28 Wishes sells ice cream in Los Angeles. Below 70°F, sales fall about 20%. The weather is out of their hands. Rent isn't.

So the owners started putting about $20 a day into Kalshi weather markets, taking the cold side. The days that keep customers away now pay something back.

This is hedging. Big companies have done it for decades, buying protection against bad weather, fuel spikes and rising rates. It used to take a broker, a trading desk, and an order size no corner shop could meet.

Kalshi opens it up. Contracts on weather, fuel prices, inflation, tariffs and regulation, starting at a few dollars. Take a position on the outcome that would hurt you. If it hits, the payout softens it. If it doesn't, the contract expires and the good month was the point.

See how other owners are doing it at Kalshi.com/hedging

This week's verdict on LinkedIn was that AI search is a distraction. Marie Haynes looked at 12 months of GA4 data across 100 sites and found ChatGPT sent 0.62% of the traffic Google organic did. SEO Works found organic search sent 39 times more traffic than AI assistants, and Kevin Indig couldn't find AI traffic hiding in Direct. Rand Fishkin then rolled all three into a whiteboard video, and the takeaway, as I read it, was to stop wasting time on GEO and get back to clicks. Search Engine Roundtable has the round-up.

I have no issue with their numbers, and I'd expect to find much the same in most analytics accounts. My problem is the question they're asking.

Measuring AI search on referral traffic is like judging a goalkeeper on how many goals he's scored, while the club wins the league on the back of his clean sheets. No manager would drop a keeper for failing to score, yet that's exactly the test AI search is being put through.

The buying journey inside an LLM works differently to search. Someone can do the whole thing in the chat window, from working out what they need to comparing options and deciding who to buy from, and the model talks back to them while they do it. That conversation feels like talking to a person, which builds a kind of trust and confidence a page of blue links never did. By the time they click, if they click at all, the decision has largely been made.

Haynes's own data shows it. ChatGPT visitors converted at 13.8% against 3.02% for Google organic, more than four times higher, because they turn up already sold.

Fishkin has built much of his profile over the last few years on telling marketers to stop obsessing over clicks, because people get what they need without ever leaving the platform. He's right about that, which is exactly why this whiteboard is so hard to take seriously.

The moment the channel in question is AI search, influence without a click apparently stops counting and referral traffic becomes the yardstick again. You can't preach zero-click one day and then write off a channel for its lack of clicks the following day. He even opens the whiteboard by asking "Is AI visibility a boondoggle? A sham? A distraction?", which gives away the answer he was heading for before he'd even started posting.

He has also compared the AI rush to the dot-com bubble and called today's AI "spicy autocomplete", and his own research argued that AI recommendations are too inconsistent to track reliably, so three studies that appear to back him up were always going to get the whiteboard treatment. It reads as confirmation of a view he already held. When someone with an audience his size tells marketers to walk away from a channel their customers are already using, that advice deserves more scrutiny than it has had this week.

This is the same Fishkin who, in a LinkedIn comment about one of the big AI visibility platforms, said his "Spidey senses" told him they were "bad news in more ways than one" (I'm unsure anyone over the age of 12 should be using the phrase "Spidey senses", but whatever). Gut feel was enough evidence for him there, but apparently AI search has to prove itself in GA4 before anyone should take it seriously.

Not everyone took the bait. Mike King at iPullRank published his response the same day, and it's a stance I agree with. He accepts the numbers, agrees that schema tricks and llms.txt hacks are a waste of time and agrees that Google still sends the most visitors. His point is that the studies measure a branding channel with a performance ruler, which he likens to judging a Super Bowl ad on its click-through rate.

He also gets into why GA4 undercounts AI's influence in the first place, which is the part most marketers miss:

And even if GA4 caught every one of those visits, it would still only show you a tiny piece of the puzzle, because it only sees the people who make it as far as your site. Every recommendation, comparison and shortlist that happens inside the chat is invisible to it. Writing off AI search because GA4 can't see it is like pointing a telescope at the one empty patch of sky and declaring the telescope broken. The telescope's fine. You're just looking in the wrong place.

King's advice to CMOs is to budget for AI search the way you'd budget for PR or brand, and to track share of voice, how often you're mentioned and what's being said about you. That's the clean sheet stat, and it's the one nobody in the press conference thought to ask about.

Some of the pushback is plain discomfort. Marketers don't like losing the data, and I get that. When I spoke to Christian J. Ward, Chief Data Officer at Yext, back in December 2025, his view was simply that you aren't going to get all the data, and he was fine with that. And this is a chief data officer talking.

We've been hooked on Google Analytics and last click for so long that we've forgotten its flaws, even though anyone who has sat through a Monday morning meeting arguing over who gets the credit for last week's conversions knows them well. A real journey might run direct, direct, paid, organic, paid, email, organic, and organic walks off with the credit. Packaging that into a neat report makes us feel better about what we're doing, though it has never described how people actually buy.

Back in 2000, plenty of business leaders and plenty of the media thought the internet was a fad. Alan Rusbridger's book Breaking News is worth a read on how The Guardian went the other way and planned for what digital would do to its business.

I've told this story before.

In the late noughties I worked with a betting client whose CEO wanted to abandon the website, and the internet altogether, because as far as he was concerned his punters were all on the high street. The iPhone had launched two years earlier, connectivity was getting better and mobile betting was about to reshape the whole industry, yet he was judging the channel on where his customers were that day.

Judging AI search on attributable traffic is the same mistake, and it'll look just as strange in a few years once agents are doing a lot of the researching and buying for us. Most people aren't sending agents out to do their shopping yet, but that's where this is heading.

Before Obsero, I spent a year sitting next to a sales team at a fintech, which is neither a small company nor an AI business. They doubled down on self-reported attribution because they knew the impact AI search was going to have, and over that year the answers to "how did you hear about us?" went from "Google, Google, Google" to "ChatGPT, ChatGPT, Perplexity".

I've said it before and I'll keep saying it: measure this through to revenue.

Even Kevin Indig, in one of the studies being used to dismiss the channel, ends up recommending the same approach. In practice that means three things:

Self-reported answers aren't perfect, because people tend to name the last thing they remember, and the event, the social post or the TV ad that started it all may never get a mention. That's fine. Chasing every last data point costs time you could spend on the things that actually drive revenue, and it's time we stopped being so insecure about not having all of it.

AI search doesn't have to beat Google on traffic to matter, any more than a goalkeeper has to outscore the striker to be worth the money. Judge it on whether you're being recommended and on the revenue from customers who tell you they found you there, and judge anyone telling you to ignore it by the same standard they'd apply to every other zero-click channel.

The brands waiting for perfect attribution before they take this seriously are making the same call as that betting CEO, and history will look back on it the same way.

So you show up in AI search. Great.

The next question is how you show up, because what ChatGPT or Gemini says about you, how it positions you against competitors and the story it tells a buyer when you're not in the room all shape the decision long before anyone reaches your site.

That question kept coming up with the brands we work with.

Being mentioned is only half of it, and whether you're being recommended or talked down matters just as much as being in the answer at all. If the first half of this edition was about finding the clean sheets, this is how you count them.

That's why we built Sentiment Analysis Plus, which went live for every Obsero customer on Monday. It scores the sentiment of every answer your brand is mentioned in, across every LLM we track, and shows you how that score moves over time.

From there you can drill down as far as you want:

AI is broadly polite about most brands, so a raw positive score can flatter you.

One brand from a sector people famously love to hate still came out 59% positive, which tells you the useful question is how you compare with others in your space, and that's where we're taking the scoring next.

The same attribute can cut both ways as well. Sustainability, for example, can show up as praise in one answer and as criticism in the next, so it's worth reading the detail beneath the headline number before you decide what to fix.

And, as ever, it comes back to the prompts you're tracking. One of the prompts behind that 59% asked what the company is like to work for, which has nothing to do with the environmental record people love to hate it for. Ask a different question and you get a different score, so a sentiment number is only as useful as the prompt set behind it.

Visibility tells you whether you're in the conversation. Sentiment tells you whether that conversation is helping you win the customer.

If you want to see it on your own brand, hit reply and I'll set you up, or read more about how it works.

That's it for this Sunday. Until next time.

Have a good one.

Andy