or generic category content. Observed

Bullets also carry a coverage duty. The constraint modifiers that would not fit in 75 or 125, the audiences, the use cases, the contexts, land here.

Description: the long-tail reservoir

Visually displaced by A+ content on most listings and still fully indexed, which makes it the most under-used field in the catalog. Write 1,200 to 1,800 characters of flowing prose, not a mirror of the bullets. It carries the honest type synonyms, long-tail modifier phrases, secondary use cases and audiences, gift and occasion language, and category-adjacent vocabulary that is true.

Identity-sabotage scan, mandatory

The most damaging copy pattern found in audits: a listing that names a cheaper substitute category or material as itself, usually added "for coverage". It teaches semantic retrieval the wrong identity and places the ASIN on a battlefield it can only lose on price. One anonymised reference pattern showed six self-identifications as the substitute across description and backend, an indexed position on the substitute's six-figure-volume head term at around 0.5% impression share, dozens of clicks and zero purchases, against a market median price roughly one fifth of the product's own. Observed Scan the description, Highlights and backend for substitute category and material names, and for offer contradictions: pack counts, sizes, colours or bundle language that do not match the actual offer.

Backend generic keywords: silent recall only

This field is byte-limited, not character-limited, and limits vary by marketplace. Amazon’s own moderator wording describes truncation: terms past the limit are not saved, and a warning appears beside the field — sellers quote the string “Please reduce your generic keyword length to less than 250 bytes.” Amazon confirmed

A correction to a claim we previously carried

It is widely asserted that an overrun causes Amazon to discard the entire field silently, with no error. We can find no credible evidence for this, and Amazon’s stated behaviour is the opposite. The sources asserting it cite nothing. Treat the real risk as the boring one: bytes are not characters, accented and special characters cost two to four each, and anything past roughly 249 bytes is simply not indexed. Verify the count in code, not by eye.

The field is for terms that have no on-page home. Nothing else. Allocate the remaining budget roughly 60% to specification language (technical synonyms, trade terms, spec phrases), 20% to category generalisation, 10% to own-brand equivalence where policy permits, and 10% to complement contexts. Own-brand terms are a small allocation because the brand field already indexes the brand.

Belongs here: high-volume misspellings, which are not lexically matched anywhere else; other-language terms with real domestic volume; technical and trade synonyms; gift and occasion intent; complement contexts. Banned: duplication of any on-page term, which buys zero additional recall; competitor brand names, which is a policy violation, and check generic-looking dictionary words against registered brands in your niche before assuming they are safe; substitute-category names; pack and size contradictions; plurals and word-order permutations, since singular and plural are usually equivalent and word order is flexible; punctuation dumps.

For an existing field, classify every term into one of five audit classes: DUPLICATE, WRONG IDENTITY, WRONG PRODUCT, WRONG OFFER, KEEP. Rebuild from KEEP only, then add justified new terms each with its source. Audited fields typically waste 50% to 70% of their bytes on duplicates. Observed

Structured attributes: the layer that ships first

Attributes drive eligibility, filters and the substrate that agentic systems score against. They ship before any copy change and are verified before the name is touched.

Source of truth is the Category Listing Report, the full category template and the All Attributes tab. Never the edit screen, which renders a subset. Catalogs look complete because the interface hides fields. Run a three-state census on every attribute: VALUED, PLACEHOLDER, BLANK, plus a wrong-value scan for values that contradict the product or use the wrong metric system for the category. Placeholders pass emptiness audits while polluting the data layer, and a wrong value is worse than a blank, because it places the ASIN in the wrong set and looks fine on every audit that only checks presence.

Priority order: identity attributes (material, type, size), then filter attributes (audience, features, use), then compliance attributes (origin, care, certifications), then logistics (weight, dimensions, package quantity). Every material claim in copy must match its structured field, and every origin claim must match country_of_origin. Assistants already detect these mismatches internally. Observed Treat enforcement as a countdown rather than a hypothetical.

Verify acceptance at 48 hours and read the three outcomes correctly: changed and held is success; unchanged with no error is a silent rejection; changed then reverted means a higher-authority contributor is overwriting you. Stop after two identical rejections and reroute, either a template resubmission or a Brand Registry case with ownership evidence.

Noun-phrase construction and the five questions

The unit of optimisation is not the keyword. It is the complete noun phrase, which carries a coherent product concept a machine can parse, compare and verify. "Large abstract metal wall art for living room" is not a keyword pile. Large is size, abstract is style, metal is material, wall art is the head noun, and for living room is context. Together they resolve to one retrievable concept.

Every phrase set on a listing must answer five questions. Missing one is a coverage hole, not a stylistic preference.

#QuestionWhat it suppliesConstruction rule
1What it isIdentity, the precondition for being considered at allHead noun, product type, category, accurate material and style language. Identity resolves before anything else can be evaluated.
2Who it is forAudience, recipient and gifting intentGenuine audience only. Connects the product to a shopper profile and a recipient.
3Where it is usedRoom, placement, use case, occasionMaps the product to real shopping missions rather than broad category terms. This is where most catalogs are silent.
4What it satisfiesThe constraint behind the clickState the constraint (dimmable, outdoor-safe, lightweight), the meaningful size, the format. The page must then prove the promise.
5Why it is trustedVerifiable evidenceCertifications and compliance evidence, reinforced consistently across name, Highlights, bullets, attributes, images and A+.
The spam line

A phrase the product cannot support is spam, and the modern penalty is not only policy risk, it is verification failure. Claims that contradict the structured record or the review corpus get cross-checked, and an unsupported claim degrades the trust of the claims around it. Model Audience and context language is permitted only where the product genuinely serves it.

The phrase taxonomy and the six copy disciplines

Every noun phrase performs exactly one of six jobs. Classify before placing, and messy optimisation becomes structurally impossible.

1 · Identity
What the product IS. Head noun and product type: "floor lamp", "ignition coil pack". Goes to the Item name.
2 · Attribute
Material, size, form, spec: "brass", "64 oz", "dimmable". Structured fields first, then the name or Highlights where genuinely defining.
3 · Context
Room, occasion, placement: "for reading nook", "bedside". Highlights and bullets, the bridging layer.
4 · Audience
Who it is for: "horse lover", "new parents". Bullets, avatar imagery, "suitable for" lines.
5 · Constraint
Filterable limits: "outdoor-safe", "lightweight", "under-cabinet". Attributes first, then copy.
6 · Proof
Trust and verification: certification and compliance language. Evidence layer, documentation on file before publishing.

Parent and child phrases

Every head noun carries a family of refinements beneath it, and the refinements are where conversational and filtered demand actually concentrates. "Wall art" is the parent. "Large wall art", "metal wall art for living room", "abstract wall art above couch" are children, and the children are what query fan-out and overview cards generate.

The parent goes in the name, once, as identity. The children are distributed across Highlights, bullets, description and attributes, each placed on the surface where it is truthful and useful. Coverage is counted at child level, never at parent level. An ASIN ranked on the parent but silent on eight children has a coverage problem that head-term rank will never reveal. The four-index diagnosis is where that gets measured.

Vague to concrete translation

The most common copy defect is language a human tolerates and a machine cannot use. Anything unverifiable contributes nothing to retrieval, comparison, or an assistant's answer. With the name capped at 75, the space for decorative adjectives no longer exists anyway.

Vague, contributes nothingConcrete: retrievable, comparable, quotable
"Premium quality""18/10 stainless steel, 1.2 mm gauge"
"Long-lasting battery""Up to 40 hours playback, 10-minute quick charge"
"Perfect for any room""Sized for living room, bedroom or entryway walls, 36 in wide"
"Easy to install""Hangs with two hooks, hardware included, no drilling into studs"
"Great for everyone""Fits standard car cup holders, dishwasher-safe for daily use"
"Highly absorbent""Chelated glycinate form, 400 mg elemental magnesium per serving"

The six copy disciplines

Fact, consequence, audience. "Waterproof shell, keeps gear dry through coastal rain, built for outdoor families." The reasoning arrives pre-built instead of being inferred.
Spec density with units. Target eight or more concrete attributes per listing, each in numerals with standard units. Spec density beats keyword density in every field.
Objection answering. Every recurring complaint in the review corpus gets a direct answer on the page, or the assistant answers it from the complaint.
Claim consistency. One source of truth per specification, propagated identically to every surface. A bullet saying 36 in against an attribute saying 34 is not a typo in this system, it is a contradiction both engines can detect.
Reviewer vocabulary mirroring. Language already validated by the review corpus is language the retrieval layer already associates with the product.
Proof before claim. Documentation on file before any certification or compliance phrase publishes. An unsupported claim degrades the claims around it.

Category formulas and worked examples

Character counts below are computed, not estimated, and the brand tokens are deliberately generic. Apply your own naming system. Clarity beats maximisation: a 54-character name that reads cleanly is better than a 74-character name that has been padded to the limit.

CategoryFormulaWorked example, with counts
Supplements (compliance-critical)Name = Brand + Ingredient + Form + Dose + Count.
Highlights = absorption or form claim + structure-function use cases + tolerability
Name (52): "Brand Magnesium Glycinate 400 mg Capsules, 120 Count"
Highlights (114): "Chelated for high absorption, supports restful sleep, muscle recovery and daily calm, gentle on sensitive stomachs"
Beauty and skincareName = Brand + Hero Ingredient + Product Type + Size.
Highlights = skin-type fit + finish or feel + what it will not do
Name (43): "Brand Niacinamide 10% Facial Serum, 1 fl oz"
Highlights (118): "Visibly minimises pores and evens tone for oily and combination skin, lightweight non-greasy formula"
Baby apparelName = Brand + Material + Garment + Size or Pack.
Highlights = certification + skin benefit + practicality + fit
Name (64): "Brand Bamboo Viscose Convertible Footie Pajamas, Newborn, 2-Pack"
Highlights (122): "Buttery-soft, breathable fabric for sensitive skin, double zipper for easy night changes, stretch fit for growing newborns"
Auto parts (fitment IS identity)Name = Brand + Part Type + OE Reference + Position or Set.
Highlights = fitment span + direct-fit claim + material or testing + verification prompt
Name (59): "Brand Ignition Coil Pack, Replaces OE 90919-02244, Set of 4"
Highlights (120): "Direct-fit replacement for 02-09 2.4L four-cylinder applications, OE-spec windings, bench-tested, verify fit by VIN"
Pet nutritionName = Brand + Species and Life-stage + Product Type + Flavour + Weight.
Highlights = complete-nutrition claim + functional benefits + suitability
Name (50): "Brand Adult Cat Dry Food, Chicken Recipe, 7 lb Bag"
Highlights (117): "Complete nutrition for cats 1-7 years, natural fiber for hairball control and smooth digestion, ideal for indoor cats"
Home and decorName = Brand + Size or Scale + Style + Material + Head Noun.
Highlights = rooms + mounting or practicality + occasion
Name (54): "Brand Large Black Metal Abstract Wall Sculpture, 36 in"
Highlights (118): "Modern statement art for living room, bedroom or entryway walls, lightweight powder-coated steel, hangs with two hooks"
Comfort goods and textilesName = Brand + Material + Product + Size (+ Colour).
Highlights = sensory benefit + sleep or skin outcome + care practicality
Name (48): "Brand Mulberry Silk Pillowcase, Queen, Champagne"
Highlights (116): "22-momme silk that's gentle on hair and skin to help reduce frizz and sleep creases, hidden zipper, machine washable"

The before and after pattern

One legacy title, 170 characters, and where every part of it lands. Nothing in the left column is deleted.

Legacy title (170 characters)Migrated structure
"BRAND Large Metal Wall Art Abstract Modern Wall Sculpture Decor for Living Room Bedroom Office Home Decoration Hanging Ornament Gift Idea Housewarming 36 Inch Black Steel"Name (54): "Brand Large Black Metal Abstract Wall Sculpture, 36 in"
Highlights (118): "Modern statement art for living room, bedroom or entryway walls, lightweight powder-coated steel, hangs with two hooks"
Attributes: size 36 in, material powder-coated steel, style abstract and modern, room living room, bedroom, entryway, mounting two-hook
Bullets: occasion and gifting, weight and installation detail, care
Description: the displaced synonym mass, "wall decor", "hanging ornament", "home decoration"

Note what happened to the two type synonyms. The legacy title claimed to be both "Wall Art" and a "Wall Sculpture". The migrated name picks one identity root and the other moves to the description, where it still indexes and no longer blurs identity. Note also that "Gift Idea" and "Housewarming" did not vanish. They became occasion language in a bullet and in the description, which is where occasion intent belongs in the taxonomy.

The migration sequence

Executed in the wrong order, a rebuild produces a visibility drop that looks exactly like an algorithmic penalty and is entirely self-inflicted. Because the failure is an indexation loss rather than a ranking loss, it is detectable at a fixed checkpoint and correctable inside the same week. That is only true if you took the baseline.

1. Baseline first. Archive eight weeks of Search Query Performance for every ASIN in scope, dated. Snapshot the Brand Analytics register and your ranking data. Tag the change date. Without this, the migration cannot be judged and a rollback decision cannot be made on evidence.
2. Term inventory. Every meaningful term in every current field gets two assignments: its function (identity, defining spec, feature, use case, benefit, audience, variant, synonym, claim) and its destination (a named attribute, the name, Highlights, a numbered bullet, description, backend, or DELETE with a recorded reason). No blank rows. Nothing leaves the listing without an address. Separate placement failures, which get re-homed, from truth failures, which get deleted and logged.
3. Attributes ship first. Anything factual moves to its structured field before it moves anywhere else, because that is where the gates read. Verify acceptance at 48 hours before touching copy.
4. Then the name. Identity root from purchase-weighted phrasing, category formula applied, compliance pass run, count verified in code. This step is what unlocks the next one.
5. Then Highlights. Now available, because the name is at or below 75. Relevance phrases, each mirrored by an attribute or a bullet.
6. Then the remainder, in order: bullets, description, backend. Backend is rebuilt last because it can only be written once you know what has an on-page home.
7. Tail ASINs before heroes. Three to five slow sellers first, to validate the pattern where it cannot hurt. Allow two weeks for indexation and four to six for the visibility read. Then heroes, one variation family at a time, identical structure across children, variant token last.
8. Never inside a deal window. Freeze windows around major events. A copy change and a deal in the same window produce a result nobody can attribute, and you have burned the measurement as well as the deal.
9. The +2-week orphan check. Re-pull SQP against baseline. Any query that has gone dark is an orphaned term. Confirm the demand is real, re-home the term, verify. This single check is the entire failure mode of the migration.
10. Verdict at +4 to +6 weeks on family-level impression share, not on single queries. Attribute nothing inside the noise window. Roll back only on confirmed indexation loss that cannot be re-homed, and restore the exact prior version, because a partial revert creates an uninterpretable third state.

Two operational paths run alongside the sequence. The bulk path is the Category Listings report, pre-populated, with uploads applying within roughly eight hours. Amazon confirmed Review the sheet before upload, because eight hours after the fact is not a review, it is an incident report. And sweep the Review Listing Changes queue weekly, comparing Amazon's drafts against your inventory sheet term by term. Retained and moved terms are fine. A dropped term is a rejection. Brand owners have 14 days. Amazon confirmed An AI draft publishing itself unreviewed is a self-inflicted incident, not an Amazon problem.

The migration law

Re-home first, shorten second, zero orphaned terms. Shortening before re-homing is how a catalog loses queries it had earned. Every term that leaves the title has an address before the title is saved.

What this means in practice

Five things worth doing this week, in this order.

Count every Item name in code and rank the catalog by the lock, not by length. Produce two numbers for your catalog: how many ASINs currently have Item Highlights disabled, and how many indexed characters that forfeits in total. Fix the ASINs sitting just over 75 first, because they are the cheapest unlocks. The title checker will do the counting; a spreadsheet formula that ignores trailing spaces will not.
Archive the baseline before you touch anything. Eight weeks of SQP, dated and stored outside the working file. If a migration is already under way without one, stop and pull what history you still have. You cannot diagnose an orphaned term against a baseline you never took.
Run the attribute census on your top twenty ASINs. Pull the Category Listing Report and the All Attributes tab, not the edit screen. Mark every field VALUED, PLACEHOLDER or BLANK, then scan for wrong values. Fix the identity attributes first. This layer ships before a single character of copy changes.
Build the term inventory for one hero ASIN, with no blank rows. Every term gets a function and a destination. The exercise usually reveals two things immediately: a second identity root that has been blurring the listing for years, and a set of honest terms with nowhere to go except the description, which is where they should have been all along.
Put the Review Listing Changes sweep on the calendar, weekly. Fourteen days is the window, so a weekly cadence is the only one that cannot miss it. Compare term by term against the inventory. Approve your version, not the enhancer's.
Diarise the +2-week orphan check on the day you publish. Not "in a fortnight, roughly". A dated calendar entry with the baseline file attached. Any query that went dark gets re-homed and re-verified inside the same week, which is the difference between a small correction and a quarter spent explaining a rank drop. See rank drop diagnosis if the check comes back ugly.

Sources

Primary sources for the confirmed and published claims above. Observations and framework inferences are labelled as such in the text and are not sourced here.

  1. Amazon Search: The Joy of Ranking Products — Sorokina & Cantú-Paz, SIGIR 2016 — www.amazon.science
  2. GeekSeller — July 2026 product title changes — www.geekseller.com
  3. Zentail — Amazon cuts product titles to 75 characters — www.zentail.com
  4. Amalytix — Amazon product title rules — www.amalytix.com
  5. Autopilot — Amazon policy tracker — www.autopilotbrand.com
  6. Amazon Seller Forums — 250-byte search terms announcement — sellercentral.amazon.com

Migrating a whole catalogue is where this gets expensive

Term inventories, attribute sequencing and orphan checks across hundreds of ASINs is grinding work with a two-week failure window. It is a large part of what we do for brands.

Get a free account teardown →
ZBD Growth

Full Amazon channel management for established brands. Catalog, advertising, inventory and cases — run by a team you know by name, reported in depth.

© 2026 ZBD Growth. Selling on Amazon since 2012.Run by Hymie Zebede →