EP 268 - AI Overviews, Brand Search & Local SEO: Cyrus Shepard on Earning the Click - from the archives

Editor's note (replay): This is a vacation-week re-run. The conversation was originally recorded and aired in April 2026 as Near Memo EP 251; it is being republished now (EP 269) because it still speaks directly to our ongoing series on brand in local. A live panel on local digital brand strategy follows next week.

Google can't tell what good content is on its own, so it watches what users do after they click — which means the durable levers are the ones that manufacture real-world demand and complete the user's task, not the ones you tune on a title tag.

Cyrus's Three Click Signals

Cyrus frames Google's post-click measurement as three nested signals, stated around (05:06–06:54). This is the spine of the episode — everything else (brand search, AI Overviews, on-site engagement) is a way of feeding or proxying these.

Takeaways

• Google was "gaslighting" us on clicks. For years Google denied using click data while Bill Slawski documented the patents and reps were under a stated no-comment policy; the antitrust trial then confirmed it's one of the top signals alongside content and links. (02:06–03:36)
• Three sources, one story. The court testimony, the Google patents, and the API leak independently line up — the leaked endpoints match what surfaced in court, giving a rare cross-confirmed picture of what data Google collects. (03:36–04:26)
• Last longest click is the conversion proxy. It's Google's stand-in for "task completed" — the user got what they wanted and never returned to the SERP; Cyrus reframes it as "I've completed whatever I set out to do." (06:36–07:15)
• Own the whole task, don't rent it. If reviews or booking live on Yelp instead of your site or profile, the satisfying action happens on someone else's property — AJ Kohn's advice is to consolidate the data you can so users never have to search elsewhere. (08:57–10:48)
• Brand search is a first-class ranking signal. Strong branded query volume "always, always" correlates with higher rankings in Cyrus's correlation studies; Google's helpful-content guidance wants either an existing audience (brand) or a clear intended one. (11:31–12:36)
• Rand proved clicks a decade before the leak. At a MozCon keynote (~2013), Rand Fishkin had ~1,000 attendees run the same search and click one business — it shot to #1 within the hour, foreshadowing the DOJ documents by a decade-plus. (12:36–13:35)
• Off-search demand counts too. Google can read popularity signals from Chrome, so traffic driven by a TV spot, billboard, or radio ad registers as demand even when you're not ranking — "there must be some demand there." (13:35–14:23)
• AI Overviews mostly benefit Google. They cut traffic to information publishers (the cost of mediocre info is now zero) while keeping users on Google, whose AIO links often point back to more Google search. (15:32–16:20)
• Google is guesstimating clicks at the edges. For long-tail queries Google reportedly built lighter systems — Cyrus cites FastSearch and a "RankEmbedBERT"-type model [ear-check] — using ~10× less data plus 20 years of history to predict clicks and verify with far fewer real ones. (16:20–17:08)
• In local, AIOs can increase downstream clicks. Near Media's personal-injury research (presented at the Vegas legal-marketing summit) shows consumers use AIOs to orient, not to finish — they aren't lawyers, so they still click through, often via follow-up brand searches. (18:07–20:33)
• Watch branded search as your AIO scorecard. Google's new branded-query filter in Search Console makes it a one-click read: rising brand search likely means you're being cited in AI Overviews; falling brand search is the warning sign. (20:33–21:23)
• Don't fake the click. A too-clickable title that over-promises lifts CTR for a couple weeks, then traffic drops as users bounce — "you're making a promise with your title and the result has to deliver that promise." (23:15–24:00)
• Navigation is an underrated engagement weapon. Testing different nav links and wording measurably moves time-on-site; Cyrus's favorite GA4 report is organic engagement, which tracks with downstream rankings. (24:00–25:18)
• Let people leave satisfied. It's fine if a visitor gets a quick answer and goes — brute-force pop-ups meant to force a "last longest click" backfire; delay any video pop-up 30–45 seconds and make it dismissible. (26:31–27:41)

Related Reading

• Zyppy Signal (Cyrus's Substack) — Cyrus's click-data and AI-search research, including the click-signals post discussed here.
• Cyrus Shepard at Zyppy — bio and the firm's SEO testing work.
• SEO by the Sea — the late Bill Slawski's patent analysis, referenced as the early source on Google's click-data patents.
• Blind Five Year Old — AJ Kohn — the "own the data in one place" argument Cyrus cites.
• The Near Memo — the full episode archive; this replay ties into the current brand-in-local series.
• Google vs. DOJ, strategic SEO recap — background on the antitrust findings on click data.

Concepts

• Clicks / long clicks / last longest click — Google's three-tier read of post-click behavior; the last longest click (journey ended here, no return to search) is treated as the strongest signal.
• Dwell time — how long a user stays on a result before returning to the SERP; sliced by query type, language, and location.
• Task completion — Cyrus's reframe of a satisfied query: the user finished what they set out to do, whether or not money changed hands.
• Brand / branded search — volume of people searching for you by name; a real-world demand signal Google reads as popularity and, per Cyrus, a reliable ranking correlate.
• Off-search demand signals — popularity Google infers from sources like Chrome traffic, so demand created by TV, radio, or billboards still registers.
• Guesstimated clicks — Google predicting click behavior with lightweight models (FastSearch / "RankEmbedBERT"-type systems) for long-tail queries, verifying with far less live data.
• Earn / get / end — Cyrus's on-site sequence: earn the click (honest, attractive SERP listing), get the engagement (answer fast, draw them deeper), end the journey (satisfy intent so they don't return to search).

Practitioner Notes

• Brand search as the AIO proxy Near Media already recommends. The show's local research (personal-injury law) and Cyrus's advice converge: since AI Overview presence is nearly impossible to track at scale, use the Search Console branded-query filter as the standing scorecard for whether AIOs are feeding you demand.
• Consolidate reviews and booking onto owned surfaces. For local operators, moving the satisfying action (reviews, booking link) onto the site or Google Business Profile — rather than sending users to Yelp — is the concrete way to earn the last longest click Cyrus describes.
• Editorial footnote for show notes: the algorithm names in the AI Overviews segment ("FastRank," "BedBERT") are almost certainly mis-transcriptions of Google's long-tail systems (FastSearch and a RankEmbedBERT-type model) — confirm against the audio before quoting them.

Pick your poison:

00:00 Replay intro
00:38 Welcome & guest — Cyrus Shepard
01:46 Clicks as a ranking signal: the backstory
03:04 The antitrust trial: Google confirmed it
04:26 API leak: three sources, one story
05:06 The three click types
06:36 Last longest click = task completion
07:28 Bringing it to local & GBP
08:57 Own the task: reviews & booking
11:31 Brand search as a ranking signal
12:36 Rand proved clicks at MozCon
14:23 AI Overviews: who really benefits
16:20 Guesstimating clicks for long-tail
18:07 Local: AIOs drive more clicks, not fewer
20:33 Brand search as your AIO scorecard
23:15 Earning, getting & ending the click

Full Transcript -->

Replay (vacation week): originally aired April 2026 as EP 251; republished as EP 269.

Editor's note: This is a cleaned reading version. Stutters, false starts, and clear transcription errors were removed, and backchannel ("yeah," "right," "mm-hmm") was folded into the main speaker's turn. Wording is otherwise left faithful to what was said. Names and brands were normalized (Cyrus Shepard, David Mihm, AJ Kohn, Substack). Speaker-side factual claims and uncertain technical terms are left as spoken and flagged in the Corrections list at the end.

Mike Blumenthal (00:10)
We are on vacation this week, so we're bringing back an April conversation with Cyrus Shepard and David Mihm that discusses what actually drives rankings in 2026. It's a great episode, and no surprise, brand plays a big role — and it relates to our continuing discussion of how brand can and should play out in local. Join us next week for a great panel that discusses specific tactics around local digital brand strategy. Thanks again for listening.

David Mihm (00:38)
Hey everyone, welcome to another episode of the Near Memo. You might recognize that this is not Greg Sterling speaking. This is David Mihm, one of the other co-founders of Near Media, and I'm a relatively infrequent guest on the podcast these days. So I'm not sure what episode number this is, but I'm here.

Mike Blumenthal (00:55)
251.

David Mihm (00:56)
251, thank you, Mike. So we're one past the nation's birthday, apparently. I don't know what that says for the state of the podcast, but anyway. We are here with a very special guest today, Cyrus Shepard of Zyppy, a fellow Oregonian with myself. We're really excited to have Cyrus on to talk about a topic near and dear to our hearts at Near Media, as a company that does a lot of user testing of Google. We're here to talk about click data, user click patterns, how Google uses that data, and the impact that AI Overviews might have on this whole ecosystem. So Cyrus, welcome to the show. As we get started, maybe you can summarize for our listeners and watchers the thrust of the article you just published.

Cyrus Shepard (01:46)
Yeah, thank you, David. I'm happy to be back on the 251st episode. I feel like this is the SEO equivalent of the old radio show Car Talk. Do you remember that? Just veterans talking about SEO. Always happy to be here.

David Mihm (01:57)
Absolutely. If we last as long as Click and Clack, that will be a real feather in Near Media's cap.

Mike Blumenthal (02:06)
Although hopefully I don't go out with dementia, as one of those poor souls did.

Cyrus Shepard (02:10)
As may we all. So to get into it — the thrust of the article, and I'll go into a little history here; I don't want to get too long-winded — was that for years, people in the SEO space highly suspected Google was using user click data in its ranking algorithms. They always denied it, but there were experiments and patents. The great late Bill Slawski would analyze patents where Google would analyze click data, and Google would always say, "Well, just because we patented something doesn't mean we're necessarily using it. Don't believe everything you read." And then the Google antitrust trial, which happened I think two years ago now — evidence started pouring out that not only are they using click data, it is one of their major ranking signals. It's one of the top things they're using to analyze results, the other things being content and links, or anchors as they call them.

Mike Blumenthal (03:04)
So you're telling us that Google was gaslighting us, Cyrus?

Cyrus Shepard (03:07)
Gaslighting is a very good word. And apparently, in the court documents, it was discussed: they said, "We don't talk about it. We have a policy. You may want to talk about clicks, but we do not talk about it. You may disagree with that policy, but we do not talk about it." Some people call it lying, but these Google reps were in a very awkward position — they couldn't tell us the truth because it was company policy. And then we had a third thing, which was the Google API leak — a data-warehouse leak someone posted online where we could see the API endpoints of what Google's using in its algorithm. It didn't tell us how they were using the data, but we could see what data they were collecting. And interestingly, the endpoints they were collecting matched perfectly what they talked about in court and also the Google patents. So we have this record of how Google might be using clicks, confirmation that they're using clicks, and data endpoints showing exactly what they're collecting. I wanted to write a post that ties all these pieces together to show how Google is likely tracking and using click behavior — watching what people click on — so we can understand it better. There's so much data and so much noise out there; I just wanted to bring these pieces together so people could understand them.

David Mihm (04:26)
That's great. And one of the things that was really compelling about your post, for me anyway, as a sort of zero-click reader — the graphics you showed around bad click, good click, last long click, or whatever the exact attribute is called, are really impressive at showing the difference between those three things. Can you take us through in a little more detail what those three click types are that you identified in the API documentation, and then how those play into the positive or negative signals your website is sending to the algorithm?

Cyrus Shepard (05:06)
Yeah, absolutely. The big idea, which came out in the court case, is that Google, with all its amazing technology, can only guess at what good content is. They have a pretty good idea, but they're not really sure, so they're relying on us to tell them. When you see the 10 blue links or the top search results, that represents Google's best guess as to what should be ranking. And then they relentlessly watch what users click on to tell them: is this a good result or a bad result? Those are clicks, and they try to judge — are these good clicks or bad clicks? Reading through all the Google patents, time and again the number one metric they're looking at is dwell time — time on site, you can call it different things — how long a user stays on a certain result. They can take that data and slice it by the query: are you searching for something simple, like "how tall is Mount Everest?" You're not going to spend a lot of time on that page. But if you're looking for stock analysis, you might spend much more time. They can slice that data by language and location, and weigh all those things against each other to understand what users are saying is the best result. And then you have probably the most important endpoint, which Google calls "last longest click." That's a long click where the user spent significant time interacting with the result and didn't go back to the search result — meaning their search journey was satisfied. Maybe they searched for something else, maybe they go spend time on X or Facebook, but they don't go back to the search result and click on something else. Those last longest clicks appear to be the most important signal you can send to Google. So those are the three main ones: clicks, long clicks, and last longest clicks.

David Mihm (06:54)
So in some sense you might say a last longest click is Google's stand-in for a conversion on whatever the query was, right? Not necessarily that they ended up buying something, but that their query was satisfied.

Cyrus Shepard (07:08)
Their query was satisfied. The way I think about it is task completion — I've completed whatever I set out to do. When we're searching Google, power users like us might have 100 or 1,000 tasks a day that we're just trying to get through before we start new searches. And those are very important, apparently, to Google.

David Mihm (07:28)
Yeah. So I want to take this back to local briefly, since that's the main focus of Near Media, and then we'll go back to some of the other points you made in the article — which I think have impact not just on businesses trying to rank for local queries but all businesses. As it relates to local, the reason I was curious to get your take on query satisfaction is that we've hypothesized for a long time in local that the number of actions users take on Google Business Profiles, in addition to the engagement those profiles get — which you can take as a stand-in for the amount of time people spend on a website — scrolling photos, reading reviews, reading services, that's probably a pretty one-to-one comparison. The thing Google has with GBPs, especially on phones, is they can also track actions: clicks to call, which Google reports on in GBP data; clicks for driving directions; and potentially, for Android users, even if they don't click to call, they can probably track the number they're calling to see which business they're checking in on. So to us in local, it absolutely makes sense that Google's been doing this across the rest of the web, and they're having to find a proxy for that conversion action they have in-house data on when it comes to GBPs. Just as an aside — I don't know if you have any thoughts on that as a parallel.

Cyrus Shepard (08:57)
Absolutely — and just a caveat that you guys know so much more about local. But those Google Business Profiles are interesting. If you think about what I just talked about, task completion — I was thinking about a local example. If I had a dentist's office, think about the user journey. Someone refers you to a dentist; you Google them; you go to their website or their Google Business Profile. What's the first question they're going to have? They're going to wonder what your reviews are. If they have to leave your website and go to Yelp to look at those reviews, that doesn't send a good signal that your website is helping them complete their task. If those reviews are on your Google Business Profile, that can help — or maybe you embed those reviews on your website to have some of those signals. Then what's the next step they're going to want to do? Book an appointment, hopefully. Your call to action is, "Hey, call us at this phone number." But information's cheap — a lot of places can have your phone number. What if you have a booking link, and you can book and complete that action? Maybe that booking link's on your Google Business Profile as well. Completing that user journey — and Google's watching that journey and where it takes place — you can decide if you want it on your website or your Google Business Profile, but you don't want anybody else to own it. So for all of those actions, you want to make sure you're consolidated and own all of those places.

Mike Blumenthal (10:26)
So you don't want Yelp doing your reservations for you on their site.

Cyrus Shepard (10:31)
Right. AJ Kohn, an SEO consultant we all know in the Bay Area, is a big advocate of this: if other people have the data and you can collect it in one place, do that, so users don't have to search somewhere else — because you want to own that experience and own the task-completion journey. That's how it ties in.

David Mihm (10:48)
Absolutely. And in particular — not that you'd copy whole cloth, but featuring snippets of reviews from Yelp reviews that have been filtered is a great double win. It's content nobody else is going to see, because Yelp hides it as best they can, and it's keeping people on the site engaging with review content, which they'd otherwise have to go to third-party sites to find. And — we'll do a segue here into AI Overviews, which I know is another piece of the equation you've been analyzing the last month or so — the AI Overview might then pick up some of these review snippets when presenting answers about your business.

Mike Blumenthal (11:31)
Before you make that transition, though — conversions is one thing, but brand search is another. In local, brand search goes back to 2009, when Google would give you a Places sticker for your door if you had a lot of brand searches. So the history of brand searches goes way back in local, but clearly the rest of the web has finally recognized their value. And as you point out, one of the metrics Google appears to be using is the quantity of brand searches, right?

Cyrus Shepard (12:04)
Yeah, absolutely. Google says, with the helpful content update that came out and is now part of the core systems, they want to see two things from any website: you either have to have an existing audience — which we can translate as brand — or a pretty strong intended audience, a business model they can recognize that you're trying to draw people in. And whenever we run a correlation study between brand search and rankings, strong brands — people searching for you specifically — always, always correlate with higher rankings. Very real-world business signals.

David Mihm (12:36)
Yeah. And I remember — I don't remember the year, let's say MozCon 2013 — Rand gave a keynote, and at the beginning he asked people in the audience, a thousand people in Seattle, to all do the same search and all click on the same business in, I think it was a 10-pack at the time, or a seven-pack. It was a lot more than three. He gave his talk, and then at the end we all did the same search, and that business had shot up to number one in the ranking. So the combination of hot brand searches, even over the course of an hour, and clicks on the same business really led to a dramatic increase in that business's rankings and performance. For all of the gag orders Google introduced on its public representatives around the importance of click data, Rand proved it well before any of these DOJ documents or API leaks came out — this was a decade-plus ago.

Cyrus Shepard (13:35)
Yeah, I remember that. I remember it was a small Ethiopian restaurant or something that shot up to the top of the rankings. And you raise a really important point, David: one of the things that came out in the antitrust trial is that Google isn't just looking at search results. You guys were talking about real-world signals — people using directions with their phone — but Google can collect any data from Chrome browsers, and apparently that's a signal of popularity. So even if you aren't driving people through search — if you're driving them through other channels, getting them to search through a TV commercial or radio advertisement or a billboard — they can measure that data and see people are going to this website. There must be some demand there; it's a popular website. Even if they aren't ranking, we should give them some consideration.

David Mihm (14:23)
Yeah, 100%. And I will circle back to brand, Mike — there was a method to my madness.

Mike Blumenthal (14:28)
All right, all right. I just felt it was so fundamental, along with conversions. Sorry for interrupting — I'll zip it from here on out.

David Mihm (14:43)
All right, so let's get back to AI Overviews for a second, Cyrus. Obviously there's been a lot of doom and gloom, to put it mildly, in the web-publisher community about the impact of AI Overviews on the amount of traffic they're receiving, the number of queries they're present for, and so on. You also wrote what felt like a companion piece to your click-data post, about the impact AI Overviews might have on this click ecosystem. Can you summarize your thoughts? Obviously, in a world with fewer clicks, Google's going to have fewer direct signals to look at for last longest click when it comes to ranking publishers. So expand on that and take us through your logic behind that piece.

Cyrus Shepard (15:32)
Yeah. There's no doubt AI Overviews are decreasing traffic to a lot of publishers, especially those that profit on information — the cost of producing mediocre information is now zero for anybody. So that's definitely a thing. And AI Overviews also seem to highly benefit one business: Google. People are staying on Google longer, and a lot of the links in AI Overviews simply go back to Google searches, so a lot of people don't have to leave the Google interface. It's interesting about the click-data point you made. When Google developed AI Overviews, apparently the story within engineering circles is they couldn't use regular search results because it took too long to generate them for these long-tail queries. So they developed new algorithms — FastSearch and FastRank and something like RankEmbedBERT [ear-check] — that use far less data, like 10 times less data. I think they're using AI and machine learning to guesstimate click data. They have 20 years of data to build on, so they can pretty accurately look at a website and guesstimate how people will click, and then they need far fewer actual clicks to verify that. I think that's the future. I used to work as a Google quality rater — we've talked about this before — manually evaluating websites, and I think they're using a lot of AI and machine learning to do that job now. They're taking humans out of the loop to a certain extent. And they might do it better than I did, to be honest — I was kind of lazy at it, but there you go.

David Mihm (17:08)
You were one of the first people in the world to lose your job due to AI. All the hype now, and this was three years ago, right? When did they fire all the quality raters?

Cyrus Shepard (17:17)
Thanks, Google. But while we're on the subject of AI — we're talking about brand and reputation and all that — right now it's still kind of the wild, wild west. When people Google your brand, there's what other people say about you and what you say about yourself. And right now there's such an opportunity to say great things about yourself, if you know what questions people are asking. I see so many businesses own those AI Overviews with their own citations, and Google doesn't seem to care. I think it's a great opportunity to just Google yourself, see what the AI Overviews say, and start creating content around that and owning that SERP. Google's starving for content for those AI Overviews. If you provide it, they'll probably use it.

David Mihm (18:07)
Right. I want to highlight one thing while we're on the subject of AI and brand — I'm going to get back to my brand soapbox, Mike. One of the things we see in our user research in personal-injury law — we presented this data in September, I think of last year, at the Lunch Hour Legal Marketing summit in Vegas, and we've since done a follow-up study with a very similar pattern in PI law — I think this is true basically in any local business category. We don't necessarily see AI Overviews reducing clicks to local business websites. What we see is consumers using them to orient themselves in the category, because they're not domain experts in law. If you get an AI Overview for a furnace-repair query, you're not going to go try to do that yourself. They're using them to figure out: am I looking for the right type of professional? What criteria should I use to evaluate these professionals? And they're not stopping the journey — that doesn't satisfy their journey. Their journey is that they need someone to solve this problem, because they're not a professional themselves. So we actually see AI Overviews driving more searches that do lead to clicks to local business websites. But a lot of times — and this is where the brand thing comes in — consumers will pay attention to the firms getting recommended in the AI Overview. Here are six or eight firms in the greater Phoenix area that all have good reviews and so on, and then they'll do follow-up brand searches for each of those. One of the things that's frustrating for a lot of SEOs is that it's very hard to track at scale how you're doing in AI Overviews, because you never know when they're going to show up — the queries are always different. But one of the proxies we've recommended law firms look at is your volume of brand searches over time. If you see that going up, chances are good you're getting cited in an AI Overview and consumers are looking for your brand as a follow-up. If your brand searches are going down, you should probably be concerned that you're not showing up in AI Overviews the way your competition may be. So even if AI Overviews are in fact reducing clicks and reducing click data in the local business world, I think Google is still backfilling that click data with brand searches — which was Mike's earlier point. I don't know if you have any thoughts on that.

Cyrus Shepard (20:36)
No, I think that's great advice — branded search volume. And Google did us a huge favor a couple months ago by adding the branded-query filter in Google Search Console. I had to recreate those for years, and a lot of people didn't do them, but now you can do it with the click of a button. I always say brand search is one of the mandatory things all SEOs and marketing teams should be looking at, because it's a gauge of strength [ear-check]. It's interesting — I've had legal clients, and the old model used to be you'd create the content you wanted to rank for, try to get as much traffic as possible, and then funnel them. I'd work with lawyers who put big pop-ups and conversion banners — "call!" — and I'd say, "Too aggressive, we've got to tone this down."

David Mihm (21:17)
Yep, believe me, those are all over the place.

Cyrus Shepard (21:23)
But now I've started to change my thinking, because the information can live in the AI Overviews. I want to really capture that secondary intent — ultimately get the conversion, the call. So now I'm thinking, let's help with that task completion. What are the next steps after that? I want to understand pricing, understand reviews, understand how do I book, when will you call me back — all those things. Emphasizing that content a bit more on the website you own is maybe a little more important now, and we can push those things to help complete those tasks.

Mike Blumenthal (22:00)
How many years have you been in business? Great photography — who am I going to be working with? Those sorts of questions.

Cyrus Shepard (22:10)
Yeah, absolutely. But I think it's still worth creating that top-of-funnel content, because you're still going to get cited. People are going to have questions. You're just not going to see it.

Mike Blumenthal (22:19)
Although now you can just do it with a listicle. Why bother with all that content? "You're the best lawyer in Phoenix — and here are the other nine guys." No, no, they're just not number one.

Cyrus Shepard (22:23)
"Here are the other guys, and they're really bad. They're really, really bad."

David Mihm (22:33)
That's right.

David Mihm (22:36)
So in our final few minutes here, Cyrus — you also had some very strong tactical recommendations, structurally similar to what we provide in our personal-injury law research, which we're not going to reveal here. But you provided recommendations in terms of earning the click, for whatever percentage of traffic AI Overviews are not eating. How can you make your organic results earn more clicks and more longest clicks once they're there? Walk people through your thinking on how you'd structure title tags, descriptions, snippets, everything.

Cyrus Shepard (23:15)
Yeah, absolutely. The thing I want to avoid people doing is starting with title-tag optimization and creating a really clickable title tag — "best lawyer in Phoenix, rated blah blah" — and then the user clicks through and thinks, "This really isn't what I was looking for." They click back, and so you got the click but created a bad click. You shot yourself in the foot. You can see this in analytics when you do title-tag testing: you roll out a bunch of new titles, your click-through rate goes up, but within a couple of weeks your traffic is lower, because you're fooling people. You're making a promise with your title, and the result has to deliver that promise. So you want an attractive SERP: your review stars if you have them, an attractive icon — I think people sleep on their icons all the time; high-contrast site icons help you stand out — and a title that shows two things: exactly what people are searching for, the keywords they used, and then the promise. How are you going to deliver? What is the search intent? "Phoenix DUI lawyer — get your questions answered now," or "we'll help you out." There's some secondary intent you're going to help them with. That's earning the click. Then, getting the click, you want to engage people. The simplest way to think about it is: what else is on the page that I can engage people with? Answer their question quickly — if they're searching for something, answer it at the top of the page; don't bury it at the bottom. And for people who stick around, draw them into your content: give them interesting things to click on, an interesting navigation that answers their question, and use the content below the answer to answer related questions. I like testing navigations — I think they're secret weapons, because a lot of people look at navigations. Testing different navigation links and language can have a significant impact on time on site and engagement. One of my favorite reports — I'm not a huge fan of GA4, but I'm starting to like it — even now, years after Google launched GA4, one of their major reports was engagement. I filter for organic traffic and see how long people spend on site, how many engaged sessions per user. It's honestly become my favorite report, because there seems to be a correlation between increasing engagement and downstream rankings, I think because of these click signals. And then finally, it's ending the journey — satisfying user intent. This is a tough one; we can't measure it directly. We have proxies: conversions, brand search. It's really hard to measure and often involves a lot of competitive analysis. Is anybody else offering this? Am I actually the best for this query? You've got to put yourself in the user's shoes and go back to the SERP and see if someone else might be answering that question better and providing a better experience. It's often tough love when I talk to clients — showing them different screenshots, like, "I'd rather shop here than on your site, because they provide a lot, and we have to do better." Earning every part of that journey, helping people complete their task, measuring it, and looking at the things that matter — that's what we do.

Mike Blumenthal (26:31)
So in law, one of the things we see a lot is software that, as soon as the user gets to the site, asks, "Do you need a lawyer?" and brings a pop-up, and then, "Let me show you a video of the lawyer," and brings another pop-up, and it just keeps going. We see that all the time — trying to push this issue of last longest click by brute strength rather than answering the questions that are either explicit or implicit in the user's journey.

Cyrus Shepard (27:07)
Yeah. It's okay if people come to your site and leave, if they got what they were looking for in that moment. They might just need a quick question answered. So there's that push and pull. You have to satisfy the user intent within the first few seconds of them landing on the page. If you have a video pop-up that's really annoying, that's probably not going to help you for those short queries. Delay the video pop-up, if you have to, by 30 or 45 seconds; make it easily dismissible. Pull them in — but you have to satisfy that user intent very, very quickly, or you're probably going to send a bad signal.

David Mihm (27:41)
So Cyrus, thanks so much for your time today. You mentioned clients earlier — what kind of clients are you looking for? Who should be contacting Zyppy?

Cyrus Shepard (27:51)
I don't do a ton of client work these days. If someone writes me and it's an interesting problem with an interesting company, I'll probably have a conversation with you — I do appreciate that, David.

David Mihm (28:02)
All right, fair enough. For those who aren't looking to hire Cyrus, where can they stay in touch with you? What's your favorite social platform?

Cyrus Shepard (28:10)
My new platform is Substack, which I'm having a good time on. I've gotten back into content publishing and enjoy interacting there, doing research and publishing my results. I'm so glad it led to this conversation today.

David Mihm (28:24)
Awesome. What's the address of your Substack? Where should people go?

Cyrus Shepard (28:27)
It's Zyppy Signal — signal.zyppy.com.

David Mihm (28:30)
Signal.zyppy.com — and we'll link to it in the show notes. Thanks so much again for joining us. I know it's early out there on the West Coast, so I appreciate you being caffeinated and joining us for such a lively conversation.

Cyrus Shepard (28:48)
Mike, David, thank you very much.

Mike Blumenthal (28:50)
Thanks, Cyrus.

Corrections

What was fixed (transcription):

• "David Mim" → David Mihm; "Cyrus Shepherd" → Cyrus Shepard (name normalization).
• "AJ Cohn" → AJ Kohn (SEO consultant, Blind Five Year Old).
• "SubStack" → Substack.
• "brand does a big plays a big role" → "brand plays a big role" (garbled duplication).
• "Do remember that?" → "Do you remember that?" (dropped pronoun).
• "the great Late Bill Slawski" → "the great late Bill Slawski" (stray capital).
• Removed stutters, false starts, and doublings ("if we if we," "apparently apparently," "it was it was," "you you," etc.); folded backchannel ("yeah," "right," "you," "mm-hmm") into the active speaker's turn for continuous reading.
• Minor: "one past the nation's birth years" read as "one past the nation's birthday" (2026 = US 250th); "before we start new results" read as "new searches" for sense. Flagged as low-confidence — revert if the audio differs.

What was intentionally left as said (speaker-side — for show-notes footnotes):

• "FastSearch and FastRank and [RankEmbed]BERT" — Cyrus's naming of Google's long-tail systems. "FastRank" and "BedBERT" are almost certainly mis-heard; Google's disclosed systems are FastSearch and a RankEmbedBERT-type model. Marked [ear-check] in the text — verify against audio before quoting, but not silently rewritten.
• "a gauge of strength" (20:36) — raw transcript read "a goal of strength," which is nonsensical; rendered as the most likely intended phrase and flagged [ear-check].
• "the antitrust trial... two years ago" and "MozCon 2013" — dates as the speakers recalled them; left as spoken.