AI Discovery on Google

Shoppers increasingly ask an AI assistant what to buy instead of typing a search query. Google's AI shopping surfaces read product data from the same Merchant Center feed you already send — and the specification has fields built for exactly that kind of answer: short benefit lines, structured specifications, and questions with answers.

AI discovery is Renable's name for switching those fields on. It is an add-on use case on a Google channel, alongside "Main product feed", "Supplementary feed", "Video ads", and "Carousel ads". Turning it on adds four fields to what the channel's Fields tab asks for — nothing else about your feed changes.

What it can and can't promise#

This matters more than the setup, so it comes first.

Google documents what these fields are, and says they help customers discover information about your products across AI-driven surfaces. Google does not publish any figure for how filling them affects visibility, ranking, or traffic — and neither will Renable. These are the fields Google's AI shopping surfaces read. Google decides what it actually shows.

Two more things worth being blunt about:

• There is no deadline. All four fields are optional parts of the product data specification. Nothing is being switched off, and nothing expires if you skip this.
• No one can tell you your "AI share". Google has announced reporting for AI-powered shopping experiences but it is not generally available yet, and Renable does not have a number to show you. [VERIFY: whether Google's AI shopping insights reporting has launched, and in which countries.]

What you can say with confidence: these fields carry more of what a shopper would want to know, in a shape Google's specification defines, instead of leaving it locked inside a paragraph of description text.

The four fields#

Google's documented limits, which Renable enforces or respects:

• Highlights are 1–150 characters each, minimum 2 and maximum 100 per product, and Google recommends 4–6. (Google's product_highlight reference) Renable exports up to 10 highlight lines per product — the limit on any multi-value channel field — which sits comfortably inside Google's range and above its recommendation.
• Product details are groups of an optional section name plus a required attribute name and attribute value. Google documents no limit on how many a product may carry; Renable caps each field at 100 blocks per product by default, which you can change on the field. (Google's product_detail reference)
• Questions and answers allow up to 30 pairs per product, up to 1,000 characters per question and per answer, and 10,000 characters combined. (Google's question_and_answer reference)

Google also asks that you don't repeat the same information across these fields — an answer that just restates the description is wasted space.

Fields Renable deliberately leaves out#

Google's specification has other attributes associated with AI shopping surfaces — related products, popularity rank, document links, variant options. Renable doesn't offer them, because it has no honest source for them in your product data: a popularity rank would have to be invented, and a related-product link needs identifiers Renable can't reliably resolve. A field filled with a guess is worse than an absent field.

Turning it on#

There are three moving parts: the use case on the channel, the prompts that write the content, and the fields in the feed. The fastest route is to ask Aimée:

"I want my products to be found when people ask AI what to buy."

She'll turn on the use case, create the AI prompts, and add the feed fields — telling you the credit cost before she creates anything. You still activate and publish the prompts yourself, and you still write your own questions and answers.

To do it by hand:

1. Turn on the use case#

Open the channel → Settings tab → the Used for card. On the Google row, switch on the AI discovery chip. This applies immediately — there's no publish step for it.

Nothing is added to your feed by turning it on. What changes is that the Fields tab now counts the AI-discovery fields among its recommended ones, each with a note saying where its content comes from — description is normally already there, since a Google feed requires it regardless. Because these fields are recommended and never required, turning AI discovery on can't make your feed look broken or hold up a publish.

2. Create the content prompts#

Two of the four fields need content that doesn't exist yet. On the project's Creative tab, click Create prompt and pick:

• Write product highlights (AI discovery)
• Write enriched descriptions (AI discovery)

There's a third, Write use cases (AI discovery), which writes short "good for…" phrases onto your products. These flow into the same product_highlight field as the highlights — Google's guidance for that attribute is benefit language, including when-to-use phrasing.

These are ordinary text prompts: same editor, same fact-checking, same review flow as your existing ad copy. The one thing that's different is that their output is tagged — ai-highlights, ai-description, ai-use-cases — and the AI-discovery feed fields ask for exactly those tags. That's what keeps them separate: adding these prompts cannot change or replace the copy your existing prompts already produce, and the new content only shows up in the fields that asked for it.

Like any prompt, they don't run until you activate and publish them. See Working with prompts.

3. Write your questions and answers#

Questions and answers are not generated. They're yours to write — an invented answer about a product's warranty or fit is exactly the kind of thing you don't want an AI shopping surface repeating.

They live as project text content, so they're shared by every product in the project:

1. Open the project → Creative tab → the Project content section.
2. Click Add text content.
3. Add a headline — that's the question — and an excerpt — that's the answer. Add one pair per item.
4. In the same dialog, add the tag faq in the Tags field. Without that tag the item is ordinary fallback content and never reaches the question-and-answer field.

Start with the questions your support inbox actually gets: delivery times, sizing, materials, returns, care. Google asks you to keep out pricing and anything time-sensitive — those go stale in a feed.

Questions and answers are project-wide: every product in the project carries the same set. Keep them to questions that are true of your whole range — a question that only applies to one product would ship on all of them.

4. Add the fields to the feed#

Back on the channel's Fields tab, a notice at the top counts the recommended fields that aren't set up yet — click Show on it to reveal them. Each appears as a tinted suggestion row with an Add button, and adding one creates a working field: the binding, and for the two nested fields their block structure, are filled in for you. Then Publish the channel as usual.

product_detail and question_and_answer are nested fields: each writes a block per item rather than a single value, which is what Google's specification expects for them. See Channel fields and placeholders.

What this costs#

Two of the fields need AI-written content, so they cost credits — the same way every other AI prompt does, charged per product each time the prompt runs (see What consumes credits). product_detail costs nothing (it reads attributes your feed already carries) and questions and answers cost nothing (you write them).

If your catalog is large, activate one prompt first, look at the results on the Content page, and only then activate the second. Turning a prompt on and letting it run across a whole catalog is the expensive step — creating it is free.

Checking it worked#

• Fields tab — each AI-discovery field shows a fill rate from the last render. A field that exists but fills for a fraction of your products is the real gap; the usual cause is that the prompt hasn't been activated, published, or hasn't caught up with every product yet.
• Preview tab → Source view — spot-check one product's raw feed entry. You should see repeated highlight lines, and nested blocks for details and questions.
• Ask Aimée — "how AI-ready is my catalog?" on the channel gives you the per-field gaps, the fill rates from the last render, and a real product as evidence.

Unapproved AI-discovery content ships to your feed like all other AI content — approval is an editorial lock, not a gate. If you want to read it before it goes out, review it on the Content page filtered to unapproved content.

FAQ#

Does this work on Meta or TikTok?

No. AI discovery is Google-only — these are Google Merchant Center attributes. The chip doesn't appear on a Meta or TikTok row.

Will turning this on break my existing feed?

No. Every field AI discovery adds is recommended, never required, so it can't make a product look non-compliant or block a publish. And the AI content is tagged, so it can't displace your existing ad copy.

Do I have to publish the channel for AI discovery to take effect?

Turning the use case on applies immediately. The fields you add afterwards are ordinary channel edits, so those need a publish like any other field change.

My products have no attributes, so product_detail is empty. Is that a problem?

It's honest, not broken — a product whose feed carries no attributes simply doesn't get that field. If you want it filled, the fix is upstream: map more of your feed's columns as attributes on the catalog's Mapping tab.

Related articles#

• Working with channels
• Channel fields and placeholders
• Working with prompts
• Required vs. recommended feed fields
• Connecting a channel to Meta, Google, or TikTok