Google's Gmail Ad Exception Gives Advertisers Nothing but a Messier Baseline

Google's privacy policy now lets Search personalize ads using summaries, excerpts and inferences from Gmail, Photos and Drive when a user connects those apps through Personal Intelligence, an exception that took effect October 16, 2026 to a no-Gmail-ads pledge Google made in July 2021. Advertisers get no new segment, report or exclusion for it, so any shift in AI Mode performance from mid-October needs its own baseline before anyone credits bidding changes.

The change was spotted by PPC Land, which compared the new text against Google's privacy policy archive and found the original sentence ("We don't show you personalized ads based on your content from Drive, Gmail, or Photos") had survived 17 versions untouched. It now carries a carve-out: unless you choose to make that data available to other Google services that support personalized ads. The Verge reported the update the same day PPC Land did.

Most of the coverage so far is framed as a privacy story, which it is. But I think the more useful read for anyone running Google Ads is about measurement, and about who actually benefits from a signal nobody on the buy side can see.

Google said "no ads" in March, then started testing in October

The timeline matters here, because it shows how fast the position moved. In March, Google told Search Engine Land there were "currently no ads for people who choose to connect their apps with AI Mode," while adding that it expected ads to work similarly for those users in the future. The future turned out to be about seven months away. Google then confirmed to Search Engine Land that it was starting to test ads for users with Personal Intelligence enabled, and Search Engine Roundtable caught the matching help-document edit on October 19.

That help page is where the careful wording lives. "Ads will not be based directly on your Gmail inbox or your Google Photos library," it says. Then, a few lines later: summaries, excerpts, generated media and inferences from connected emails and files may be used.

The word "directly" is doing a lot of work in that sentence.

Think of it like a hotel concierge who promises never to read your mail, then reads the note your assistant wrote summarizing it. Technically true. Practically, the concierge still knows you have a flight to Denver on Thursday. And as PPC Land points out, a "summary" of a short email can easily be the whole email.

What the advertiser side actually gets (spoiler: nothing new)

Here is the part that surprised me a little. For a change this big on the user side, the advertiser side is basically blank. Per the help page and Search Engine Roundtable's reporting, advertisers will not have access to Gmail or Photos data. There is no "connected apps" audience, no toggle to opt out of that traffic, and no reporting dimension that tells you an impression was shaped by an inbox inference. Google also hasn't said which markets the test runs in or which ad formats are involved.

So the signal flows into Google's ad selection, and the only place you would ever notice it is in aggregate results. On paper, that sounds like free performance. And sometimes it probably will be. A user whose inbox holds a hotel confirmation for next week is a better prospect for an airport parking ad than someone who typed "airport parking" out of curiosity.

The less comfortable version: your own transactional emails could become targeting fuel for someone else. If a customer's order confirmation from your store sits in Gmail, and that customer has connected Gmail to Search, an inference like "recently bought running shoes" is exactly the kind of thing a competitor's broad match or AI Max campaign could end up matching against. Google hasn't said it works this way, and I wouldn't claim it does. But nothing in the policy text rules it out either, and that's worth sitting with for a minute.

How big is the affected audience? Nobody outside Google knows

Google hasn't published how many people have connected apps. Some context: AI Mode had more than 75 million daily active users when Shopping ads arrived there in February, and Personal Intelligence has since expanded to nearly 200 countries (the help page puts the connected-apps feature in 188 countries and territories, 18+ personal accounts only). In an interview with PPC Land, Google's Nick Fox described the feature as opt-in, with not everyone turning it on.

My guess is the connected slice is still small relative to total Search volume. Which, honestly, makes the measurement problem worse rather than better. A small, high-intent cohort that suddenly gets better ad matching can nudge conversion rate a few points on accounts heavy in AI Mode traffic, and it will look exactly like your last bid strategy tweak working.

Draw a line at October 16 before you read any Q4 results

This is where I'd spend 10 minutes this week. None of it requires Google to cooperate.

• Annotate October 16 in every report you use for Q4 decisions. Looker Studio, your GA4 annotations, the shared doc your client reads. If AI Mode traffic is a meaningful share for you, treat everything after that date as a new regime.
• Set a do-not-credit threshold. My rough rule: if conversion rate on broad match or AI Max campaigns moves more than about 15% week over week with no structural change on your side, don't attribute it to the bidding change you made that week. Hold the change for two more weeks and see if it survives.
• Read the search terms report for "personal context" queries. Look for longer, oddly specific queries that reference dates, bookings or past purchases. Those are the ones most likely to be riding on connected-app inferences.
• Don't stack changes in late October. Black Friday prep already tempts teams to change three things at once. With an invisible signal entering the auction, one change per campaign per week is the only way you'll know what did what.

If you're also dealing with Google's other recent defaults, like PMax pulling images from your site starting December 4, the same logic applies. Q4 is already crowded with platform-side changes you didn't make, and it's easy to mistake them for wins.

The Gemini app is probably next

When Personal Intelligence launched in January, Google said there were no ads in the Gemini app, and it said the same about AI Mode with connected apps in March. One of those two statements has now expired. The policy edit explicitly names Search or Gemini Apps as services users can let read their other Google data, so the plumbing for the second one seems to already be in place.

My prediction: Google tests ads for Gemini app users with connected apps within 12 months, and it does so with the same zero-reporting setup advertisers are getting now. I'd put that at around 70%. If I'm wrong, it'll likely be because the Verge-style privacy coverage gets loud enough to slow it down, not because Google changed its mind about the signal's value.

And to be fair, none of this is entirely new. Google's policy still described analyzing emails for tailored advertising in late 2017, dropped that language in 2018, promised in writing in 2021, and now has an exception. Call it a return with better branding. What's actually new is that the inference layer sits inside an AI answer, where it's even harder for anyone outside Google to see what triggered what.

I don't think there's anything to panic about here. I'd just be a lot slower than usual to take credit for a good November.