AI Overview Optimization One answer box. Eight ways into it.
Google assembles the answer above the results from whatever it trusts on the question. Sometimes that is a guide. Just as often it is a product page, a video, a pin or somebody else mentioning your brand. Most stores own one of those, so they qualify for one kind of question and are invisible for the rest.
Yes. Ceramic window film rejects infrared heat without darkening the glass, so a lighter 70% film can block more heat than a darker dyed one. Tint percentage describes visible light, not heat rejection.
Five sources, five different kinds of thing. One of them is an article. The rest are a video, a product page, a forum reply and a pin. A store that only publishes articles was eligible for one of those five slots.
The question decides the format.
Not you.
Nothing gets made until we know what Google currently puts in the box for each question, and which kind of source it pulled that answer from.
AI Overview
source citedCeramic film rejects infrared without darkening the glass, so a 70% film can block more heat than a darker dyed one.
The slot goes to whichever source answers the question soonest and cleanest. Owning more than one kind of asset is what puts you in the running more often.
AI Mode
kept on follow-upA conversation, not a query. The source that answered the first question well is the one still there three questions later, which is why depth beats breadth here.
AI search filters
attribute presentGoogle cuts the category down using chips built from product attributes. A product missing the attribute is removed from the filtered set even though it matches. The shopper never sees it to rule it out.
The answer above the results
is now the result
Both figures below are Google's own, said on stage at its developer conference, and both are linked.
monthly active users of AI Overviews. The box above the results is not a pilot any more, and for most queries it is the first thing a buyer reads.
Sundar Pichai, Google I/O 2026 keynotemonthly active users of AI Mode, within a year of launch. A second surface, assembled the same way, from the same sources.
Sundar Pichai, Google I/O 2026 keynoteBeing in the box is not a bonus on top of ranking. For a question that triggers one, it is the result. Google has said it keeps adding inline links and site previews inside these answers, so the slot is worth holding. The only question is which of your assets is eligible to fill it.
You wrote the guide.
The box wanted a video.
A store with three hundred articles and no video answers every question that wants prose and loses every question that wants a demonstration. The content was never the problem. The shape of it was.
Years of guides, comparisons and how-tos. Real work, and genuinely useful.
A how-to question. The kind your articles were written for.
For a how-to, Google would rather show somebody doing it than describe it. The format is chosen by the question, not by what you happen to own.
Your article was never in the running. Not because it was weak, but because it was the wrong kind of thing.
Writing a three hundredth article does not help here. Opening a door you do not have does.
A brand mention
The cheapest slot in the box is somebody else saying your name. Google quotes forums, roundups and review sites constantly, and none of that content is yours to write.
- Consistent name, price and details everywhereA store listed three different ways reads as three weak entities instead of one strong one.
- Be where the category is discussedPrivate reviews on your own site are invisible to the box. A public answer in the place buyers ask is not.
- Never buy or fake a mentionIt breaks platform rules and, when it is spotted, it takes the trust you did earn with it.
- Mine what people already sayThe doubts repeated in your reviews are the questions the box is trying to answer.
Running 70% ceramic on the windscreen and it made a real difference in summer. Got it from VinylFrog.
This store already has something that can fill this slot. The work is making it the version an engine can lift cleanly.
A product page
For a specification question the box quotes a product page, because that is where the number lives. It also builds its filter chips from product attributes, and a page missing the attribute is dropped from the filtered set even when it matches.
- An answer in the opening two sentencesWhat it is, who it fits, and the one number that settles the question. That is the line an engine can lift.
- Thirty to sixty real attributesMaterial, fit, terrain, tint percentage, warranty. Google builds the filters from these, so a missing attribute is a silent exclusion.
- A title that survives truncationBrand, product, key attribute, fitment, front loaded and under about sixty characters.
- Complete and honest schemaProduct, Offer, Variant and Review, matching what the page actually says.
Blocks up to 98% of infrared heat. 70% visible light transmission, legal on windscreens in most states.
This store already has something that can fill this slot. The work is making it the version an engine can lift cleanly.
A collection page
For a which-one question the box quotes the page that helps somebody choose. A grid of products is not that page. A grid with a real buying guide under it is.
- Headings in buyer languageThe words somebody types, not the words your catalogue uses internally.
- A guide below the gridTwo sentences above it, the real explanation underneath. That prose is what gets quoted.
- Comparison tables, marked upOrganized by the situation a buyer is in, not by SKU order.
- Filters with counts and real valuesA filter showing how many products sit behind it signals a curated category rather than a dumped feed.
Five VLT levels from 5% to 70%. Which percentage is legal where, and which blocks the most heat.
This store already has something that can fill this slot. The work is making it the version an engine can lift cleanly.
A guide or comparison
For an explanation question the box quotes prose, and this is the door most stores already have open. It is also the one where original evidence decides everything, because every competitor has the generic version.
- The answer first, not the payoffA line an engine can lift, in the opening, before the history of the product category.
- Original evidence only you haveWhat you see when a product fails, the test you actually ran, the mistake customers repeat. This is the part a content mill cannot copy.
- The vocabulary the topic requiresThe specialist terms a real answer contains. Missing them reads as a summary of other summaries.
- Upgrade before you addMost stores gain more from fixing what is published than from publishing more.
Ceramic film rejects infrared without darkening the glass, which is why a 70% film can outperform a darker dyed one.
This store already has something that can fill this slot. The work is making it the version an engine can lift cleanly.
A YouTube video
For a how-to question the box wants to show somebody doing it. This is the door almost no ecommerce store has open, and it is the one with the largest demand behind it.
- The term in the title and the transcriptYouTube and Google both read what is said. A video that never names the thing cannot be matched to it.
- Chapters that map to sub-questionsEach chapter is separately quotable, which is how a single video answers five questions.
- Demonstration, not presentationThe question asked how. Show the hands, the tool and the failure mode.
- Mapped to the same catalogueOne keyword bank feeds the video, the article and the product page, so they stop competing with each other.
window tint · 50,000 est. monthly searches on YouTube alone
This store has nothing eligible for this slot today. Every question that wants this kind of answer is currently going to somebody else.
A TikTok clip
For a comparison question, short video is where the argument actually happens, and TikTok is itself a search engine that Google indexes.
- The term spoken in the first secondsTikTok transcribes the audio and matches searches against it.
- The term set on the frameRead by OCR, and the only layer a viewer gets with the sound off.
- One clip per real searchRepeats of the same idea compete with each other rather than with anyone else.
- Indexed twiceGoogle puts short videos on its own results page, so one clip is eligible on two engines.
window tint vs vinyl wrap · 7,200 est. monthly searches
This store has nothing eligible for this slot today. Every question that wants this kind of answer is currently going to somebody else.
A Pinterest board or pin
For a planning question the answer is an image, and Pinterest pages are public, so what you build there is eligible for Google image results as well as for Pinterest search.
- The board is the unit that ranksA pin is one impression. A board about one clear thing keeps collecting saves.
- Titles in the words buyers typeAlmost no Pinterest search names a brand, so the words decide whether you appear.
- Built from the same keyword bankThe board, the article and the product page are answering one question, not three.
- Public pages, so Google reads them tooA board is a page with a title, a description and captioned images.
window tint · 14,800 est. monthly searches on Pinterest
This store has nothing eligible for this slot today. Every question that wants this kind of answer is currently going to somebody else.
A Meta video or page
Meta AI answers inside Instagram, Facebook and WhatsApp, and the same catalogue work feeds it. For most stores this door has never been opened at all.
- The same map, a different surfaceThe catalogue and the questions are already mapped. Meta is another place to answer them.
- Answer in the caption, not just the clipMeta AI reads text. A caption that names the thing is what makes the post retrievable.
- Consistency with everything elseThe same category vocabulary across every surface is what makes the brand one entity rather than five.
- Counted in coverageA shut door is tracked as shut. That is the number this page reports.
This is a real gap on this store. Three surfaces mapped, three still shut.
This store has nothing eligible for this slot today. Every question that wants this kind of answer is currently going to somebody else.
Three readers, one box
Every source in an AI Overview is read three times, and content made only for the third of these is the reason most stores never appear in the first.
Assembles the box from what it can extract cleanly. It has no loyalty to a format: it takes the video, the product page or the forum reply, whichever answers soonest.
Reads three or four lines and a row of sources. If your name is in that row you are in the consideration set before anybody has visited your site.
Still there, still sending traffic, and still worth winning. The box sits above it rather than replacing it.
Same catalogue.
Eight doors instead of one.
No new products. What changes is how many shapes of question your store is eligible to answer, and that is a number you can count.
The state column is the real position of the store used throughout this page: 4 doors open, 4 still shut. It has the catalogue, the collections and nearly three hundred articles, and nothing published on video or visual search. That count is the thing this service reports, because it is the thing that can be counted. Rankings and citations are not ours to promise.
What people ask before starting
Is this the same as normal SEO?
So this is a content service?
Do we have to be on every platform?
How is this different from LLM Optimization?
How long before an engine cites us?
Will you invent statistics to make the content look authoritative?
What are AI search filters?
Find where AI can cite your store
Six engines, your gaps, and what to do next. Or talk to a human.
- Google AI
- ChatGPT
- YouTube
- TikTok
- Meta AI