Writing product titles for AI agents

A Shopify product title has two readers. One scans for size and color. The other ingests for embedding. The first wants brevity; the second wants identity. Pick one and the other suffers — but the suffering is asymmetric. A title that satisfies the AI agent and inconveniences the human costs a fraction of a second per scan. A title that satisfies the human and confuses the AI agent costs the surfacing entirely.

This guide is the editorial pattern that holds across catalogs — the structure, the trade-offs, and the eight rewrites that fix the most common failure modes.

The pattern

Five components, in this order, with brand-specific judgment on which to omit. An apparel example annotated:

Written out as a string:

[Brand] [Product Name] — [Defining Attribute] [Material/Style] [Variant Detail]

Examples, with each component labeled:

• Allbirds [brand] Wool Runner [name] — Mizzles [defining attr] Merino [material] Charcoal [variant]
• Vacheron Constantin [brand] Overseas [name] — Automatic [defining attr] Stainless Steel [material] 41mm Blue Dial [variant]
• Patagonia [brand] Better Sweater [name] — 1/4 Zip [defining attr] Recycled Polyester [material] Birch White Medium [variant]

The em-dash separates the identity (brand + name) from the descriptors. The descriptors are ordered by what an AI agent weights most: defining attribute, material, variant.

What changes by category

The order is not universal. Three category-level adjustments:

• Apparel uses the order above. Material and variant detail carry significant weight in shopping-intent queries.
• Beauty moves the active ingredient earlier: [Brand] [Product Name] with [Active] — [Variant]. Example: The Ordinary Niacinamide 10% + Zinc 1% — 30ml.
• Electronics drops material and adds model number: [Brand] [Product Name] [Model] — [Defining Attr] [Variant]. Example: Sony WH-1000XM5 — Noise-Cancelling Wireless Headphones, Black.

The eight rewrites

The patterns that come up most often when auditing Shopify product titles, with the rewrite that fixes each:

1. The “marketing copy” title

Before: The Best Sweater You'll Ever Own

After: Acme Outfitters Heritage Sweater — Cable-Knit Merino, Forest Green

Why: marketing-led titles embed against marketing-intent queries (“best sweater”), not shopping-intent queries (“women’s merino sweater”). The agent surfaces what gets the embedding match.

2. The keyword-stuffed title

Before: Wool Sweater Cashmere Blend Soft Warm Winter Pullover Crewneck Womens Ladies

After: Acme Outfitters Cashmere Crewneck — Wool Blend Pullover, Women's

Why: keyword stuffing was a 2010s SEO tactic that AI agent embeddings penalize. The dense list of synonyms reads as low-quality and pushes the embedding toward generic queries.

3. The size-forward title

Before: Medium Wool Sweater

After: Patagonia Better Sweater 1/4 Zip — Recycled Polyester, Birch White, Medium

Why: size belongs in the variant slot, not the headline slot. Size-forward titles surface in size-specific queries but lose general shopping-intent surfacing entirely.

4. The variant-first title (the parent-product trap)

Before: Black Medium Cotton T-Shirt

After: Acme Outfitters Heavyweight T-Shirt — 100% Organic Cotton, Black, Medium

Why: parent-product page surfacing requires the parent’s identity in the title. Variant-first titles imply this is a variant page; the agent treats it as one.

5. The bare-name title

Before: Heritage

After: Acme Outfitters Heritage Sweater — Cable-Knit Merino, Forest Green

Why: catalog systems sometimes export only the model name. The agent has nothing to embed against. Even when the on-page H1 carries the context, the title is a separate field and gets used independently.

6. The all-caps title

Before: ACME OUTFITTERS HERITAGE SWEATER

After: Acme Outfitters Heritage Sweater — Cable-Knit Merino, Forest Green

Why: most AI agents normalize case before embedding, but several ranking signals downgrade all-caps fields as a low-quality marker inherited from web SEO. No advantage to caps; small downside.

7. The dimension-overloaded title

Before: Acme Outfitters Sweater 38" Chest 26" Length 23" Sleeve

After: Acme Outfitters Heritage Sweater — Cable-Knit Merino, Forest Green (move dimensions to attributes)

Why: dimensions are attribute-table data, not title data. Putting them in the title pushes out the components that drive surfacing.

8. The promotional title

Before: 🔥 SALE — Acme Outfitters Sweater — 30% Off!

After: Acme Outfitters Heritage Sweater — Cable-Knit Merino, Forest Green (sale data lives in offers, not the title)

Why: promotional decoration in titles never helps surfacing and often hurts. AI agents read the title for product identity and the offers schema for pricing — moving promo content to where it belongs.

Where to put what got cut

Material, dimensions, fit, gender, country of origin, certifications — these belong in attributes (Schema.org additionalProperty, Shopify metafields, GMC feed attributes), not the title. The title is identity; the attributes are specification.

The right title length is the length that fits the pattern. The character count varies by category — apparel titles tend to run longer than beauty titles, for example, because more components are needed to express identity. The target is identity completeness, not a character-count quota.

The contrarian take

Most ecommerce SEO content recommends titles 50–60 characters because that’s what Google’s SERP truncates to. AI agents do not truncate at 60 characters. They embed the full string. A 90-character title that includes the right components surfaces better than a 55-character title that’s been cut to fit a SERP display constraint.

The trade-off: longer titles look worse in Google’s SERP. Shorter titles look better in Google’s SERP and surface less in AI shopping. For brands where AI traffic is now larger than direct-Google traffic, the call is to optimize for surfacing, not display. The SERP display can be controlled separately through og:title if needed.

Title quality is one of the six dimensions Lumio scores at 20% weight in the AI Readiness Score — tied with Description density and Conversational fields as the highest-weighted dimensions in the base framework.

Where it breaks

Three failure modes the pattern doesn’t fully solve:

• Single-product brands. A brand with one product faces a different trade-off — the title carries marketing weight that a multi-product catalog distributes across pages. The pattern still applies, but the brand description may need to absorb more identity work.
• Highly visual products with weak text identity. Some categories (art prints, decor, jewelry one-of-ones) don’t have descriptive attributes that match shopping-intent queries. For these, multi-modal optimization (image alt text, image structured data) carries more weight than title optimization.
• Multi-language catalogs. Translation tools often produce titles that pattern-match poorly across languages. The pattern’s component order may need to flex by language; manual review of the top 50 products per language outperforms full automated retranslation.

What to ship this week

1. Audit the catalog’s top-revenue 50 products against the pattern.
2. Rewrite any that fail.
3. Push the changes through Shopify and update the GMC feed.
4. Wait two weeks. Re-check ChatGPT and Perplexity for the affected product types. Lift is typically visible in a 14-day window for catalogs above 60 readiness.

The remaining catalog can wait. The top 50 carry most of the surfacing weight, and the pattern is easy enough to apply to the rest as products come up for refresh.