Zero-click search: why your clicks are falling and how to tell if your brand is losing ground in AI

Between August 23 and September 19, 2026, Google showed searchbrand.ai about 62,000 times and we got roughly 239 clicks. Around 22,000 of those impressions came from long, question-shaped queries, the kind AI Mode's fan-out generates, and not one of them produced a click. One of them, in Spanish, was the question behind this article: "how do I know if my brand is losing ground to competitors in ChatGPT's answers?" Google showed it 1,220 times, at position 9.5. Zero clicks.

That is zero-click search seen from the inside. This article unpacks it using SearchBrand's Micro-Training #4, recorded on November 26, 2025 with Andrés Kloster (founder and CRO of the Argentine agency Eleven), and gives you a tool to diagnose your own site from a Search Console CSV. The session is in Spanish; quotes below are our translation.

What zero-click search is

A zero-click search happens when the search engine or AI assistant resolves the query on its own screen and the user never visits a website. Andrés opened the session with a Google Search Console screenshot (6:30): impressions flat or rising, clicks falling, CTR shrinking in between.

His example: search "what to see in Santiago, Chile in two days" and the AI Overview, built with Gemini, lays out day one and day two for you. The links sit off to the side. For many people, he said, that paragraph is enough. The same happens with comparisons like "iPhone 14 or iPhone 15".

Google has been doing this for more than a decade. Ten years ago it already answered "how tall is Justin Bieber" without sending you to a celebrity site, and it showed Maps for places, Shopping for products and a 10-day forecast for "weather". Andrés searched for a glasses emoji and Google displayed it directly; he used to end up on Emojipedia every time (9:04). AI Overviews stretch that logic to queries that used to need three or four sites.

The crocodile effect

SEOs call that chart the crocodile effect: the two lines drift apart like an opening jaw. The "teeth" are the weekly peaks, since almost no site gets the same traffic on a Monday as on a Sunday.

"So far I haven't found a single website that doesn't have this shape in its clicks." — Andrés Kloster, Founder & CRO at Eleven 12:07

He was talking about the teeth, the weekly seasonality. About the jaw itself, he said the pattern shows up across the sites he has reviewed regardless of niche, country or size. That is an agency's field observation, not a study. Further down you'll find the cases where falling clicks have nothing to do with AI.

Refreshing content won't fix it

Germán asked in the chat how much of the drop comes from sites not updating their content. Very little, Andrés answered: sites that publish and refresh constantly show the same pattern, because Google is "putting itself in the middle between users and the destination site" (15:01).

"The people who want to buy mattresses are the same as before." — Andrés Kloster 15:01

Demand is still there. The exploratory part of the purchase (learning, comparing, shortlisting) now happens inside Google or ChatGPT, and fewer, more decided people reach your site.

Check your own crocodile

Export the Performance report from Search Console with 16 months of data, open the Dates tab and upload that file here. The detector groups days into Monday-to-Sunday weeks to flatten the teeth, compares the average of the first 8 weeks with the last 8, and tells you whether the jaw opened, from which week, and how many clicks you're missing versus your starting CTR. Below it sits the four-layer pyramid so you know where to start. The file is processed in your browser and never uploaded.

How to read the result:

Jaw index	What it means
Above 1.3	Wide-open crocodile: impressions are growing much faster than clicks
1.1 to 1.3	Early crocodile: split your queries and check which ones are already answered on the SERP
0.9 to 1.1	No effect
Impressions and clicks both down more than 20%	Not zero-click: check indexing and seasonality

If your index is above 1.3, the next step is to see which of your pages AI already treats as a source. Our AEO and GEO audit guide walks through it. And if you want to know what AI answers say about you where you already appear without a click, you can track them in ChatGPT, Gemini, Perplexity and Claude with 10 free credits, no card required.

The September 2025 change is not a recovery

In mid-September 2025 Google stopped supporting the &num=100 parameter that many rank trackers used to pull 100 results at once. Plenty of properties saw impressions drop overnight while clicks held steady, so CTR "improved" in a day. The webinar was recorded two months later and doesn't mention it.

The detector looks for a weekly impression drop of 30% or more in weeks starting between September 8 and 21, 2025, with stable clicks. If it finds one, the index is computed only from that week on and you get a warning. Without that cut, a site with a real crocodile can look healthy.

Case study: our own Search Console, step by step

We used searchbrand.ai data from August 23 to September 19, 2026. It's our property, sc-domain:searchbrand.ai, with numbers exactly as the dashboard shows them.

1. We exported Dates. With 28 days the detector drops two partial weeks (Sunday, August 23, and the week ending Saturday, September 19), which leaves 3 full weeks. It says there isn't enough data to compute the jaw. Correct. With fewer than 8 weeks you're reading noise, which is why the tool asks for 16 months.

2. We looked at the total. About 62,000 impressions and 239 clicks give a CTR of 0.39%. That's low for a first-year B2B site, and the average hides two different sites inside one.

3. We split queries into three groups. Brand (searchbrand and variants), non-brand, and prompt-shaped: seven words or more, or starting with what, how, which or why.

Group	Impressions	Clicks
Prompt-shaped (AI Mode fan-out)	~22,000	0
Non-brand	—	17 (average position 22)
Everything else, mostly brand	—	~222

Prompt-shaped queries were a third of all impressions that month and produced zero clicks. Of 239 clicks, 17 came from searches that didn't include our name. Everyone else searched for us by brand.

4. We read the questions one by one. The "losing ground to competitors in ChatGPT" question had 1,220 impressions at position 9.5. Google treats us as an eligible source for it, but AI Overviews or AI Mode answers it and the click never comes.

5. We changed the dashboard. We stopped treating fan-out impressions as a goal. We still track them, on a separate panel, as a sign that Google sees us as a source. The metric that counts is now non-brand clicks over 28-day windows.

6. We ran the pyramid. Content came out weakest. "answer engine optimization" showed us 7,001 times at position 27, and we don't have a page that answers that query head-on. This article is part of the fix.

7. We checked those questions in the assistants. If Google shows us for a question, what matters is whether the assistants name us when they answer it, and with which attributes. Search Console can't tell you that.

8. We picked the channel. We sell to B2B marketing teams in Chile and Latin America, and our bet is that they compare tools with AI assistants before booking a demo. So the budget goes to pages that answer those questions and to being present in the sources assistants cite. We'll revisit it against conversion data by channel.

How AI decides which brands to show: Andrés's pyramid

Andrés showed an Ahrefs chart plotting many sites' organic traffic against how often ChatGPT mentions them: sites that rank well on Google also get mentioned more in ChatGPT. Wikipedia and media outlets such as as.com sit above the line, cited more than they rank, because in his view the models have a "conservative bias" toward sources that work as official references (21:05). His takeaway: SEO and AEO strategies are "first cousins", and the tactics are what change.

To organize those tactics he used a four-layer pyramid (24:06).

1. Technical: if Google can't read you, neither can ChatGPT

"Chatting with ChatGPT is like having an assistant that goes and googles for you." — Andrés Kloster 24:06

Ask ChatGPT for "the five best mattress stores in Buenos Aires" and it runs several searches and comes back with a ranked list. If your site can't be crawled and indexed, it isn't on that list. Andrés added that some AI crawlers can't render JavaScript-heavy sites. That held for several bots in 2025; in 2026, check it yourself by viewing the source of your key pages, which is exactly what the first pyramid question asks.

2. Content built for questions

People talk to ChatGPT in problems. His example: "my neck hurts, what can I do?" The assistant might say it's the mattress and cite a mattress store's article. If that article doesn't exist on your site, there's nothing to cite. Andrés argued that schema (FAQ, recipe, article) matters more for AEO than it did for SEO. That's his opinion; he didn't show data for it.

3. Authority

Models play it safe: when in doubt, they recommend the official source. E-E-A-T signals count, author pages with real people count, and they count most on what Google calls YMYL topics, like health and finance. A bad article about mate tea, Andrés said, is less serious than a bad one about medication.

4. Popularity

The top layer is other people talking about you. Andrés said ChatGPT has answered him by quoting scores from Clutch, Trustpilot, Capterra and other ranking sites. Reddit carries weight because it works as a forum. Instagram and Facebook, in his reading, don't count as authority, and neither does X.

Hussam Sufan, SearchBrand's co-founder, saw it differently in the same session (30:02): ChatGPT and Perplexity increasingly use social media as a source, which opens the door for false content to show up as fact. His defense is the reputation you've already built. Our position: work on Reddit and reviews before Instagram, but watch what's said about your brand on social, because it can end up cited.

The rule in the self-assessment comes from the pyramid itself: fix it from the bottom. If Technical scores under 60, the tool won't recommend working on Popularity even when that's your lowest layer.

When AI mentions you badly

Google hands you results. An assistant hands you results with an opinion: "the best one is this tool, and then there's this other one, which is quite a bit more expensive." Being in the second half of that sentence is bad news.

"If you're being mentioned negatively by someone people trust as much as ChatGPT, it's even more harmful; you'd rather not be mentioned at all." — Andrés Kloster 33:07

That's why counting mentions falls short. You need to log the attributes the assistant gives you (expensive, slow, reliable, built for small businesses) and the tone. Andrés compared the difficulty to measuring billboards or TV.

Why ChatGPT answers you differently

Nicolás asked in the chat whether ChatGPT is biased when he asks about his own company. Always, Andrés said, by context: tell it "I'm the founder of Eleven" and it's more likely to recommend Eleven. Hussam suggested a test: type "what do you know about me?" and look at everything it has stored in memory, location included.

So your own account is a poor instrument for measuring your brand. SearchBrand queries the models through their APIs, with no user memory. That doesn't produce "the true answer", since the consumer app may run a different model or search setup, but it does give you controlled conditions you can compare month to month.

How to tell if you're losing ground to competitors

This is the question Google showed us 1,220 times. The short answer: with two measurements, one in Search Console and one in the assistants.

In Search Console, the detector above tells you whether clicks are falling faster than impressions. In the assistants, build a fixed list of 20 to 50 questions your customer would ask, run it monthly in ChatGPT, Gemini, Perplexity and Claude under the same conditions, and log four things per question: whether you appear, in what position, next to which competitors, and with which attributes. You're losing ground when a competitor climbs those lists while you slide, even if your clicks haven't moved. If you're still picking a tracker, here's our comparison of the best AEO tools for 2026.

First, ask where your users are

Eight years ago, Andrés said, his first question to a client was which keywords their customers searched. Now he asks where they are (36:49). For most niches the answer is still Google. For expensive or trust-heavy purchases, AI assistants. In Argentine retail, many searches start on Mercado Libre, much like Amazon elsewhere. Others live on YouTube or in the app stores. One brand asked him for TikTok because "it's what people use now"; its customers were over 45 and weren't there.

Travel insurance case: fewer visits from AI, more sales per visit

Andrés shared data from a large travel and travel-insurance site Eleven works with (39:25):

Channel	Sessions	Conversion to sale
Paid Search (Google Ads)	6.5 million	0.4%
Google organic	1.3 million	0.8%
ChatGPT	30,000	1.5%

It's one site, with no published period, country or conversion definition. Use it as a prompt to look at your own numbers, not as a benchmark. To build the same table, tag AI referrals in GA4 first (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com).

Andrés's explanation: the person arriving from ChatGPT didn't search for "travel insurance". They typed something like "I'm traveling abroad with my family, what should I keep in mind?", and the assistant explained why they needed a policy. That nurturing used to be the brand's job. Hussam described it as a funnel that flattens: the assistant does the discovery and the comparison, and the visitor arrives ready to buy.

Andrés also told the story of buying a wine fridge that cost several hundred dollars. He had a long conversation with ChatGPT about cooling zones, whether he needed 12 or 24 bottles, power use and noise, then googled the model and brand. To that brand, the sale came from a branded search. It actually started in ChatGPT. If your branded searches are rising and you can't explain why, this is a hypothesis worth testing.

On timing, in November 2025 Andrés said fewer companies were doing AEO than SEO or paid, especially in Latin America:

"The best time was six months ago; the second best time is now. Don't wait." — Andrés Kloster 42:05

The great filter: what you lose when the click never comes

Andrés closed with what he called the great filter: Google and the assistants keep the decision on their side and send you only the visitor who's ready. That visit converts better, but you lose everyone else. People who weren't buying that day used to land on your site, and you could show them a popup, get them on a newsletter or retarget them. If they stay in ChatGPT, that's gone.

"We have to be smarter with the data points we have." — Andrés Kloster 48:03

In practice: tag AI referrals in GA4, add "how did you hear about us?" to your forms, and read branded search trends alongside AI mentions.

What we don't recommend

• Measuring success with impressions from prompt-shaped queries. Ours added up to 22,000 and brought 0 clicks.
• Comparing two days against two days. A Monday against a Saturday always looks like a drop; compare full weeks, or 28 days against 28.
• Blaming AI when impressions and clicks fall together. That points to indexing or seasonal demand.
• Reading the September 2025 CTR jump as a recovery.
• Measuring your visibility from your own ChatGPT account, which already knows who you are.
• Spending on reviews and listicles while your key pages are invisible without JavaScript.
• Adopting the travel case's 1.5% as your target.

If the detector flagged a crocodile, take the questions Google already shows you for without a click and check what the assistants answer. You can do it in SearchBrand with 10 free credits, no card required.

Sources

• SearchBrand Micro-Training #4, "Cómo posicionarte en las respuestas de IA | El Paradigma del Zero Click", with Andrés Kloster (Eleven) and Hussam Sufan (SearchBrand), November 26, 2025 (Spanish).
• Google Search Console for searchbrand.ai (sc-domain:searchbrand.ai), August 23 to September 19, 2026.