Checking Site Exposure in Personalized AI Answers

Google recently started using connected account data to personalize ads in AI Mode, not just the answers themselves. That raises the odds that AI answers and ads now look different from person to person. This guide covers what to check when personalization makes it harder to tell whether your site is actually showing up.

Why personalization complicates exposure checks

A query used to return roughly the same answer no matter who asked it. Once connected account data and personal interests started shaping both answers and ad selection, a single person's results stopped being enough to judge the full picture.

Login status, which apps are connected, and past search history are all variables now. The same person can see very different results just by changing their account state.

This shift goes beyond different wording in the answer. Which sources get mentioned and which ads appear alongside them can both vary by person. A single check can no longer tell you with confidence whether your site "shows up well" or "doesn't show up at all."

For instance, if a site that was consistently mentioned for a given query last week disappears when checked from a logged-in account this week, but still appears when logged out, the issue likely lies in that account's personalization settings rather than the content itself. Narrowing conditions one at a time like this helps pin down the actual cause.

Compare across different account states

The most basic method is running the same query once logged in and once logged out. If the two results differ significantly, that's a signal personalization is affecting exposure.

Where possible, running the same query across multiple accounts with different interests and search histories helps too. If results stay consistent across accounts, exposure is relatively stable; if they diverge sharply, it's worth narrowing down which conditions are driving the difference.

Ask questions from a real customer persona

Framing queries the way an actual prospective customer would ask them is another useful method. Results from an account that reflects your real customer base's interests tend to be more informative than results from an account used only for work logins.

Setting up a few personas in advance, each with its own list of queries, saves you from reinventing the questions every time. Logging results by persona also makes it much easier to compare later when something changes.

Separate ads from organic answers

Both Google and OpenAI have been expanding ad tests inside AI answers recently. A slot filled by an ad and a slot where the AI genuinely cited a source during its answer process are two very different signals, so it matters not to conflate them when judging exposure.

When capturing screenshots, note whether an ad label or sponsor tag is present alongside the result. That way, when you look back at exposure trends later, you won't lose track of which result was an ad and which was an organic citation.

Control timing and variables

As personalization grows, a single check matters less than repeating the same check under the same conditions on a regular cadence. Fixing variables like day of week, time of day, and device type where possible makes comparisons more reliable.

Logging login status, device, and exact query wording in the same format every time makes it much easier, weeks or months later, to figure out which condition changed when exposure shifted.

Set up a logging template in advance

Deciding in advance what to record at each check makes later comparison far easier. A reasonable baseline template includes the check date, login status, persona used, exact query wording, sources the AI mentioned, and whether an ad appeared.

Turning this into a table you fill in the same way each time lets patterns become visible at a glance once you've gathered a few weeks of data. Too many fields makes consistent logging hard to sustain, so start with only the essentials and add more later.

What to do when you spot a change

When an exposure pattern looks different from usual, it's safer to recheck under the same conditions before drawing conclusions. That helps distinguish a temporary fluctuation from an actual algorithm or policy change.

Rerunning the same query at a different time, or checking similar related queries, helps gauge how widespread the change is. It's worth noting whether the shift is limited to one query or shows up across several related ones.

Running this as a team

If multiple people share responsibility for exposure checks, keeping the persona list and logging template in a shared document helps. Results collected from different accounts need to land in one place, or the full picture gets lost.

Documenting clearly which accounts and personas are in use lets the process continue consistently even when the person doing the checking changes. Over time, this accumulated record becomes a useful asset in its own right.