When the assistant has read the shopper's inbox, there is no single answer to rank for

For most of the history of search marketing, there was a thing called the results page. You could look at it, screenshot it and put it in a report. It varied a little by location and device, but two people typing the same words saw roughly the same list. A lot of early AEO work carried that habit over: ask ChatGPT a question, see whether you're named, call it your rank.

That habit is getting harder to defend. The assistants that matter most for shopping are being built, on purpose, to give different people different answers.

What changed this month

On October 27, Search Engine Journal reported that Google's privacy policy, in a version that took effect on October 16, now carves out an exception to its long-standing line about not using Gmail, Photos or Drive content for personalized ads. The exception applies when users choose to make that data available to other Google services that support personalized ads. In practice that means people who switch on Personal Intelligence in Gemini or AI Mode. Google has also started testing AI Mode ads for those users.

The ad part got the headlines. The bigger shift for brands happened in January, when Google brought Personal Intelligence to AI Mode. Robby Stein, Google's VP of Product for Search, described the shopping case plainly: AI Mode "considers the types of items you buy and where you shop." His example was a coat for a March trip to Chicago, found by reading the flight confirmation in Gmail and the brands the user prefers.

It's like a personal shopper who already knows your itinerary and the vibe you're going for.

Google isn't alone, and that's the point.

Everyone is building the personal version

Assistant	Personal context, in the company's words or as reported	Status
Google AI Mode with Personal Intelligence	Connected Gmail and Google Photos; "the types of items you buy and where you shop"	Opt-in Labs feature for eligible US AI Pro and Ultra subscribers since January 2026
Google AI Mode ads	Summaries, excerpts and inferences from connected apps may be used to show ads, per Google's help page as cited by SEJ	Testing for Personal Intelligence users; policy effective October 16, 2026
Amazon Alexa for Shopping	"Your personal preferences, shopping history, and conversations from across both Amazon.com and Alexa"	Launched May 2026 in the US
Meta Muse	Users connect it to email and calendars, per Mark Zuckerberg on the Q3 2026 call as reported by TechCrunch	More than 10 million users in the US and Canada

Amazon's launch post for Alexa for Shopping calls it the "most personalized AI assistant for shopping" and says it blends product knowledge with "your personal preferences, shopping history, and conversations." On Meta's Q3 call, Zuckerberg said other popular uses of Muse beyond shopping include connecting the agent to email or calendars, as TechCrunch reported.

Each of these is opt-in or limited in some way, and the reach of each is still small next to plain search. Personal Intelligence in AI Mode is a Labs feature for paying subscribers in one country. We're not claiming most shoppers see personalized AI answers today. We're saying the three companies with the most shopping data have all decided this is where the product goes.

Why "rank in ChatGPT" was already shaky

Even before personal context, AI answers weren't stable. The same prompt asked twice can name different brands, which is why we've argued that tracking a single AI rank misleads you. What personalization adds is a reason for the variation that isn't noise. A shopper whose inbox shows three orders from a clean-label brand, or whose photos are full of trail runs, is meant to get a different answer from someone else.

That breaks the mental model in a specific way. In ten blue links, if you were third, you were third for nearly everyone. In a personalized answer, you might be first for people whose history fits your product, absent for people whose history doesn't, and named with a caveat for people whose context raises a question your page doesn't answer ("is it safe with a nut allergy?").

So the thing to win is no longer a position. It's the share of contexts in which you're the reasonable pick.

What a brand still controls

That sounds like a loss of control. Mostly it's a change in which controls matter.

The facts every answer starts from

A personal shopper still needs to know what the products are. Whatever context Google, Amazon or Meta adds, it's matching that context against product facts: ingredients, sizes, certifications, use cases, price, stock. If those facts are missing, thin or inconsistent, no amount of personal context will put you in the answer, because the assistant can't tell you fit the person. The work on making product facts match everywhere becomes more important, not less.

Fit statements, not slogans

Personalization runs on fit. The useful sentences on a product page are the ones that say who a product is for and who it isn't for: "suitable for low-FODMAP diets", "not recommended for children under 4", "works in cold brew without clumping". These give an assistant something to match against a person's context. A brand line about "wellness for every lifestyle" gives it nothing.

History you earn, not buy

Two of the three personal signals in the table are about what someone has already bought or done. That's a quiet advantage for brands with repeat purchase, which is most of CPG. A brand already in a shopper's order history, inbox receipts or reorder list is the one personalization tends to bring back. It's also why getting the first purchase at a retailer that feeds its own assistant matters more than it used to.

Ads still run on your feed and pages

For the ad side, SEJ notes there's no new targeting toggle, no audience segment and no dedicated reporting for the Personal Intelligence test. Eligibility follows the existing AI Mode rules: Performance Max, Shopping, and broad match Search including AI Max. Advertisers never see the inbox. The inputs a brand controls are the same ones as before: the product feed and the landing pages. We cover that side in our AI Mode ads checklist.

How to measure something that's different for everyone

If there's no single answer, a single screenshot isn't evidence of anything. It's one sample.

The better approach is closer to market research than to rank tracking. Write a handful of realistic shopper profiles for your category: the parent buying snacks for a child with allergies, the runner looking for electrolytes without sugar, the person who always buys the store brand. Ask each assistant your category questions in a way that reflects each profile, many times, and track how often you're named and whether the facts are right. You can't log into someone else's Gmail, so you'll be approximating context with what the prompt says, which is an honest limitation. It still beats one logged-out screenshot.

There are tools for this kind of repeated, multi-assistant checking. XENA Intelligence asks assistants such as ChatGPT, Perplexity, Gemini and Amazon's assistant shopper-style questions and scores products on how they show up; SearchPanel, its product view, shows the actual answers. Whatever you use, report a rate across contexts, not a rank.

When an assistant gets a fact wrong for one of those profiles, that's the actionable finding. Our tip on what to do when an assistant gets your product wrong applies directly.

The uncomfortable part

There's a version of this future that's bad for challenger brands. If assistants lean heavily on purchase history, incumbents with years of orders in people's inboxes get a head start that's hard to buy your way past. We think that risk is real, and it's one reason the famous brand isn't always the recommended one argument matters: fit can still beat familiarity, but only if the assistant can see the fit.

There's also a privacy question that brands don't control and shouldn't pretend to. Google says the feature is opt-in, that it doesn't train directly on the inbox, and that sensitive categories are excluded from ad personalization. Shoppers will decide how comfortable they are with that, and some will switch it off. A brand strategy that only works for people who've connected their email is a fragile one.

What we'd do now

None of this calls for a new team or a new budget line. It calls for treating the product page as the source of truth that every assistant, personal or not, reads from. Make the facts complete. Write down who each product is for. Keep the feed and the page in agreement. Measure across many people, not one. The answer will keep changing from shopper to shopper. The facts it's built from shouldn't.