Google Shopping Titles: Rewrite Product Titles With AI Safely

To rewrite Google Shopping titles with AI safely, give the model each product's real attributes, ask for a front-loaded title under 150 characters with no promotional text, and send it in Merchant Center's structured title attribute marked trained_algorithmic_media. Then test the new titles on a matched group of products before changing the rest.

Titles are the most useful feed field to rewrite and the easiest to get wrong. They decide which searches your product matches, and Google says a specific, accurate title helps show the product to the right customers. A model can tidy 500 titles in an afternoon. It can also invent a compatibility claim in all 500, so the safeguards in this tutorial are the point, not an afterthought.

Follow me on Instagram@sagnikteaches

The rules Google sets for the title field

Everything below comes from Google's Merchant Center help page for the title and structured title attributes. Rules change, so read the page again before a big upload.

Connect on LinkedInSagnik Bhattacharya

• Length. 1 to 150 characters. Longer titles are truncated and flagged with a warning. Google recommends using all 150 where useful.
• Front-load. Put the most important details first, since most shoppers see only the first 70 or fewer characters, depending on the screen.
• Describe the product on the landing page. The title does not have to match the page word for word, but it must refer to the same product.
• Distinguish variants. Put the distinguishing details in the title (colour, size) and also in the matching attributes. For features with no attribute, such as flavour, the title is the place.
• No capitals for emphasis. All-caps shouting is called common in spam. Capitals for real abbreviations are fine.
• No promotional text. No price, sale price, sale dates, shipping, delivery dates, other time-related information or company name.
• No gimmicks. No symbols, HTML tags, extra white space or decorative foreign characters.
• Language. Most product data must match the data source's language; foreign words only where well understood.

The label that AI titles must carry

This is the part most tutorials skip. Google's generated-content policy says titles created with generative AI must go in the structured title attribute, not the title attribute. That attribute has two parts: a digital source type, set to trained_algorithmic_media for AI-written text, and the content, which is the title itself. Titles that were not AI-generated can use either attribute, with the digital source type left as default.

Subscribe on YouTube@sagnikteaches

In a text feed, Google's example for an AI title looks like this:

trained_algorithmic_media:"Stride & Conquer: Original Google Men's Blue & Orange Power Shoes (Size 8)"

A title that was not written by AI is just the content after an empty first value: a colon and the text. Note two details from Google's page. If the content contains a colon, wrap it in double quotes; if it contains a double quote, double it. And if you send both a structured title and a plain title for the same product, Google says it uses only the plain title. That means a lazy fix, leaving the old supplier title in place while adding the new one, quietly throws your rewrite away.

What counts as written by AI

Google does not draw a bright line between "AI wrote it" and "AI tidied it", and I would not guess where it falls. A safe rule: if a model produced the words that are in the final title, label it. If a person typed the title and a spell-checker fixed a typo, it is a human title. When in doubt, label it; the cost is a field value, not a violation.

Where to put it on Shopify

A store on Shopify usually sends products to Merchant Center through the Google and YouTube app or a feed tool, and the product title in Shopify becomes the feed title. Changing the Shopify title also changes the page heading and search listing, which you may not want. Many stores use a supplemental data source in Merchant Center, keyed on the product ID, to override feed fields without touching the shop. Check in your own account whether the source you use accepts the structured title attribute and how it handles the colon-separated format, and try it on three products first. Google's page shows the format for text sources, XML and the Merchant API.

Title structure by category

Google gives the rules; it does not give one formula for every category. The patterns below are my suggestions for the order of details, built on its examples (such as "Sleeveless Polka Dot Skater Dress, Small, Black", and a 15-inch Android Squishable in organic cotton for a product with no variants). Test them. A sensible order is the way a shopper would type the query.

For parts and accessories, the "what it fits" detail is the one most worth checking by hand. An AI model will happily write "fits all models" because it sounds helpful. Google's page has dedicated requirements for compatible and refurbished third-party products and printer cartridges, so look up the rule for your category in the live page before you apply a pattern.

Special cases Google calls out

• Software subscriptions. Include the word "subscription" and the duration, for example "(1-year subscription)".
• Phones or tablets with a contract. Say in the title that a contract of a certain duration is needed, such as "with contract, 12 monthly payments".
• Bundles and multipacks. Describe the whole bundle or the pack count in the title.
• Samples. If you sell a sample of a bulk-only product, begin the title with "sample" and submit the sample's own price in the data source.

A main worked example: 300 kitchenware titles

Take an illustrative wholesale distributor of catering and kitchen supplies with a Shopify store of 300 products, mostly imported lines with supplier titles. A typical raw title reads "NEW 2026 HOT Stainless Steel Cooking Pot Soup Stock Pan Cookware with Lid Free Shipping 24cm". It breaks four rules: capitals for emphasis, promotional text ("Free Shipping"), a year used as a hook, and a feature stack with no brand or order.

The team's attribute sheet has columns for brand, product type, size, material, colour, pack size and compatibility (for example "induction compatible: yes"). They feed the sheet, not the old title, to the model.

The prompt

You write Google Shopping product titles from a data row.
Rules:
- Maximum 150 characters. Put the most important words in the first 70.
- Order: brand, product type, key attribute (size or capacity),
  material, colour, pack size.
- Use ONLY the values in the row. If a value is blank, leave it out.
  Never guess a compatibility, material or capacity.
- No promotional words: free, sale, best, new, hot, cheap, top quality.
- No price, shipping, dates or store name.
- Normal capitalisation, no symbols.
Return a table: product ID, new title, character count.

Row: ID 1042 | brand: Acme | type: stockpot | size: 24 cm / 8.5 litres
| material: stainless steel | colour: silver | lid: yes | induction: yes
| pack: 1

Illustrative output: "Acme Stainless Steel Stockpot with Lid, 24 cm, 8.5 Litres, Induction Compatible, Silver" (87 characters). Everything in it comes from the row. The model did not add "dishwasher safe", because the row did not say so.

What you check before upload

1. Character count. Ask for the count and recompute it yourself in a spreadsheet with LEN. Models miscount.
2. Invented facts. Filter for words that are not in the source row: "premium", "heavy-duty", "commercial". A short formula or a second prompt can list them.
3. Banned words. Search the whole column for free, sale, new, best, shipping, discount, and any currency symbol.
4. Variants. Make sure two variants do not end up with identical titles. A duplicate title for different colours breaks Google's rule that variants must be distinguishable.
5. Landing page. Open five at random and check that the title describes the product that the page shows.

Illustrative result of the review: of 300 titles, 17 had a banned word the model copied from somewhere, 9 exceeded 150 characters after the pack size was added, and 6 duplicated another variant. Two checks and a few edits fixed all 32. Without the checks, that is 32 products with a warning or a disapproval.

A before and after you can reuse

The second row is the instructive one. "Water-Resistant" was added only because the attribute sheet said so. If the sheet had been silent, the correct output would have omitted it.

Building titles from a sheet, then letting AI handle the exceptions

A model is not always the right tool for every title. For a catalogue whose attributes are clean, a spreadsheet formula builds the title more reliably than a language model: it cannot invent a word. A formula such as brand, then type, then size, then colour joined with commas gives a consistent title in seconds, and a title that no model wrote does not need the generative AI label.

Use AI for the rows a formula handles badly: products where the supplier type is vague, where two attributes conflict, or where a feature belongs in the title but has no column. A practical split for 300 products is 220 built by formula and 80 handled by the model. That halves the number of titles needing the generative label and cuts the risk of invented detail in the bulk of the feed. It also makes the review lighter, because you check the 80 carefully rather than all 300 loosely.

A prompt for the awkward rows only

These product rows have a blank or conflicting field. For each,
write a title using the pattern: brand, type, key attribute, colour.
If a field is blank, do not guess; put the field name in a column
called "needs input" and leave the title blank. If two fields
conflict (for example size 24 cm and size 28 cm), flag it and
leave the title blank.

Illustrative output: 80 rows in, 61 titles written, 19 returned with a "needs input" note such as "colour blank" or "size conflict". Those 19 go back to whoever owns the data. The model has done the sorting; a person fills the gaps.

A second example: a software reseller's subscription titles

Take an illustrative B2B software reseller listing licences in Shopify. The supplier title is "Pro Suite 2026 Business License Key Instant Delivery". Instant delivery is shipping information and a year in the title is a hook, so both go. Google asks that a software subscription title includes the word "subscription" and its length. With the attribute row (brand Acme, product Pro Suite, seats 5, term 12 months), the title becomes "Acme Pro Suite Business, 5 Users (1-year subscription)". The model could write it; so could a formula. The part to check is the term, because a licence that is a perpetual key and a licence that is a 12-month subscription need different words, and a model told to "make it consistent" will blur them.

When a title is right but the product still does not show

Titles are not the only reason products miss out. Check three things before blaming wording.

1. Status. Is the product approved, or sitting in the Needs attention tab? A perfect title on a disapproved item does nothing.
2. Price and availability. A mismatch between feed and landing page causes disapprovals unrelated to titles.
3. Identifiers and images. Missing GTINs and non-compliant images limit reach; for images, see the image requirements and AI metadata rule.

Fix those first, then judge titles. Otherwise a title test measures the noise of an unrelated fault, and you may drop a good title for the wrong reason.

Handling long-tail variants without drowning in titles

A shirt in 6 sizes and 5 colours is 30 variants. Each one needs its own title with the distinguishing details, as Google's skater dress example shows with its six titles for two colours and three sizes. Do not write 30 by hand and do not ask the model for 30 free-form titles. Give it the base title and the variant table and ask for output in a fixed pattern. Then check that the list has 30 unique titles by sorting and looking for duplicates, which takes a minute. A single duplicate in a 30-row block can be the cause of confusing performance data, because two variants will compete for the same query.

Mistakes that look like improvements

Three failures show up again and again.

• Keyword stuffing. A model asked to "maximise keywords" writes "Shoes Sneakers Trainers Running Jogging Gym Sport". Google asks for professional, grammatically correct language and a clear description of the product, not a word list. One product type, named once.
• Copying a competitor's pattern. It may include brand names you do not sell. Brand in the title must be the brand of the product, and the brand attribute must agree.
• Losing the variant. The AI shortens "Blue, Large" to fit and the colour disappears. Check that every variant keeps its colour and size in the title or the matching attributes.

A test group that gives a fair answer

New titles should earn their place. Guessing from the first week's numbers is a common trap, because seasons, promotions and bids all move. A fair test compares like with like in the same weeks.

1. Pick a pool. Choose 60 products of similar type and price with enough traffic, say each with at least 200 impressions in the last 28 days.
2. Split by pairs. Sort the pool by last month's impressions and alternate: product 1 to the test group, product 2 to the control, product 3 to the test, and so on. That keeps both groups the same size and the same strength.
3. Change only the title. Do not change prices, images or bids for either group during the test.
4. Run for four weeks, no fewer. That gives enough clicks to compare, and covers a few weekly cycles.
5. Compare. Impressions, clicks, click-through rate, conversions and revenue per product, test against control.

Illustrative result: the test group gets 18% more impressions and a click-through rate of 1.9% against 1.7%, with conversions roughly level. That suggests the new titles match more searches, though the conversion figure is too small to call. With few sales, accept the impression and click signal and roll the pattern out in stages. If the test group shows a drop, look at the lost terms first: the old title may have contained a search word the new one dropped.

Because Google may customise titles with details from your landing page, what shoppers see can differ from what you send. Look at your listings in a real search as well as in the reports. If you would rather Google did not rewrite your Shopping ad titles, the page says to ask your account manager or support to opt out of Product Data Customization, with requests typically processed in 3 to 5 business days. Think about it before opting out: dynamic titles may match some searches better than yours.

A weekly routine for new products

Once the catalogue is done, new products are the maintenance load. For each batch of new items, complete the attribute row before writing anything, run the prompt, check the five points, label the AI titles and upload. Fifteen minutes for ten products is realistic once the sheet exists.

Titles interact with other fields. A missing GTIN can limit where a product appears whatever the title says, which the tutorial on missing GTINs in Merchant Center covers. Bulk page titles on the shop itself are a separate job, handled in fixing Shopify titles and meta descriptions. If the same text will also appear on product pages, see which AI store pages are at risk under Google's spam policy, and if an account problem has stopped products showing at all, start with the suspension audit.

Title questions that trip up feeds

What is the character limit for a Google Shopping title?

Google's title attribute accepts 1 to 150 characters, and longer titles are truncated with a warning. Google advises putting the most important details first because shoppers usually notice only the first 70 or so characters, depending on screen size. Using close to all 150 characters is recommended.

Do I have to tell Google a title was written by AI?

Yes. Titles created with generative AI must be sent in the structured title attribute, with its digital source type set to trained_algorithmic_media, instead of the plain title attribute. If you send both attributes, Google says it uses only the plain title, so leave the old one empty for those products.

Can I put free shipping or a sale price in the title?

No. Google's requirements say not to include promotional text such as price, sale price, sale dates, shipping, delivery dates or your company name in the title. Send those through their own attributes, such as sale price and shipping, where Google displays them properly.

Will Google change my titles anyway?

Possibly. Google may promote key attributes using content from your landing page and other signals, and shows a dynamic title where it fits a search better. You can ask your account manager or support to opt out of Product Data Customization for Shopping ad titles; requests are typically processed in 3 to 5 business days.

Further reads

• How to Write Shopify Product Descriptions With AI That Convert — Write the descriptions that sit behind these titles.
• Google AI Mode Shopping: Merchant Center Fixes for a Small Store — Feed fixes for shopping results led by AI.
• How to Create Product Photos With AI on a Small Budget — Product photos with AI on a small budget.
• Does Writing Etsy Listings With AI Hurt Your Search Ranking? — How AI-written listings fare in marketplace search.
• Google Ads for Dropshipping With AI: What to Let It Write — Check ad claims after titles are fixed.
• Performance Max vs Standard Shopping for a Small Online Store — A small-store way to choose between Performance Max and standard Shopping: budget floor, six-week test, fair comparison and a decision rule.
• AI Product Title Generator: How to Test One Before You Trust It — A side-by-side test of an AI product title generator on 20 of your own listings, with accuracy and keyword checks before any go live.
• WooCommerce to Shopify Product Import: Variants and Redirects With AI — Map WooCommerce export columns to Shopify's import file, fix variants and images, build a redirect list and audit a sample before go-live.
• Perplexity Merchant Program: Should a Small Online Store Join? — What Perplexity and feed-tool guides say about the free programme, who can apply, what a product feed needs, and a sum to decide if the prep time pays.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Merchant Center Help (title and structured title attributes, generated content policy, product data customisation), Google Ads Help.