AI Shopping: Ecommerce Teams Test Product Titles in Merchant Center

Write each title as a clear noun phrase followed by disambiguating specs such as model, size, color and material, and put AI-generated titles in Google’s structured_title attribute. Apply the pattern to a sample of SKUs, then run a feed preview or an agent simulation before you touch the rest of the catalog. That one habit serves shoppers and AI shopping agents at once.

TL;DR:

• Google requires titles generated by AI to use structured_title, with digital_source_type set to trained_algorithmic_media; titles written by people use the standard title attribute.
• Google display surfaces often truncate titles at 50 to 60 characters, so front load the brand or product type, model, and decisive specification.
• Pilot 20–50 SKUs across categories, verify claims against product data, and inspect Merchant Center previews for truncation and feed warnings before broader changes.
• Then publish to a group, leave comparable SKUs unchanged, and track agent visits, click through, search success, and conversion for at least two weeks.

Why AI-aware titles matter now

A shopper scanning a results page can quickly read a title. An AI shopping agent reads it differently: it parses the title alongside your structured metadata, compares it with what the user asked for, and decides whether your product is a credible match. Both readers need the same string to work, which is why product titles for AI cannot be written for machines alone.

Think about what each reader needs. A person wants to recognize the product type fast and spot a detail that signals fit, like a size or a brand they trust. An agent wants unambiguous attributes it can map to a request such as “waterproof trail running shoes, women’s, size 8.” A title that opens with a clear noun phrase and follows with specs gives the person a quick read and gives the agent clean tokens to match.

This is also where many catalogs quietly lose ground. Titles built for old keyword-stuffing habits (“BEST Cheap Shoes Free Shipping Sale”) confuse agents because they bury the actual product under noise. Titles that are too short (“Trail Runner”) fail the other way: they give an agent little to disambiguate from similar products.

The shift is also measurable. AI-driven visits can be tracked as their own channel, so a title change is no longer a matter of taste. You can compare agent selection rates and attributed traffic before and after, and decide with evidence. Our broader guide on optimizing listings for AI shopping agents covers the full picture beyond titles, but the title is the best place to start because it is the smallest change with the widest reach. It appears in your feed, on your product page, in search results and in the context an agent weighs when it ranks options.

Naming strategists make a related point about AI products themselves: names work best when they explain what the product does rather than sound futuristic, a theme Fast Company’s naming coverage develops. The same principle holds for listing titles: descriptive beats clever, every time.

Google rules: title attributes and handling AI-generated titles

If you sell through Google Merchant Center, two attributes matter here: title[title] and structured_title[structured_title]. The title attribute is for text written by a person: your merchandisers, copywriters or brand team. The structured_title attribute is for text produced by a generative AI model.

The distinction is not cosmetic. Google’s product data specification says AI-generated titles belong in structured_title, with the digital_source_type sub-attribute set to trained_algorithmic_media. It also requires the content to stay accurate and free of promotional text. Skipping this step is the kind of feed mistake that can get listings downgraded or removed, as Google’s feed requirements make clear.

Here is how to decide which attribute to use:

• Written by a person, even if an AI tool suggested edits you rewrote: use title[title].
• Generated by an AI model and used as produced or lightly edited: use structured_title[structured_title] and set digital_source_type to trained_algorithmic_media.
• Mixed workflows: pick one rule for your team and document it, so reviewers do not guess. If a model drafted the string, treat it as AI-generated.

The content rules apply either way. Keep titles factual and describe the product, not the offer. That means no promotional text such as “Free shipping” or “Best price,” no ALL CAPS shouting, no decorative symbols, and no gimmicks meant to catch the eye. Those tricks do not help an agent and they make human shoppers distrust the listing.

Accuracy is the quiet requirement underneath all of this. If your generation tool invents a material, a capacity or a compatibility claim, the title is wrong no matter which attribute carries it. Always check generated text against your product data before it reaches the feed.

One practical tip: run a small batch of titles through a Merchant Center preview first and read the diagnostics. Catching a flagged attribute early is far cheaper than discovering it across the catalog. The next section gives you the pattern to draft against so that what you submit is already compliant.

A practical hybrid title pattern you can implement immediately

The pattern is simple enough to put on a sticky note: [Primary noun phrase] + [disambiguating specs]. The noun phrase tells a human and an agent what the product is. The specs separate it from every similar product in your category and in your competitors’ catalogs.

A format such as “Insulated Water Bottle, Stainless Steel, Matte Black, Model [number]” does both jobs. The first words name the product plainly. Everything after them is data an agent can match against a query and a shopper can use to confirm the choice. Use a comma or a plain separator rather than decorative symbols, so parsers do not trip on odd characters.

Follow these rules when you draft:

• Lead with the thing itself. Start with the product type or the brand plus product type, depending on how your shoppers search. If buyers hunt by brand, brand first; if they hunt by need, noun first.
• Front-load exact-match terms. Brand, model number and the one or two specs that most often decide a purchase (size, capacity, compatibility) go early.
• Use real attributes only. Material, color, size, weight, quantity and model number. Skip adjectives you cannot verify.
• Cut promotional language. “Premium,” “best” and “amazing” tell an agent nothing and take space from useful specs.
• Keep one canonical order. Pick the sequence of attributes for each category and hold it across the catalog, so titles are predictable to parse.
• Avoid parsing hazards. Skip emoji, repeated punctuation, stray slashes and stacked brackets.

For writing the human-facing side of titles, the principles in this practical guide to writing SEO titles complement the pattern well, particularly on balancing keywords with readability.

Pro Tip: Read each title aloud as if a shopper typed the product into a search bar. If it sounds like a search query with the right nouns and specs, it will usually serve both a person and an agent.

Do not try to make one title carry every variant. Color and size variations belong in structured attributes and variant-level titles, which the next section explains alongside truncation behavior.

Character limits, truncation behavior, and front-loading rules

Where a title is cut off decides which words people and agents see. Google’s documentation notes that product titles are typically truncated at around 50 to 60 characters on display surfaces, and it recommends front-loading critical keywords and distinguishing attributes such as brand, material, color and size within that window. Treat that range as a design constraint rather than a hard cap: your stored title can run longer, but the visible part has to carry the sale.

So build the first 60 characters as a complete, self-sufficient title. If a shopper saw only that fragment, they should still know the brand or product type, the model or key spec, and the single attribute most likely to decide fit. Everything after is a bonus: useful for matching and for agents that read the full string, but it will not do the persuading on a crowded results page.

A practical priority order for the first 60 characters:

1. Brand or product type (whichever your shoppers search first).
2. Model number or product line.
3. The decisive spec: size, capacity, compatibility or material.
4. Color or pack size, if there is room.

Anything that is not a deciding factor drops to the back half or moves to a dedicated attribute.

Long variant families are where titles bloat. A jacket sold in many sizes and colors does not need a title that lists them all. Put color, size and material in the structured fields your platform provides, and give each variant a concise title that names only the attributes that change. The parent product keeps a clean, stable noun phrase.

Resist the temptation to pad titles to a fixed length. A concise title that is accurate and specific beats a longer string stuffed with synonyms, because stuffing hurts readability and can trigger feed warnings. Check your shortest and longest titles in a Merchant Center preview, and confirm that the cut-off lands after the words that matter. If a truncated title ends mid-spec (“Stainless Steel Wa…”), reorder it.

Feed and data hygiene that prevents “no results” and improves AI matching

Good titles sit on top of clean data. If the underlying product names, model numbers and attributes are inconsistent, no title pattern will rescue discovery.

The problem starts with how people search. Google’s guidance notes that shoppers often do exact-copy searches, pasting a product title or model number straight into a search box, and that missing or inconsistent model numbers degrade the result. Baymard’s research on e-commerce search query types documents the same weakness from the UX side: many sites fail to support exact-match or synonym queries, which hurts discoverability when model numbers or product-type words differ between listings. A study summarized by Baymard’s search usability findings points to missing model numbers, inconsistent titles and poor synonym handling as recurring causes of failed search tasks.

An agent facing your catalog runs into the same wall. If listings use inconsistent model-number formatting for the same bottle, a precise query may find one and miss another.

Use this hygiene checklist before you generate or rewrite anything:

• Canonical product name. One approved name per product, used in the title, on the page and in the feed.
• Model numbers everywhere. Include the manufacturer model or SKU in the title where shoppers would search for it, and keep the exact formatting, hyphens included.
• Synonym coverage outside the title. Capture alternate terms (“sneakers” and “trainers,” “couch” and “sofa”) in a secondary title field, taxonomy or search synonyms, rather than cramming them into one string.
• Structured attributes for the details. Color, size, material, pack count and compatibility live in their own fields, so the title stays short.
• Consistent units and spelling. Pick “oz” or “ounce,” “in” or “inch,” and apply it catalog-wide.
• No contradictions. The title, the attribute fields and the page copy should state the same facts.

Doing this work first pays twice. Your own site search improves, and the product data agents read becomes cleaner and easier to match. Our overview of product listing optimization operations goes deeper on keeping that data consistent at scale.

Workflow: generating titles with AI tools, constraints to use, then test and iterate

AI tools can draft titles quickly, but speed without guardrails produces inventions and policy violations. Run a fixed sequence so every batch gets the same checks.

1. Preflight your data. Confirm the canonical product name, mandatory attributes (brand, type, size, material, color) and known model numbers for every SKU in the batch. Fill gaps before generating, because a model will happily invent whatever you leave blank.
2. Write constraints into the prompt. Give the tool a character ceiling, the exact fields it must include, the order to use, and a list of banned phrasing: promotional claims, ALL CAPS, symbols, and any attribute not in your data. Tell it to output only values you supplied.
3. Generate in small batches. Start with 20 to 50 SKUs across different categories rather than the whole catalog, so patterns of error show up early.
4. Check compliance and accuracy. Compare every title with your product data. Route anything AI-generated into structured_title with digital_source_type set to trained_algorithmic_media, as Google’s product data specification requires. Reject titles with unverifiable claims.
5. Preview and search-test. Use Merchant Center previews to confirm how titles display and truncate, then paste titles and model numbers into your own site search to confirm exact matches return the right product.
6. Roll out to a small slice and measure. Publish to a limited set of SKUs, keep a control group untouched, and observe agent-attributed visits, click-through and conversion over a fixed window.
7. Iterate. Keep what wins, revert what loses, and feed the lessons back into your prompt constraints.

Pro Tip: Save your best prompt and your rejected outputs together. The rejected titles show you exactly which constraint to add next, and over a few rounds your prompt becomes a reusable style guide.

If you want to test how agents respond to a store before and after changes, the approach in our guide to making product pages work for AI shopping agents shows how to validate changes against agent behavior rather than guessing.

Copy-ready examples and templates across common categories

Concrete before and after pairs make the pattern easier to adopt. Each “after” leads with a plain noun phrase and follows with verifiable specs, and each keeps its most important words inside the first 60 characters.

Apparel

• Before: “AMAZING Summer Dress Best Seller!!”
• After: “Women’s Linen Midi Dress, Sage Green, Size M, Short Sleeve”

Electronics

• Before: “Super Fast Charger Cheap Free Shipping”
• After: “USB-C Wall Charger, Fast Charging, Multiple Ports, White”

Home goods

• Before: “Luxury Bedding Set Perfect Gift”
• After: “Cotton Duvet Cover Set, Queen, Light Gray”

Consumables

• Before: “Healthy Tasty Snack Yum”
• After: “Dark Chocolate Almonds, Resealable Bag, Unsalted”

Notice what changed. The “before” titles lean on opinion and urgency. The “after” titles state what the item is and the attributes a buyer or agent would filter on. Nothing in them is a claim you would need to prove.

Use these short templates as starting points, replacing the brackets with values from your product data:

• Apparel: [Brand] [Gender] [Product type], [Material], [Color], [Size], [Fit or sleeve]
• Electronics: [Brand] [Product type], [Key spec], [Connector or compatibility], [Model number], [Color]
• Home goods: [Product type], [Material], [Size], [Pattern or finish], [Color], [Pack count]
• Consumables: [Brand] [Product], [Flavor or variety], [Net weight or count], [Package type], [Dietary attribute]

Adapt them with these notes:

• Variants: put the variant-defining attribute (size, color) in the variant title and structured fields, and keep the parent title free of per-variant lists.
• Bundles: state the bundle contents plainly (“Starter Kit: Brush, Comb, Case”) and keep counts accurate, rather than writing “value pack.”
• Compatibility: if a product only works with certain models, name that compatibility in the title, because it is a deciding spec and a common agent query.
• Brand placement: keep the brand first when shoppers search by brand, and after the noun phrase when they search by need.
• Regional wording: match the terms your shoppers actually use, such as “sneakers” versus “trainers,” and cover alternates through synonym fields.

If a template produces a title longer than your visible window, trim the least decisive attribute rather than shortening the noun phrase.

How to validate title changes and which metrics to watch

Treat every title change as an experiment with a defined start, a control and a stopping rule. Without that, you cannot tell whether a lift came from your titles or from a seasonal swing.

Watch these metrics:

• AI-attributed visits: sessions arriving from AI shopping agents and assistants. This is the closest read on whether your titles help agents choose you.
• Organic and collection click-through rate: shows whether the human-facing title works on results pages and category listings.
• Site search success rate: the share of searches, especially exact-copy and model-number searches, that return the right product.
• SKU-level conversion: confirms that more clicks are turning into orders rather than curiosity.

For the rollout, change a small, representative slice of SKUs across several categories, and leave a comparable group untouched. Measure for long enough to cover your normal buying cycle and at least a couple of full weeks, so weekday and weekend patterns do not distort the result. Set guardrails before you start: if conversion or click-through drops past a threshold you choose, revert that slice and review.

Avoid these common mistakes:

• Changing titles, images and prices at once, which makes cause impossible to isolate.
• Rewriting the whole catalog before the pilot reads out.
• Ignoring feed diagnostics after publishing, where policy warnings can quietly suppress listings.
• Optimizing for agents at the expense of readability, so humans bounce.
• Never refreshing: agents and shopper language shift, so schedule a periodic review.

Log each change with its date and the SKUs it touched. That record turns a messy year of tweaks into a clear history you can learn from.

Why we simulate before we scale

Our view is simple: do not guess what an AI agent will do with your titles, watch it. Titles feel like a creative decision, yet their effect shows up in selection behavior that you can observe. When we simulate how agents evaluate a listing, the debate over word choice turns into evidence about which version wins and why.

That is why we push simulation first and small rollouts second. A simulation reveals which title elements an agent leans on before a single customer sees the change, and a limited rollout then proves the effect with real traffic. Together they replace opinion with measured wins, and they keep a bad rewrite from spreading across thousands of products.

We also think teams should stop treating AI visibility as a passing trend. Agents are becoming a real route to purchase, which makes visibility with them a channel worth owning, measuring and improving on a schedule like any other.

— Xhurian

How we help you run scans, simulate agents and publish fixes

Everything above maps onto what we built. We simulate AI shopping agents against your store, diagnose in plain language why your products win or lose, and suggest publish-ready rewrites for titles and other listing fields. Here is how it fits your workflow:

• Scan first: run a free scan to see how agents currently treat your listings.
• Diagnose: see the title and data gaps that cost you selections against competitors.
• Rewrite: review suggested titles built on the hybrid pattern and your own product data.
• Publish: push approved changes to Shopify or WooCommerce in one click.
• Test: use our UCP playground to test agents.
• Track: rescan to confirm that your selection rate and attributed traffic are moving.

You can see how this works in our AI product listing optimization feature. Start with a free scan, or compare plans on our pricing page: Compete is $14.90 per month, Scale is $44.90 per month, and Dominate is $149 per month.

FAQ

How do I name an AI company?

Pick a name that explains what the product does rather than one that only sounds futuristic. Naming strategists recommend writing a short set of constraints first, such as length, pronounceability and a check that the name is available, as Fast Company’s naming analysis describes. Then test candidates against those rules.

What is the best structure for an AI-friendly product title?

Use a clear noun phrase followed by disambiguating specs such as model, size, color and material. This keeps the title readable for shoppers while giving agents the data they need to match the product to a request.

Where do AI-generated product titles go in Google Merchant Center?

AI-generated titles go in the structured_title attribute, with digital_source_type set to trained_algorithmic_media. Human-written titles use the standard title attribute, and both should stay factual and free of promotional text, per Google’s product data specification.

How long should a product title be?

Make the first 50 to 60 characters work on their own, because display surfaces often cut titles off around that point, according to Google’s title guidance. Put brand, model and decisive specs first and let secondary details follow.

Sources

• Product data specification - Google Merchant Center Help
• Ecommerce Search UX Best Practices 2026 – Baymard