Product Titles for GEO: Entity + Differentiator + Use Case

Two weeks before a spring promotion, the ecommerce lead at a power-station brand selling into the US and Germany exports the top 40 titles. One reads 2026 New Upgraded Portable Power Station Solar Generator 500W Camping Emergency Backup Battery. That week an assistant is asked whether it runs a 1,500 W espresso machine. It answers with three rivals and a 300 W solar panel: the model had filed the SKU under solar generators, where 500 W reads as panel input, not inverter output. The spec was right. The title put it on another shelf.

A title is not a keyword container; it is the first identity record a model reads — one continuous entity name, a differentiator that fits a comparison column, and a use case a real question can match, in that order.

Four ways a title loses its entity

Each of these passes a coverage check and breaks the noun phrase a model resolves.

Anti-pattern	Typical string	Model reads	Fix
Category stacking	Solar Generator Portable Power Station Camping Battery	Four peer nouns, no head	One noun; rest are attributes
Promo prefix	2026 New Upgraded Premium Portable Blender	Adjectives before the noun	One adjective max; noun up front
Abbreviated entity	PS500 Pro portable power supply	Model searchable, category not	Category spelled out, model next
Exclusive scenes	Espresso machine for home office RV hotel	Four scenes that cancel out	One scene plus one exclusion

All four answer “does anyone search this”, not “what is this”. Keyword tools measure demand; models assign categories. Get it wrong and coverage becomes noise: one noun heads the phrase, four become modifiers, and the SKU drifts onto a neighboring shelf. That work belongs to content and semantics.

The formula: entity, differentiator, use case

The title has three jobs in order: confirm the category, give a comparable reason, attach to one question.

Brand + category noun + model + one or two comparable differentiators + primary use case or constraint

A model reads left to right, building one noun phrase; adjectives wedged inside can split it in two. The entity must be continuous and written in the buyer’s words, not your internal category name or the marketplace back-end label. Differentiators must fill a comparison cell: watt-hours, folded length, material, compatible model, certification class. “Premium” fills none. Two at most — a third dilutes the rest and the model retreats to the spec table. Values live in master data, so fixing master data comes first; each needs a plain reading, the work of attributes as buying reasons.

Length is a symptom, not a target

Judge length by what survives deletion, not by a character count. Drop the use case: does the entity survive? Drop the differentiator: is a comparison still possible?

One product needs three strings — site H1, listing, feed — with different limits. Never copy the longest everywhere or treat the tightest as a standard; keep the entity identical.

Locale	Head noun	What to do
English (US, UK)	after modifiers	Keep the phrase unbroken; model number after the noun
German	inside a compound	Write it in full; no spliced abbreviations
Japanese	after modifiers	Use the local term, not the loanword

Localization is reordering, not translating. Literal translation pushes the head noun to the end and leaves US units on an EU listing: convert inches to centimeters, state 120 V or 230 V where voltage decides compatibility. Keep the category noun; buyers search it and taxonomies index it.

Rewrite the misread twenty, not the catalog

Sequence by the cost of the error, not by SKU count. Old strings live in listings, feeds, reviews, and roundups, and contradict the new one for months. Rank: wrong-category answers, then high-traffic SKUs with a model number, then abbreviated nouns, then long-tail.

Step	Action	Owner	Passes when
1 Extract	Test 50 SKUs against five real buyer questions in two assistants; note each answer’s category	Growth	Mis-attributed SKUs listed
2 Label	Entity, differentiator, or use case missing — or all three	Content	Each label names a word in the title
3 Rewrite	Top 20 only; hide the brand name and reread	Content, category ops	Still answers “what is this”
4 Sync	Site, listing, feed together; retest in two weeks	Ops, engineering	One entity string in three places

Skip step 1 and the ranking is taste; skip step 4 and you own tidier titles with the same problem.

Example: one power station, three titles

Example — hypothetical brand and figures.

Before: 2026 New Upgraded Portable Power Station Solar Generator 500W Camping Emergency Backup Battery

• Site H1: Vela P500 portable power station, 500 W AC output, 1,024 Wh, two days of laptop and camera charging; not for high-draw appliances such as espresso machines or induction cooktops
• Marketplace listing: Vela P500 Portable Power Station, 500 W AC Output, 1,024 Wh, Camping Power
• Feed title: Vela P500 Portable Power Station 500 W 1,024 Wh for Camping

Five changes: one continuous category noun; “solar generator” deleted because no panel ships in the box; the model number beside the noun; two comparable values; one scene plus one exclusion. The entity is identical in all three; only length and order differ. A model merging site, listing, and feed into one record reads three variants as three products, and may cite none. How records connect is the Product Graph; whether it worked is an AI visibility baseline question.

Where this breaks

First, the marketplace fixes the structure. Some categories dictate the order of brand, category, and attributes, plus a character cap. Comply, then optimize what is left.

Second, the model number is the asset. Accessories and compatible parts are searched by number: put it first, move the category explanation to a subtitle.

Third, the title is not the bottleneck. If crawling fails, structured data is missing, or price and stock disagree, the layer below cancels title gains.

Fourth, third parties cite the old name. Reviews and rankings will not follow; build an alias map before changing the string.

Fifth, machine translation at scale. It turns category nouns into literal calques and swaps the entity — worse than leaving the title alone.

This month, take one hero category and 20 SKUs, run five real buyer questions through two assistants, and record the category each answer assigns. Rewrite only the mis-filed ones — one category noun, one comparison value, one scene, one exclusion — then sync site, listing, and feed that week and retest two weeks later.