Men, Say Goodbye to Eyebags, Dark Spots & Wrinkles
Particle Face Cream is a 6-in-1 formula engineered specifically for men's skin. It reduces eye bags, dark spots, and wrinkles, restores firmness, hydrates deeply, and revives dull tone. Multiple premium anti-aging ingredients, clinically researched, built into one product that actually fits your routine.
Over 1,000,000 men have added Particle to their daily routine. Easy, effective, and worth the two minutes. Try it risk-free with a 30-day money-back guarantee.
On Tuesday Mike and I ran Obsero's first webinar, How to measure AI search, and if you couldn't make it, this edition is the write-up.
The short version is that the customer journey has moved inside the AI answer, which means referral traffic only ever shows you the trickle that leaks out of it. So we walked through what to measure instead and, because CFOs are now asking the question as well as CMOs, how to connect it to revenue.
Your customers have moved, your measurement hasn't
Mike opened with two numbers. 77% of US consumers used AI to help them shop in the last six months, according to Exploding Topics, while clicks on Google's number one result fall by 58% when an AI Overview appears. Ranking first in Google isn't the traffic driver it used to be, and most brands are already seeing that in their organic numbers.
His point, which I agree with completely, is that most bad decisions in AI search come from jumping to tactics before understanding how the customer journey has changed. He set out five shifts that all head in one direction, into the answer:
Search terms to prompts. We all learned to search the same few head terms, whereas prompts are long, conversational and personal, so the value has moved into the long tail.
Google to platform fragmentation. ChatGPT alone has 1.2 billion weekly users, and Claude, Copilot, Gemini, Perplexity and Google's own AI surfaces each do different jobs for different people.
Links to answers. Research, comparison and the decision can all happen in a single conversation, which makes being mentioned and cited the new click.
Brand-owned narrative to LLM-owned narrative. Your website used to be where customers decided whether you were for them, and now the answer tells them who you're for, who you're not for and what you're good at.
Checkout to agentic commerce. Mike's view is that ChatGPT's new business profiles exist to get more product data into the platform, which brings forward the point where people buy without ever leaving the chat.
Every stage of the journey has changed, so measurement has to change with it.
Why referral traffic is the wrong metric
You'll have seen the "GEO is a distraction" argument doing the rounds on LinkedIn, which I wrote about a few weeks ago. Judging AI search on the traffic it sends is like signing a goalkeeper and then asking why he hasn't scored any goals, because you'll get a disappointing answer to the wrong question.
Google was built to send you to websites, because the click was the product, whereas LLMs are built to keep you talking, because the conversation is the product. Across the 10 million AI answers we've tracked at Obsero since September 2025, only 12% include an outbound link once you set citations aside, so there's often nowhere to click at all.

Only 12% of AI answers include an outbound link, across 10 million answers tracked in Obsero.
Think about how you use these tools yourself. I'll ask ChatGPT or Claude for a recommendation and then open a new tab and search for the brand, or type the URL, which means the influence lands in analytics as branded search or direct and only rarely as a referral. For an FMCG brand it can be even less visible, because the answer might not lead to a visit at all, just a choice in the supermarket aisle a week later.
Referral traffic measures the leak from the bucket, not what's in it.
The four metrics that track AI search
The rule we gave everyone was to measure what happens inside the answer rather than what leaks out of it, and that starts with how you set up your prompts:
Track topics, not single prompts. Answers change from run to run, so one prompt is an anecdote while a topic of 25 to 50 prompts shows you the trend. Fewer than that isn't reliable, and more is wasted spend.
Split brand and category prompts. "What are the best jeans for men?" tells you whether you get discovered, while "Are Levi's good quality?" tells you what AI says about you, so report them separately.
Track more than one platform, but not all of them. Go where your customers are, because B2B buyers use Claude and jeans shoppers mostly don't.
From there we track four indicators, which we call leading because they move weeks or months before any traffic does. We used Levi's as the example throughout.
Visibility: are you in the answer?
This is the share of your tracked prompts where the brand is mentioned, across platforms and markets. Splitting it by platform is where it gets useful, because being strong in Google's AI Overviews and weak in ChatGPT points you towards a different set of sources to work on.

Levi's appears in 59% of ChatGPT answers but only 26% of Google's AI Overviews.
Prominence: where are you in the answer?
Being the tenth brand named isn't the same as being the first. Google taught us that higher means recommended, and people stop reading halfway down an answer in the same way they stopped scrolling a results page, so we score mention position and whether you're recommended or merely listed.
Sentiment: what is the answer saying about you?
You can appear in 90% of answers and still be described badly. I saw an example recently where an affiliate was recommending a competitor over our client because it was supposedly cheaper, which wasn't true, so fixing it was a simple correction request to the affiliate. When the negative sentiment is accurate, that becomes a conversation for your product or brand team instead, and either way you'd never spot it in a traffic report.
When a prompt is grounded, the model searches the web and cites sources, whereas something like "Who won the 1966 World Cup?" (Mike's choice of example, as a proud Scot 🏴) comes straight from training data. For grounded prompts the citations are the lever, because if you change what gets cited, prominence and sentiment follow.

The sources behind AI answers about jeans are mostly fashion magazines, with Reddit and YouTube close behind.
I looked at a popular running app recently (Strava, as I owned up to in the Q&A) and only 5% of its citations came from its own website, with the rest coming from the App Store, Trustpilot, TechRadar and other third parties.
Put those four together with a share of voice figure against your named competitors and you have a CMO report that doesn't rely on traffic at all. The rhythm we recommend is to track daily, optimise weekly and report monthly, benchmarking against named competitors rather than against yourself.
None of this replaces having a product worth talking about. AI answers are built from what customers, reviewers and the press say about you, so the brands that win are usually the ones people already recommend.
How to prove it to your CFO
Three to six months ago that's where the webinar would have ended. The question has moved on, because money is going into GEO and CFOs want to see the return, and you can't attribute your way out of this.
For years, multi-touch attribution made everything easy to justify, but it relied on cookies and the click, and we've lost both. Someone who asks ChatGPT for the best jeans on day one and types levi.com to buy on day ten gets credited to direct, and the AI answer that made the decision gets none of the credit.
Instead, we use three lenses.
1. Direct and branded traffic
If your AI visibility is working, branded search and direct traffic should rise with it, while non-branded organic search falls as AI answers take its share. Set a baseline, take action to grow visibility and build the correlation over time. For most of the brands we work with this is the main way they tie impact to action, because the website is still where the transaction happens and the traffic is simply arriving from a different place.

When AI visibility grows, branded search and direct traffic should follow it while non-branded organic search falls.
2. Experimentation
Pick one product, audience or market, change what gets cited through content, PR and third-party pages, and hold a comparable one back as a control. Then watch visibility, prominence, sentiment and citation share move and see whether traffic, enquiries and sales follow. Country is the easiest split if you operate in more than one market, and every test adds to your own GEO playbook at the same time.
In the illustrative example we showed (not a real client result), one product line had its fit guide rewritten and earned three publisher roundups over eight weeks. Its visibility went from 34% to 51% while the control barely moved, and its traffic, enquiries and sales rose 18% against 2% for the control.

An illustrative GEO experiment: the test product line moved while the control stayed flat.
3. Self-attribution
Add "AI assistant (ChatGPT, Gemini…)" to your "how did you hear about us?" question. It feels very 1990s, and any single month will be noisy because nobody really knows why they chose one brand over another, but the trend over time is the signal. If you're measured on something like organic demo bookings, our view at Obsero is that this is the best source of attribution you've got today.

Add an AI assistant option to "how did you hear about us?" and watch the trend over months.
A few things from the Q&A
Being featured vs being recommended. Small brands can win in AI search in a way they couldn't in Google, because prompts carry far more context about who's asking. Track the prompts you have a genuine right to show up for and focus your work there. We'd love Obsero to appear for "best martech in 2026", but "best AI search analytics platform for a UK mid-market brand" is where we can win, and do.
How long do wins last? Content gets cited faster and decays faster than it did in SEO, which calls for rapid experimentation and regular updates. Just like core updates, you can't rest on your laurels, and a model change or a spike in OpenAI's search bot visiting your site is a signal to act on.
AEO or GEO? I don't care what you call it, because time spent arguing about terminology is time not spent thinking about the customer. The market has settled on GEO, and the useful difference is the goal: people judge SEO on winning the click and GEO on being in the answer. The techniques and tactics overlap heavily, so most teams already have the skills and just need to measure something different.
Does it differ by vertical? Massively. In three pitches we ran the week before, five of the top 10 cited sources for a fintech were affiliate aggregators, five of the top 10 for an airline were social platforms and seven of the top 10 for a fashion brand were fashion publications. Local and ecommerce need their own approach too, and affiliate and PR teams care more about being the cited source than the recommended brand.
What about apps? The same principles apply, with app stores, tech publications and query fan-out all worth watching, and connectors being pushed to users are starting to become a growth lever for app teams.

The sources that shape AI answers change completely by category.
FREE AI VISIBILITY SNAPSHOT
Want to see your own numbers?
Get a free, human-reviewed snapshot of how your brand shows up in ChatGPT, Gemini, Copilot, Google AI Mode, AI Overviews and Perplexity, emailed to you with no account needed. It covers:
Your share of AI answers to the questions your customers ask
How you compare with the competitors named alongside you
Which third-party pages the models read to build their answers
Where you're invisible and what to fix
Next up in December is the GEO playbook, covering which tactics actually move these numbers, and we'll email the date out shortly.
🗣️ This week's stories

What caught my eye in AI search this week.
Salesforce has launched an AI visibility tool, and Marc Benioff is calling it "goodbye SEO". Palmata shows what ChatGPT, Claude, Gemini and other AI tools say about a brand, the sources shaping each answer and what to change, and claims to simulate the impact before you publish. My take is that Salesforce is late to the party and "goodbye SEO" is pure nonsense. The techniques overlap, so much of the work that earns you rankings and referral traffic in the ten blue links also earns you a recommendation in AI answers. Telling brands to drop SEO is fearmongering to sell a product. Bolting AI visibility onto a CRM is a very Salesforce move. When we picked a CRM for Obsero, Salesforce didn't even make the shortlist, because the interface still looks like it was designed in 2005 (with a UX to match), so we went with Pipedrive. Marc Benioff on X
Google is putting AI Mode on its homepage. A collapsible sidebar on the desktop homepage now lists your recent AI Mode conversations, so you can pick one back up, start a new thread or connect a Gemini Notebook without ever running a classic search. Google's Robby Stein says it's rolling out globally because people are bringing Search more questions that "aren't just one and done". For a page that has barely changed in 25 years, this is a huge shift, and Google is actively steering people from typing a query towards having a conversation, which is exactly the change in behaviour we covered in the webinar. Search Engine Roundtable
OpenAI has already rewritten the ChatGPT Business Hub terms. Days after launch, it revised them when lawyers warned the original wording could limit businesses' ability to bring legal claims against it. If you claimed your business after last week's edition, it's worth reading the new version. Digiday
Anthropic says no to ads, but Wall Street may have other ideas. Now that it has public shareholders, a huge compute bill and an expensive free tier to fund, Digiday asks how long Claude's no-ads pledge can last. Digiday
Anthropic is now a public company. It raised $81 billion after pricing its IPO at $83 a share, above its $70 to $78 range, for a valuation of around $1.5 trillion, and closed its first day of trading 17% higher. The Information
Meta is pricing its Muse Charm at $300, with a year of free 5G. The pocket-sized device comes with up to 12 months of Verizon data so people can talk to Meta's Muse assistant wherever they are, which is one more way into AI answers that never touches a browser. The Information
Personal agents are already doing the shopping. Instinct is approaching $1 billion a year in transactions on a small invite-only user base, half of it travel, with its agents searching and booking through browser startups Kernel and Browserbase, whose revenue is climbing fast as a result. The Information
That's it for this time. If you sent in a question we didn't get to, drop Mike or me a message on LinkedIn and we'll pick it up.
Have a good one.
Andy



