Shopping Title Structure and Attribute Enrichment

What a title is for: Retrieval, not voice. Google truncates the visible title at roughly 70 characters on mobile Shopping surfaces, so the first 70 decide whether an item is matched and clicked and the rest decides only whether it is matched. We rewrite titles for items that already draw impressions and lose them. An item with almost no impressions across 30 days does not have a title problem, it has an attribute problem or a price problem, and part of this work is telling you which one you have before anybody edits a string. What we read before writing anything: Search terms from Shopping and Performance Max, sorted into query classes: brand plus model, category plus attribute, occasion, and the Spanish "para" queries that carry most kids and family demand, regalo para niña de 3 años, mochila para primaria. Then the Merchant Center best sellers and price competitiveness reports at product level, because a title rewrite on an item losing to a benchmark price 22 percent below yours changes nothing and we would rather not bill for it. Token order: Brand first where the brand carries the category and buyers type it. Category first where demand is unbranded, which in Mexican kids and family retail is most of it. Size and color enter the title only when they are the query. For children's apparel talla is the query. For furniture and most home goods it is not, and putting it there costs you the characters that matched. The attributes titles get blamed for: size_system and size_type declared per country feed, age_group and gender aligned to how Google separates infant, toddler and kids, material and pattern where unbranded demand uses them, item_group_id so variants compete as one item instead of cannibalising each other, product_type kept as your own merchandising taxonomy and not as a duplicate of google_product_category, and custom_label_0 through 4 carrying margin band, weeks of stock cover, season and price band so bidding can act on the things the title cannot say. Sizing and color in Spanish: Tallas arrive mixed: 6-9 meses, 2 años, talla 4, and the US 4T sitting next to it. We unify to one system per country feed and declare it rather than letting Google guess. Color names get a controlled list behind them: rosa mexicano, vino, arena and hueso stay in the title where they carry the brand's voice, while the color attribute holds the normalized value the matching engine needs. How we know it worked: The affected set is split by a stable hash of item_id, one arm held back, three to four weeks, and impressions and click-through read per query bucket rather than as one revenue number. If the set is too small to separate the effect from noise, we say so before the month is spent, not after. Not included: Campaign management and bidding. Product photography and image editing. Product detail page copy. Translation beyond es-MX and English. Building the attribute data itself when it does not yet exist anywhere, which is enrichment work we do separately and will not fold into a test. Who this is not for: Catalogs where size, color and material have never been captured as fields. Brands that want machine-generated titles across the catalog with no holdout, because that is a text volume exercise and it is not what we sell. Teams who will overrule a winning variant because it does not sound like them; that decision is theirs to make, and it makes the test worthless.