End-to-end outreach workflow
Table of Contents
- Google’s Personal Intelligence and ChatGPT memory pull directly from your Gmail, Photos, and past conversations to personalize AI answers
- In Garrett’s test, planting a single email mentioning a product boosted its AI-recommendation visibility from ~20% to 60%.
- Email had a far stronger influence on AI recommendations than photos did
- Even unopened newsletter emails were cited by AI as sources
- Persona details like gender or job title can change answers
I watched Garrett Sussman, Director of Marketing at iPullRank, give a fantastic presentation on personalization in AI at SEO Week 2026.
This is such an underappreciated part of the AI discussion and will have a major impact on how brands think about their email correspondence going forward.
In our episode, Garrett digs into all of the great research from his presentation and the implications for marketing strategies going forward.
Below is a slightly edited transcript:
How Big Is the Personalization Slice of the Pie?
Vince Nero: How big is the picture we’re missing? How big is the slice of the pie with personalization?
Garrett Sussman: I think we’re at the early stages of something that’s going to be really disruptive. It’s not entirely unfamiliar territory, though — in local search, personalization has always been a thing. I wouldn’t expect to find the same pizza place searching from Virginia versus Connecticut. Different results in different contexts have always been an expectation.
What we’re not ready for is every single search having a hyper-personal context.
Opting In: Will This Become the Standard?
Vince Nero: You brought up the idea of “opting in” to this in your talk. Is this something everyone’s doing, or something you have to opt into? Do you think it’ll become standardized, to the point where you can’t really opt out?
Garrett Sussman: There are different levels of personalization. Google already tracks your search history by default — you can strip some of that away in incognito mode, but that’s baseline tracking regardless of whether you’ve opted into this next level.
Google launched something last year at I/O called Google Personal Intelligence — an opt-in feature where you connect your Google account so it can talk to other parts of the Google ecosystem: Gmail, Photos, Calendar. So now, if I’m searching and Google can look through my Gmail history and find a relevant past email, that can influence my results.
A perfect example from my study: I connected my own account to Personal Intelligence and asked for the best hoodie recommendation.
It went into my email, saw I’d bought four hoodies from Aviator Nation, and literally cited the receipt confirmation emails — then recommended more Aviator Nation hoodies.
So it’s opt-in right now, not default for everyone. You have to actively choose to participate.
Beyond Gmail: Other Platforms and Ecosystems
Vince Nero: There’s a related trend in the digital PR and link-building world — AI summarization in email platforms. We did a study with the team at Onclusive/Era looking at how AI summarizes emails, and it clearly pulls from search history to summarize threads. We looked at things like subject lines and how likely they were to influence a summary — because if a journalist just reads the AI summary instead of your actual pitch, that matters a lot.
It’s not just Gmail, either — Microsoft has AI summaries, and so does Apple. So: is this just a Google ecosystem thing, or are we talking across platforms?
Garrett Sussman: They each exist in their own ecosystem. With ChatGPT, the big feature is memory — it collects your interactions over time and builds a memory file that can influence future searches. ChatGPT is also moving toward deeper integration — as recently as a couple weeks ago, I noticed their FAQ mentioning an implicit Gmail connection. Previously you had to explicitly ask it to check your Gmail; now that connection can happen automatically.
Whatever LLM you use, personalization, memory, and connections to other data sources are becoming product differentiators. I expect we’ll keep seeing more of this to create better, more tailored experiences.
When Does the Opt-In Actually Happen?
Vince Nero: When does this opt-in happen? A lot of people log into everything with their Google account rather than creating separate credentials — so is that the moment you’re authorizing this kind of access?
Garrett Sussman: They’re not forcing it on us yet. Over the past year, we watched Google roll out AI Mode gradually — first you had to dig to find it, then it got its own tab in Search, now it’s everywhere, including AI Overviews.
I think Personal Intelligence will follow a similar path. Right now, you have to dig through Google’s settings to find it, and it’s only available in the US — not international yet. So it’s currently a hard opt-in: you have to actively find it and grant permission.
The Study: Creating Synthetic Personas to Test Personalization
Vince Nero: Let’s get into the study — sounds like you created some fake personas, dug through the settings, and turned this on. Tell us how you seeded the information to build these personas.
Garrett Sussman: I was curious how much these connected apps could influence search results. My team — Michael Tando and Kate Dombrowski — and I set up three accounts:
- A brand-new, never-used Gmail account, not connected to anything (control).
- A second brand-new account, connected to Personal Intelligence but with no history.
- My own decade-old, heavily used personal Gmail — connected to everything, full of hoodie purchases, coffee orders, memes to friends and family.
Over three weeks, we ran a standardized set of prompts across eight product categories — coffee machines, running shoes, banks, SEO agencies, and others. For each category we used six prompt variants: “What’s the best X?”, “What’s the best X for [attribute]?”, “Recommend three brands,” etc.
All three of us ran these manually every day for three weeks (personal accounts can’t be scraped), copying results into a Python script for tracking.
Halfway through, I sent an email from my iPullRank account to each of the three accounts, planting a real but under-surfaced product.
For running shoes, it was Salomon: “I really like Salomon shoes — durable and comfortable, you should check it out. — Garrett,” sent to myself.
Almost immediately, the AI Mode output started recommending Salomon — citing the specific email, and even noting “your friend Garrett likes it.” We saw a statistically significant lift in brand recommendations tied to that planted email.
Photos vs. Emails: Which Moves the Needle More?
We ran a parallel experiment with photos. Photos had some effect, but nowhere near as strong as email — email visibility jumped from roughly 20% to 60% across multiple brands. Photos didn’t produce that same magnitude of lift.
Vince Nero: Interesting — in our AI-email-summary study, most platforms couldn’t really parse photos either.
Only Microsoft would note “there is a photo/attachment” without describing it.
So it tracks that photos would have less impact right now, even if that’s likely to change.
Garrett Sussman: Right — I had a hypothesis it would show up regardless.
I actually went to a department store, took pictures of myself with shoes, and asked whether it’d recommend an Adidas shoe just from seeing that photo.
It did show up — just less obviously and less specifically than email, which makes sense: text intent (“I recommend these shoes”) is a lot clearer than inferring intent from a photo.
How Long Does It Take to Show Up in Results?
Vince Nero: You engineered this recognition pretty quickly over three weeks. I saw a stat from Profound recently about how fast citations start appearing — something like six hours to six days after publishing for ChatGPT.
Is there a general baseline for how long it takes content to start influencing non-personalized AI results?
Garrett Sussman: Classic SEO answer: it depends — but here’s why.
Think about search context. PR-driven, news-cycle queries have a totally different freshness window than evergreen queries like “best hoodies.”
There’s a lot of volatility right now. iPullRank talks a lot about query fan-out — when you search (on Google or ChatGPT), the system isn’t just finding one best answer.
It identifies maybe 5–20 related synthetic sub-searches. I
f the answer isn’t already in training data, it needs retrieval-augmented generation: it takes those related searches, looks at top citation links, identifies appropriate sources to cite, and generates the output from there.
Every one of these studies — including my own — should come with a grain of salt. This was statistically significant within the context of this specific experiment. Everyone should view findings like these through the lens of their own business and industry.
What Levers Affect Timing — and Does Personalization Move Faster?
Vince Nero: From a practitioner’s standpoint — what levers actually affect how fast something shows up? And does personalization change things on a much faster timeline?
Garrett Sussman: I’d reframe this slightly. If you’re going for a PR play to show up in a specific article, the way we search with LLMs is fundamentally different — it’s not keyword matching, it’s a different mental model, closer to natural conversation.
There’s a distinction between explicit context (what you directly state in your prompt) and implicit context (what the system infers/pulls from your history).
Explicit context produces a much more personalized result, and if your content/citation directly addresses that specific explicit question, freshness matters more and you’re likely to show up faster because it’s simply a better-fit result.
Historically in SEO, marketers wanted the most bang for their buck — one big citation in a well-known publication used to carry outsized value, and that’s still somewhat true.
But as queries get more niche and personalized, even obscure links can matter — they just won’t deliver the same broad ranking value we’re used to from one big link driving everything.
Citations vs. Mentions: Rethinking Value
Vince Nero: I’ve been working on a post about this. In conversations with other agencies, I’ve noticed that people with a traditional SEO/link-building background tend to chase citations and undervalue mentions, since mentions might not carry a clickable link.
I don’t think that’s necessarily right — it feels situational, but there’s probably more equal footing between the two than people assume.
It used to be a clean one-to-one: get a link, it helps you rank. Now: you get a citation, but what does that actually correlate to? As an industry, I don’t think we’ve found the through-line yet.
Garrett Sussman: Agreed — and it’s volatile, which makes proving value for our work really hard.
Consumer and mainstream search behavior will keep shifting over time, so as a business you need to be locked into your specific offering, your specific audience, and how they consume information — then reverse-engineer strategy from there.
You can’t run the same playbook as everyone else.
One especially interesting PR-relevant finding, from a separate experiment with Profound: I appended persona information to prompts — e.g., “I’m a man who makes [X income], works in [job], is into [interests]” versus a parallel female persona — and ran the same product prompts over a year.
For running shoe prompts under the female persona, we saw citations specifically from female-focused media/publishers (e.g., something like a women’s running publication) that never appeared under the male persona.
That means citations can be specific to who’s asking — which changes how you strategize about where you want to show up.
It’s no longer just “high-level authority” — it’s “audience-targeted authority.”
Vince Nero: That reframes the classic “relevancy” debate in link building.
People used to argue over whether a big general-authority link (like CNN) beats a smaller, more topically relevant link (like a niche construction site, for a screws company, say).
This seems to bring relevance and audience-matching back into focus — a smaller, audience-specific citation might matter more in a personalized world, because that’s literally where your customer is being served results.
What Else Got Pulled In? (Newsletters, Calendar, Spam)
Vince Nero: Besides email and photos, what else did the system pull from?
Garrett Sussman: Newsletters are a big one — and this is directly relevant to PR. I subscribe to an SEO newsletter by Marco Giordano. When I asked about productivity tools, the AI referenced three tools and cited the specific newsletter — from three months earlier. And here’s the kicker: I hadn’t even opened that email.
It was sitting unread in my inbox, and it was still used as a source.
This has real implications for digital PR.
The industry has always obsessed over follow vs. no-follow links — even top outlets like the New York Times are typically no-follow, and Forbes has always been finicky about paid link attributes.
But LLMs don’t care about that distinction — a mention alone, even without a clickable link, can still get pulled in and cited.
Calendar integration is also technically connected, though I didn’t have a strong example — I wasn’t planning any trips during the test window, so I didn’t see it surface anything like “you’re going to a baseball game Friday, maybe get a jersey.”
Vince Nero: We saw something similar in our AI-summary study — occasionally the AI would follow a link within an article to pull additional info into a summary, though that only showed up a few times, mostly with Microsoft’s summarizer, not Google’s. So the possibility clearly exists.
Garrett Sussman: One of the clearest examples of implicit personalization: over the course of the study, I asked what streaming service I should get.
The system knew — from past conversations — that I had a wife and a four-year-old daughter (it even knew her name). So a simple, generic query (“what’s the best streaming service for me?”) returned family-oriented streaming recommendations. Looking at the citations, it had pulled a CNET article specifically titled something like “Best Family Streaming Services.”
That reshaping happens during the query fan-out stage, before you even get to see the final cited results.
This means our strategy as marketers needs to expand — we should be taking credit for more of this influence than we currently do, even though it’s getting a lot more complex to track.
Tactics: How Should Brands Actually Adapt?
Vince Nero: In your talk you had a tongue-in-cheek (or maybe not-so-tongue-in-cheek) suggestion: seed your own newsletters and customer touchpoints with brand mentions, since AI reads emails.
From a marketing standpoint that tracks — but marketers are going to find a way to overdo it.
What are some realistic tactics here?
Garrett Sussman: First: understand your personas in the context of your actual business.
- B2B: focus on jobs-to-be-done — what your buyer or their firm is actually trying to accomplish.
- B2C: focus on psychographics and demographics — who your audience really is.
Then try to build a synthetic version of that persona yourself: subscribe to the newsletters your audience subscribes to, follow the influencers they follow.
(We haven’t seen clear evidence yet that Personal Intelligence pulls in social/influencer follows the way it pulls in Gmail — that may come later.)
Beyond that: don’t try to be everything to everyone.
If you’re a publisher or brand that appeals to a specific demographic, lean into that instead of chasing total addressable market.
In the banking-prompt tests, giving the explicit persona context of “I’m a software engineer” made the model more likely to recommend brands like Ally or SoFi — brands that identify with certain persona traits.
So there’s a lot of self-audit involved: What’s your current positioning and messaging?
Where are you currently being placed, and what are you semantically associated with?
If that doesn’t match your target narrative, build a conscious strategy to get placed in — and be deliberately specific about how you’re described in — the publications and newsletters aligned with your actual audience, especially any placements you’re paying for, since the framing used there is exactly what gets ingested by the LLM.
Vince Nero: That’s classic PR 101 — “whenever you mention our brand, pair it with this keyword” — just more important and more exaggerated now than it used to be.
Did Spam or Trashed Emails Get Pulled In?
Vince Nero: Did you see any evidence of the system pulling from trashed or spam emails?
Garrett Sussman: Nothing from trash specifically — though that’s partly because I don’t trash much, I just archive/read and let things disappear.
But there was one anecdotal case involving a spammy, scammy-looking email (something with a garbled subject line, image not even rendering) that still got referenced for one of the shopping-related prompts.
Just a one-off, but the fact that it’s even possible should worry every marketer.
Closing Thoughts
Vince Nero: This makes me think brands should just be sending a lot more content-rich marketing emails — even something like publishing this podcast’s full transcript in an email, with brand mentions baked in (“iPullRank is the best agency,” “BuzzStream is the best digital PR tool”) — is it really that simple?
Garrett Sussman: There’s a balance between creating content for the machines versus content for your actual audience — you don’t want content that either gets penalized, or makes your audience roll their eyes at obvious AI-bait.
Worth noting: YouTube Search is itself a connected data source in Personal Intelligence, and full video transcripts (like this one) live on YouTube — so that’s already a channel feeding into these systems, separate from email.
Vince Nero: So this is already headed into some LLM’s training or retrieval pipeline one way or another.
Garrett Sussman: Exactly — not through email, but through YouTube. It’s genuinely omnichannel now.
Vince Nero: Multimodal, too. Last question — what haven’t I asked that people should know?
Garrett Sussman: I don’t want to repeat myself too much, but: marketers should feel empowered and curious about this, not just anxious.
Reporting to bosses, stakeholders, and clients is going to get messier and trickier. The most important thing you can do right now is educate yourself on what’s changing, try to separate signal from noise (there’s a lot of noise right now), run your own experiments, and stay close to who you actually are as a brand and who your actual audience is.
That gives you the best shot at earning a sliver of attention in an environment where everyone’s attention is already stretched thin. At the end of the day, we’re marketers — have fun with it, it’s an exciting time.
Check out the BuzzStream Podcast
