Original insights by Dan Petrovic
30-second rundown
Key learnings:
- Google AI Mode chooses a container before serving the answer: Internal tags can turn an answer into a map, comparison, product list, calculator, file, chart or follow-up question.
- Useful facts need the right shape: Prices, locations, availability, specifications and other decision-making details must be clear enough for AI systems to place them into structured cards and tools.
- These tags are clues, not ranking buttons: Dan Petrovic’s findings reveal how answers may be assembled, but they do not prove that adding the tags to a webpage improves visibility.
This week: Choose one commercial page and the AI response format that best fits it, such as a map for a local shop or a comparison for software. Check that the page clearly states its price, availability, location, service area and essential specifications, then correct the most important missing or contradictory detail.
Coming 3 months: Build a workflow that checks important commercial questions in Google AI Mode daily and compares the answer’s format with the information on your pages. Flag pages where prices, locations, availability, specifications or other useful facts are missing or hard to extract for the maps, lists, comparisons and product cards AI Mode uses. Assign an owner to update each flagged page and check the answer again after the change.
Dan Petrovic has been crawling around the service corridors of Google’s AI Mode with a flashlight and little regard for locked doors. He found tags that tell the interface whether to ask a question, draw a map, build a calculator, compare products or hand the user a file.
That matters because the answer is no longer always a paragraph with a few blue citations nailed underneath. The interface can repackage information into an action, and the business whose facts fit that action may reach the customer while everyone else is still polishing prose in the parking lot.
1. The Answer Is Becoming a Room, Not a Paragraph
Petrovic documents a <FollowUp> tag that wraps targeted questions at the end of a response, asking for missing details such as budget, location or required amenities. It interrogates the customer like a motel clerk who wants cash before sunset.
The <Generate> tag goes further by instructing the system to build an interactive tool, such as a compound-interest calculator. Layout tags can produce maps, side-by-side comparisons or visual inspiration galleries, while separate tags can render charts, files, videos, events and quizzes.
AI Mode can choose a format that changes the customer’s next step. A hotel query may become a map or carousel. A software query may become a comparison table. A financial question may become a calculator that keeps the user inside the answer instead of sending them across ten websites.
2. Structured Cards Feed on Specific Facts
Petrovic’s <List> and <Entity> examples show text being organized into cards for products, hotels, local businesses, software and events. An entity is a recognizable thing, such as a company, store or product, with facts the system can attach to it.
The source lists 30 entity types and connects them to grounding details. A hotel needs prices and reviews. A local business needs an address and opening hours. Software needs licensing and compatibility details. Vague marketing copy is thin soup for a structured answer.
Consider a company selling commercial espresso machines. A page chanting “premium performance for modern hospitality” gives the system perfume and smoke. A page stating the model number, boiler capacity, dimensions, warranty, price and delivery time gives it the ingredients for a comparison card that can survive contact with a buyer.
The same rule runs through the product and shopping-offer tags. Identity and manufacturer details sit at one layer, while merchant price, stock, shipping and checkout links sit deeper. If those facts conflict across your site and public listings, the system inherits the mess.
3. The Renderer Has Rules, and None of Them Care About Your Feelings
These modules have guardrails. Petrovic says carousels need at least three similar child items. Lists and carousels require homogeneous entities, meaning the objects must belong to the same type. If the data fails validation, the interface can fall back to ordinary text rather than forcing a broken display.
A rendering rule is not a ranking factor. Petrovic’s article exposes clues about how the interface assembles answers, but it does not demonstrate that copying these internal tags into HTML will earn inclusion, improve rankings or summon a golden product carousel from the desert floor.
The useful move is less theatrical. Make the underlying facts complete, consistent and current, then monitor how important customer questions are actually rendered. When a comparison omits your strongest specification or a map forgets your location, fix the source data rather than painting occult brackets on the walls.
The old search result was a list of doors. This new contraption may build the room, arrange the furniture and produce a calculator before the customer sees daylight. You cannot command the floor plan, but you can stop arriving with missing measurements and a brochure full of fog.
