Product data guidelines of Google, Meta, TikTok and Pinterest: rules with sources and AI titles

Product data guidelines are the rules an ad platform sets for every product in a catalogue: how long the title may be, what it must not say, which image and which price are required. A product that breaks a required rule is rejected. A product that breaks a recommendation may be shown less often or less well. This article shows where the rules come from, how to see which ones your products break and how the AI title suggestion works.

Why do product data rules matter for sales?

Shopping ads on Google and catalogue ads on Meta, TikTok and Pinterest are built from your feed. If a product has no image, price or link, the platform rejects it and shows it in no ad. If its title is full of capitals and “SALE -30%”, it may be shown less often. Product data is the precondition for an ad to have anything to show at all.

The catch is that each platform has its own rules, spread across many help pages, and they sometimes differ. A few examples from the vendors' documentation:

Google has the strictest length limit here (150 characters), Meta the shortest recommendation (65). One title for every platform works best with the most important information first.

What is on the Guidelines page?

The Guidelines page is one base of 50 rules from four platforms, grouped by topic: required fields, identifier, title, description, link, images, availability, price, brand and identifiers, GTIN, condition, category, apparel, labels. Each rule has a platform, a weight, its text, a link to the vendor page and the check date. Each shows how many of your products it affects today.

There are three weights:

• Required — the platform rejects a product that breaks the rule.
• Important — the product may be shown less often or risks rejection.
• Tip — an opportunity to improve.

At the top are three tiles: rules in the base (from how many vendor pages), rules your products break and active products. Filters let you pick a platform and a topic and turn on “Only rules my products break”. That last switch is the fastest way to a to-do list.

Only rules a vendor page states outright made it into the base. Basketly does not add numbers of its own as “Google requirements”. A few product checks are Basketly's own tips (for example a title that is too short or a description identical to the title); they are labelled as tips with no vendor rule and claim no source.

How do the guidelines connect to the product diagnosis?

The product diagnosis is 44 checks run with every read of the shop or feed, with a score from 0 to 100. The guidelines say which vendor rule each check comes from. The number next to a rule is the products that break it according to the latest diagnosis, as the shop gives them.

An example: the check “promotional phrases in the title” enforces both Google's rule “no promotional text” and TikTok's rule “no promotional text or emoji”. A product counts once per rule, even if it breaks it for several reasons.

Next to each check of a rule is a button that opens the Products page filtered to products with that problem. On the product record every problem with a vendor rule shows its sources (links to the platform pages) and the check date.

Some rules have no automatic check because feed data cannot judge them, a clean image for example. The page then says that Basketly does not check it automatically and you should check it yourself. That beats a false “zero problems”.

How many problems fix themselves? The automatic feed fixes remove HTML from descriptions, capitals and sales phrases from titles, shorten long titles on a word boundary and drop an invalid GTIN, among others. The numbers on the Guidelines page show the state “as the shop gives it”, so some of those products are already fixed in the served feed; the result after fixes is on the product record.

What happens when a vendor changes its page?

Every Monday Basketly downloads all 20 vendor pages, keeps only the visible text and compares it with the previous read. When the text changes, the page is marked “changed — to review”, and its rules keep their current wording until a person rereads the page.

That is deliberate. A changed page may mean a new rule, but just as well a typo fix or a new example. “Rewriting” rules automatically from the page text would be guesswork. On the Guidelines page a rule whose page changed says so: it is being reread, and until then it stands as written.

A failed download changes nothing and shows the error. The whole base was last compared with the vendor pages on 25 October 2026; that date shows next to every rule.

How does “Suggest a title” work?

On a product record, someone allowed to edit connections can ask the AI for 1 to 3 better titles. The model gets the title rules of all four platforms with their sources and only the product's own data. Every proposal goes through Basketly's title checks again, and those that fail are dropped.

What exactly the model gets: title, brand, product type, Google category, colour, size, material, gender, age group and the first 600 characters of the description. This data is marked as untrusted shop data the model must not take instructions from. No customer or order data.

A proposal is dropped when it is over 150 characters, written in capitals, carries sales phrases or symbols, or equals the current title. Each proposal shows which rules it follows; only rules that are really about the title and were sent to the model are kept.

Nothing is saved by itself. Only “Apply” stores the chosen title as the product's manual fix, and from then on every feed with that product serves it. The next read of the shop does not erase that fix.

Limits: requests count against the organisation's daily AI budget (refused once it is spent) and at most 30 per organisation per hour. Without a MiniMax key on the server the panel says suggestions are off.

How it looks in Basketly

The page is in Work → Guidelines. Turn on “Only rules my products break” first and pick the platform you advertise on. Next to each rule click the button with the product count to jump to the list in Products. Fix titles on the product record with “Suggest a title”.

Where to fix the data depends on the cause:

• A problem across many products (e.g. no brand in the title) — a feed rule in Feeds.
• A single product — a manual fix on its record.
• A mistake at the source (wrong price, no image) — in the shop, because that is where the true data belongs.

More on rules, fixes and feed formats: product feeds. Once the data is in order, check whether products are actually shown in the article on products not shown in catalogue ads.

Limits

• The base is 50 rules, not the full documentation. It covers the main fields, not every special case or advertising policies.
• No rejections straight from the platforms. Basketly checks data against the rules but does not yet read the diagnostics of Merchant Center or Meta's Commerce Manager. A product without errors in Basketly can still be rejected for another reason.
• A page change is not a rule change. “To review” only means someone has to reread the page.
• An AI title is a proposal. Read it before applying it: the model does not know your range better than you do, and Basketly's checks catch rule breaks, not factual mistakes.
• Counts follow the shop. The number of products breaking a rule comes from the diagnosis of the shop's data, before feed fixes.

Next step

Open Guidelines, turn on “Only rules my products break” and start with the rules weighted “Required”: those are products the platforms will not show at all. Then check in the article on ROAS and MER whether a cleaner catalogue turns into sales, and see the shop's connections on the connection map.