Two thirds of searches end without a click. 39% of teams lost traffic but held conversions. The question is which traffic you lost.
Everyone shares the first half of the HubSpot story, and everyone stops at the same comfortable conclusion: the lost traffic didn’t matter anyway.
The people who actually lived it correct that in the very next sentence, and the correction is the useful part.
Plus new industry data showing 39% of teams lost traffic while holding or growing conversions, and my own measurement of AI Overview coverage in Turkey versus the US. There’s an 18-point gap, and it tells you how much time you have.
I got a lot of replies to last week’s issue. Most of them said some version of the same thing: the same is happening to us, impressions down, clicks down, and we don’t know what to do about it.
This week I found the best answer in a place I wasn’t expecting. HubSpot’s own podcast, from HubSpot’s own executives, while they were making fun of themselves.
Grab a coffee.
Losing organic traffic is no longer the exception. It is the industry average. But losing traffic and losing business are not the same thing, and teams that can’t tell the difference are putting budget in the wrong place.
HubSpot lost 140 million visits over a twelve-month period, shed roughly 80% of its organic traffic, and in the same quarter posted its highest revenue ever.
The reason is simple. The pages it lost were things like “most popular movie quotes” and “how to find the shrug emoji.” The pages it kept were CRM, automation and software.
New Search Engine Journal research says this isn’t one company’s luck: 39% of respondents saw traffic fall or stay flat while leads and conversions held or improved, and only 15% lost both.
So the right question isn’t “why is my traffic falling.” It’s “which of my traffic is falling.” And no dashboard hands you that answer. You have to sit down and split it yourself.
Amanda Natividad of SparkToro brought the 2026 zero-click study to HubSpot’s Marketing Against the Grain podcast (episode 454, 17 September 2026), drawing on SparkToro’s clickstream study. Here is what happens after a Google search now:
| What happens after a Google search | Share |
|---|---|
| Someone clicks a link | 32% |
| Nothing happens | 39% |
| The user runs another search | 29% |
The bottom two add up to 68%. Two out of three Google searches end without a click.
Now look closely at that 32%, because that’s where the real story is. Roughly 6% of it goes to a paid ad and roughly 27% goes to Google’s own properties or AI Mode. The remaining 66% reaches the open web. The rough figure given in the episode: for every 1,000 US Google searches, only about 232 clicks land on the open web.
And none of this is new. In 2016, 45% of Google searches ended without a click. Today it’s 68%.
Natividad’s framing is the part worth keeping: “People think AI disrupted search. That’s not actually true. If anything AI is accelerating that trend a lot, but this has been happening.”
I’d underline that twice. Blaming a ten-year trend on a two-year-old technology is the most common diagnostic error I see right now, and it leads straight to the wrong fix.
One more number, because it reframes where you should even be looking: around 80% of search still happens on traditional search engines, mostly Google. After that comes ecommerce sites and social platforms. AI tools account for roughly 3%. Which also means there’s about a one in five chance your customer is looking for your solution somewhere other than Google entirely.
I like this section because it’s rare. A large company’s CMO and VP of Marketing, on their own channel, talking through their own decline.
Kieran Flanagan, HubSpot’s SVP of Marketing, gives the number: “We lost 140 million visits in over a 12-month period of time.”
Natividad fills in the rest: roughly 80% of organic traffic gone, in Q3 and Q4 of 2025, and it went viral in SEO circles as a cautionary tale.
Then the line that makes it interesting: in that same quarter, Q4 2025, HubSpot reported all-time-high revenue.
Because the pages that lost traffic were “most popular movie quotes” and “how to find the shrug emoji.” And the queries they kept were CRM, automation, software. The business itself.
Natividad’s summary: “SEO used to be about this bigger game of let’s get a bunch of organic traffic and then we’ll grow from there. But now it’s become more you just need to get the right traffic.”
This is where the story usually gets cut when it circulates, and it lands on a comfortable conclusion: the low-quality traffic didn’t matter anyway. HubSpot’s own executive corrects that, and nobody quotes the correction.
Flanagan says two separate things.
The first is honest self-criticism. They really were ranking for things they shouldn’t have ranked for, because when you try to grow traffic at a scale no other B2B company has reached, you get over your skis. They lost traffic that didn’t convert, and that loss was earned.
The second is the one that matters. That low-intent traffic was still generating demand. It converted at around 0.5%, but the volume was so large (at one point they had more monthly traffic than TechCrunch or Entrepreneur.com) that even a tiny rate produced real pipeline. In his own words: a hole opened up that had to be filled.
My takeaway: when you tell this story to a client, don’t skip the second sentence. “The traffic we lost was junk anyway” is comforting and wrong. The accurate version is this: some of what you lost was genuinely junk, and some of it was cheap but real demand that now needs replacing.
The full deck for this issue, built in NotebookLM: open it here.
Kipp Bodnar, HubSpot’s CMO, takes the conversation somewhere that I think is the most durable idea in the episode.
For fifteen years you could monetise a transient audience. Someone runs a random search, finds you once, buys something, leaves. That era is closing. To grow now you have to build a loyal audience and accumulate trust, influence and credibility, and then you monetise that audience at a far higher rate.
Flanagan compresses it into one sentence: “There is now a fundamental difference between building an audience and building the right audience.”
Bodnar’s image is good too. You used to stand in the stream and pull a few fish out as they swam past. Now you have to build the pond.
For those of us working outside the US, here’s the practical translation: the era of putting monthly sessions on the cover of the client report is ending. What replaces it is harder to screenshot. How many people chose to hear from us this month, how many came back, how many asked us something.
Here’s where I can add something the US-based writers covering this story can’t, because almost everyone measuring AI search is measuring the US results page.
I ran the test on Turkey. I built 39 matched commercial query pairs across eight sectors: healthcare, finance, products and ecommerce, travel, legal, B2B software, education, home services. Each topic was queried in Turkish and in US English, using whichever phrasing is natural in that market. One question only: does AI Overview trigger?
Method. Semrush’s Analytics API, the phrase_these report, the Fk column, which returns every SERP feature a query triggers. Semrush encodes those features as numbers, and code 52 is AI Overview. I didn’t guess the mapping, I verified it against Semrush’s own developer documentation. Databases tr and us, desktop, data pulled 20 September 2026.
| Market | Queries with an AI Overview | Trigger rate |
|---|---|---|
| Turkey | 25 of 39 | 64.1% |
| United States | 32 of 39 | 82.1% |
An 18-point gap. Roughly one in three commercial queries I measured in Turkey still returns no AI summary at all.
The eight topics where Turkey lags the US are these: hair transplant, mortgage calculator, robot vacuum, cheap flights, things to do in Bali, online courses, boiler repair, roof insulation. Not one of them is an academic question. Every one is a query where money gets spent.
One more detail. In the Turkish sample, none of the six queries that returned a local pack also returned an AI Overview. In the US, three of the eight local-pack queries showed both side by side.
Three limits, stated plainly. First, six queries is a signal, not a finding. Don’t generalise from it and neither will I. Second, a translated query pair is not a perfectly matched intent pair. Third, and most important: this data comes from Semrush’s scraping layer, and Google has been blocking scrapers far more successfully since around 13 September. Nozzle reported roughly an 80% drop in the data it could collect, and Sistrix posted a formal notice that its collection is running at reduced rate. This is a single point in time. I won’t speak with certainty until I’ve repeated it.
But the direction is clear enough to plan around. The kind of traffic HubSpot lost hasn’t eroded as far in Turkey yet. That isn’t an exemption. It’s a delay. And a delay is time you’ve been given to prepare.
If you run search in a non-US market, run your own version and you’ll have a number nobody else in your market has. Send it to me and I’ll fold it into the October study.
The weakness of the HubSpot case is obvious: it’s one company, telling its own story. So let’s look at the industry.
Search Engine Journal published its State of Search 2027 research this week. 39% of respondents saw traffic fall or stay flat while leads and conversions held or improved. Only 15% lost both. SEJ’s own reading: traffic still measures reach, but it no longer measures business value by itself.
Three more numbers from the same study, and read them together because the picture is uncomfortable:
The side effect SEJ flags: as GEO climbs the investment agenda, content refreshes and technical SEO are moving down. Both of those were among the activities most often credited with producing strong results.
I want to put a warning here. This pattern is familiar and it doesn’t end well. If you’re shifting budget toward work you can’t measure while deprioritising work you can measure and know is working, then in twelve months you’ll have neither the proof nor the result. I say this as someone who does GEO for a living: fund GEO, but not out of the technical SEO and content refresh budget. These are not alternatives to each other.
One reassuring number to close the section: 81% expect SEO investment to increase or hold steady next year. Even among those reporting the most harm from AI search, 44% expect investment to increase and only 16% expect a cut. The industry isn’t retreating. It’s adapting.
Clara Soteras published a good round-up in SEJ that gets at exactly this. Three findings I keep coming back to.
1. AI summaries trigger by question type. In the Spanish media study she co-authored with MJ Cachón and University of Barcelona researchers Mari Vállez and Carlos Lopezosa (CUVICOM project), query length is a direct predictor. Four-word queries trigger an AI Overview 48.1% of the time, and three and four-word queries together account for nearly 70% of all AI-generated results. The classic journalism question words are close to a guarantee: “why” 92.3%, “what” 85.7%, “who” 68.4%.
2. AI summaries and breaking news are separate ecosystems. AI Overviews appear in 34.6% of evergreen searches and almost vanish for immediate current events at 1.1%. Don’t run those two layers with one strategy.
3. And my favourite: the blue ocean is still there. In 43.6% of the analysed SERPs there was neither an AI Overview nor a Top Stories module. That’s a substantial space where traditional organic ranking still works at full strength. My Turkey number above is the same story measured in a different market.
The eye-tracking work (Soteras with Diego Criado and MJ Cachón) turned up something I didn’t expect. Images, product listings and the AI Overview “Explore Further” button capture up to 100% of visual attention and produce 0% click-through. People scan them in milliseconds and move on. The actual decisions happen in two places: the AI Overview text itself at 86.4%, and traditional organic links, still the undisputed king of the results page at 95.5%.
Worth knowing alongside that: Google’s own AI Mode research says people now search across five verbs. To explore, decide, learn, create and do. AI Mode queries run about three times longer than a classic search, AI Mode has passed a billion monthly active users, and follow-up questions are rising around 40% month over month.
Run your own HubSpot split. It takes half an hour.
Then treat the two boxes differently:
Then open a third box, because almost nobody does: the pages you never lost and never noticed. That’s where the blue ocean sits.
HubSpot published a free resource alongside the episode: an AI visibility audit built as 15 prompts for Google AI Mode. I read the whole thing and it’s well constructed. Running all of it is a day’s work, but here are the two most diagnostic:
Competitive positioning map:
“Who are the main players in [your category]? For each one, tell me what they’re best at, who their ideal customer is, and what their reputation is in the market.”
Full buyer journey:
“Walk me through how a [job title] at a [company size] [industry] company would evaluate, select and implement a [category] solution. At each stage, what resources, companies or content would they most likely turn to?”
The second one maps your visibility across the entire buying journey in a single response. Where you appear, where you go quiet, and which stage a competitor already owns.
There’s also a useful distinction in HubSpot’s framing: being mentioned and being cited are not the same thing. A mention is AI naming your brand inside the answer. A citation is AI linking to your content as a source. They need different strategies and most teams track neither.
If you’ve been reading this newsletter a while, that should sound familiar. It’s the practical face of the memory versus citation split. Mentions are the memory game and they’re won off your site. Citations are the retrieval game and they’re won on it.
One caution, because this is a marketing page. HubSpot cites some striking figures on that same page: that over 2.5 billion people now use AI-powered search by default, that over 90% of those searches don’t result in a site visit, that ranking first in AI Mode gives you only a 17% to 36% chance of being cited, and that cited brands get 35% more organic clicks and 91% more paid clicks. None of those numbers carries a source.
The prompts are good and the framework is good, but pass those figures to your client as “HubSpot’s claim,” not as data. I apply my own rule to the sources I like too.
Cloudflare will now write your robots.txt for you, and by default. Slobodan Manic’s piece is the technical read of the week. Bot Preference Sync takes whatever bot policy you set in the Cloudflare dashboard, turns it into robots.txt lines, and prepends it to your file. The problem is that the policy has three categories (Search, Agent, Training) and cannot express a per-crawler position.
Manic’s own example is the sharp one: he allows GPTBot, Anthropic’s crawler and PerplexityBot, and blocks Bytespider and meta-externalagent. All five train models. The first three send him visibility in return. There is no setting that describes that. For new domains the defaults changed on 15 September: Training and Agent blocked on ad-bearing pages, Search left allowed.
Do this week: open your robots.txt, open Cloudflare’s AI bot policies, and compare them. If they contradict each other, some crawlers treat that as grounds to ignore you entirely.
Google has started paying publishers for AI usage. The AI contribution pilot is invitation-only and pays when your content contributes significantly while an answer is being generated in Gemini, AI Overviews or AI Mode. The detail that matters: links or fact checks delivered after the answer don’t qualify. Earnings appear in a new AI earnings widget in the Search Console dashboard, and the calculation method hasn’t been published. Google plans to run the test for twelve months before deciding on a wider rollout.
In the same round-up: Search Profiles, the personal profile feature that feeds your Knowledge Panel, has reached 10,000 and is rolling out to new markets. Directly relevant if you’re building GEO around a personal brand.
Three measurement problems landed in the same week, and that’s the real story. Google has been blocking scrapers and third-party tracking tools far more effectively since around 13 September (Nozzle reported an 80% data drop, Sistrix posted a formal status notice). Search Console’s crawl stats report is missing a day of data from 15 September, across all profiles. And rankings have been unusually volatile for two weeks running.
When all three overlap, every decline chart on your desk is suspect. Verify from two independent sources before you tell a client you lost something on a given date.
Last week I wrote that your traffic is falling but your sales don’t have to. This week we got the strongest possible evidence for it. A company loses 140 million visits and sets a revenue record in the same quarter.
But please don’t read this as a consolation story. HubSpot’s own executives say two things clearly: part of what they lost was a deserved loss, and part of it was real demand that they then had to replace.
The lesson I’ll keep is that one sentence: building an audience and building the right audience are now two different jobs. For fifteen years we could monetise a passing crowd. That period is ending. What replaces it is harder, slower and less measurable. People choosing to come back to you.
You don’t have to like it. I don’t like it every day. The upside is that most of your competition won’t do it, because it’s hard.
So, a question: do you know the 20 pages that lost the most traffic in the last twelve months, and how many of them actually made money?
Talking about traffic loss without those two numbers is like examining a patient without taking their temperature.
Subscribe to our newsletter for weekly SEO insights, join the discussion in our community, or engage with professionals on our Twitter group.
