A phone, a television or a washing machine is shortlisted on what the engine could read, not on what the brand knows about it. By the time a shopper sees a price, an assistant has already resolved their question into six comparisons: the specification that matters for their use, the price today and where, the warranty and its exclusions, the after-sales service, what reviews say after months of use, and whether the seller is official. A product whose specifications sit in an image or a downloadable sheet, with a price only a marketplace states, gets described second-hand, or left out of the shortlist entirely.
Six questions, asked in one breath
A prompt like best budget smartphones in India right now does not stay one question. It fans out into six: which specification actually matters at this price, what the model costs today and from whom, what the warranty covers and for how long, who repairs it and how quickly, how it has held up for people who have owned it for a while, and whether the seller is official or a grey import. The assistant answers all six from whatever it can read, then writes one shortlist.
The specification sheet is where most brands think the comparison happens. It is where it starts. A shopper who has been told that two phones have the same camera and the same chip moves straight to the questions the sheet does not answer, and those are the questions that decide.
Almost none of the material the assistant reads has to come from you. Review sites, comparison videos, marketplace listings, support forums and technology publishers all describe your product whether or not your own page does, and they describe it in their words.
Why is a specification in an image not a specification?
Because an engine reads text, and a table rendered as a picture or packed into a PDF is not text. The product exists, the numbers are correct, and the answer cannot see them. So it takes the numbers from the nearest source that is readable, which is usually a retailer listing or a comparison site, and those are maintained by somebody else on their schedule.
The model number is the quiet half of the same problem. Reviews, forum threads and comparison videos refer to a product by its model code, and an engine matches a review to a product by that code. A product page that carries a marketing name and no model number cannot be matched, so the long-term review that would have earned the recommendation gets attached to nothing.
Variants make it worse. A television sold in three sizes and two panel types, with one page and one image for all six, gives the engine six products it cannot tell apart. The answer picks one set of numbers and states them for the whole line, and the shopper discovers at the store that the size they wanted is a different panel.
A warranty without its exclusions is a promise the shopper does not have
The warranty is the question a shopper asks last and the one that settles the decision. It is also the one this category publishes least carefully, because the warranty lives in the box, on a registration card and in a service policy, and rarely on the product page.
An engine that cannot read the exclusions does not skip the question. It infers an answer from whatever source mentions the warranty, and it will sometimes infer in your favour. That is the worse outcome: the shopper arrives at the service centre believing the screen is covered, and finds out that physical damage never was.
The service network has the same shape. A shopper in a smaller city asks who will fix it and how fast. If the service locator is a widget and the turnaround is unstated, the answer names the brand whose service page is a readable list of cities and a stated repair time. The rival was not better at service. The rival was legible about it.
What does an assistant do with a price it cannot see?
It quotes the price it can see, which is whichever marketplace listing it read most recently, including one from an unofficial seller with no warranty attached. The shopper then asks a direct question, where to buy your brand products with an official warranty in India, and the answer has nothing from you to work with.
Brands keep the price off the product page for reasons that made sense in a world where the shopper clicked through to a retailer. In an answer there is no click. The number on the page is the number in the comparison, and no number on the page means a number from somebody else.
The fix is not to publish every price. It is to publish the current price or the official retailers that carry the product, by region, as text. An engine can quote a stated price and point to an official seller. It cannot read a buy button that opens a drawer.
The six places the answer finds its evidence
An answer about your product is assembled from sources, and only one of them is yours. Product pages and specification sheets supply the specifications, the price and the warranty, when they are readable. Review sites and comparison videos supply the verdicts an answer cites. Retailer and marketplace listings supply the price, the seller and the warranty status, and restate everything else. Support and community forums are where faults, repairs and service experience become public evidence. Technology publishers define what a specification means. Service and warranty pages are where the engine checks what is covered and how to claim.
You control two of those six directly. The rest you influence by being accurate and easy to verify.
So a marketplace listing that misstates your warranty term is not a listing problem. It is a sentence that will be repeated in answers given to people who never visit the marketplace.
How do you know what the answer says about your product?
You read the answer. There is no proxy for it, and the traffic figures will not tell you, because the shopper who was shortlisted elsewhere never arrived.
So measure it as text. Run the prompts shoppers actually type, such as which television brand has the best after sales service in India or your brand warranty terms and how to register a product, across ChatGPT, Gemini, Perplexity and AI Overviews, and capture the answer, your position in it, and the domains cited behind it. A wrong warranty term or a stale price stops being a suspicion at that point and becomes a sentence you can act on.
The progression to aim for is a ladder. The answer compares brands and you are absent. You are named among options, with no model attached. Your product page is cited as the source for a specification. Your model is in the side-by-side the shopper is deciding from. Your model is recommended as the fit for this budget.
Each rung is earned by publishing what the rung below it was missing: the specifications as text with the model number, the price or the official sellers by region, the warranty term with its exclusions and how to register, the service network with how to book a repair and how long it takes, and a buy path that takes a city and returns an official seller with stock.