Your Traffic Is Shrinking. Your Sales Shouldn’t.

Twelve months of client data: impressions down 38%, position up nine places, CTR up 32%, clicks still fell. The fight moved from rankings onto the page.

I’ve wanted to write this one for months but didn’t have clean enough data. Now I do.

A client’s search data over twelve months: impressions down 38%, average desktop position up nine places, click-through rate up 32%, and absolute clicks still fell. Good SEO isn’t the problem. The market got smaller.

That doesn’t end SEO, but it changes the arithmetic: you now have to convert more of the fewer people who arrive.

The good news is that conversion work no longer needs an expensive tool stack. Claude Cowork, free Microsoft Clarity, the GA4 you already have, and your own eyes will do. I’ll walk through the setup step by step, and show you three blockers I found on an e-commerce signup form in two minutes.

Most zero-click conversations start from the same place: your SEO must be slipping. I want to show you a set of numbers that says the opposite, and still ends badly.

I opened up twelve months of search data for one of my clients. I am not naming the brand, but I can tell you the sector: an online video editing tool, a global software company. Millions of organic clicks a month, a well-worked site.

My first reaction to the numbers was confusion. Because the table does not say “SEO is going badly.” It says the opposite. And traffic is falling anyway.

Grab a coffee. This issue covers both the table and what to do about it.

The thesis

Organic traffic is shrinking, but not for the reason you would assume. Over twelve months in a client’s data: impressions fell 38%, while average desktop position improved by nine places and click-through rate rose 32%. So the SEO is not wrong; the site is doing better work. And absolute clicks still fell 18.5%.

Here is why: AI Overviews and AI Mode are absorbing the informational long tail, leaving behind a smaller but more intentful pool of demand.

The practical consequence is a multiplication problem: to hold the same revenue, you have to grow conversion rate by 22.7%. The fight is no longer in the rankings. It is on the page.

The good news: conversion work today does not need an expensive tool stack. Claude Cowork, free Microsoft Clarity, the GA4 you already have, and your own phone are enough. My rule fits in one line: numbers, behavior, and eyes. If all three do not agree, you have not found it yet.

Look at the table first

Read that table twice.

More than a third of impressions evaporated. But click-through rate went up. And so did rankings: average desktop position moved from 24.8 to 15.7, mobile from 14.7 to 8.7. The site climbed roughly nine positions on desktop in a year.

This is not a picture of SEO going badly. This is a picture of SEO going well inside a shrinking market.

The real detail is in the device split

On desktop, more than half the impressions vanished. On mobile the loss is about a quarter of that.

My read: desktop is where AI Overviews take up the most screen real estate and where chatbot use is heaviest. And most of the impressions that disappeared were already sitting past position 20, informational queries that never produced clicks. That long tail got swallowed.

What is left is a smaller but cleaner pool of demand. That is why click-through rate rose: the site now appears on fewer queries, but higher up and in more intentful searches.

I am not claiming causation. This is one site, in one sector. Google shipped several updates in the same window, and the site’s own content work continued throughout; I have not separated the two. I am not saying “because of AI.” I am saying “here is the table.” But do go open your own Search Console and look at how impressions and clicks have come apart over the last twelve months. If the gap is widening, this issue is for you.

The market-level picture points the same way. In Search Engine Land’s analysis of US website traffic, chatgpt.com grew 48.38% year over year to 1.09 billion monthly visits, making it the 9th most visited site in the US, ahead of DuckDuckGo and Bing. On the same list Amazon fell 15.31%.

The math changed

Here is the uncomfortable part. A simple multiplication:

Visitors × Conversion rate × Average order value = Revenue

If visitors fall and you want to hold revenue, you have to grow one of the other two. Average order value is a pricing and product problem; it does not move in a week. That leaves conversion rate.

Let us put numbers on it. Say your conversion rate is 2%:

Now put the client above into it: clicks fell 18.5%. To hold revenue, they need to grow conversion rate by 22.7%. A 2% rate has to become 2.45%. That sounds small, and that is exactly why it matters: this is a gap you close with a few weeks of friction cleanup, not a year of SEO work.

When I first looked at this table I said “impossible.” Then I realized: doubling conversion rate is not impossible on most sites, because on most sites nobody has looked.

For years, nearly every budget went to bringing people to the site. Almost nobody watches what happens to those people once they arrive. When traffic was plentiful this did not show. Now it does.

My read: the table above already says it. AI is reducing search traffic but raising intent. Someone who comes through a chatbot is not a random click; they got their answer, put you on a shortlist, and came to check. Potential value per visitor went up. Losing that visitor to a broken form is far more expensive than it used to be.

The three-source rule

Conversion work used to be expensive: a heatmap subscription, testing software, an agency. Not anymore. But cheap tools created a new mistake: people look at one source and decide.

I work with three. Each one covers the others’ blind spot.

My rule: if a finding does not show up in all three sources, it is not a finding. It is a guess. I do not hand guesses to clients as reports.

Let me unpack the second layer, because most people do not have this installed. Session replay tools automatically catch two things:

• Dead click: someone taps, nothing happens. The page does not respond.
• Rage click: someone taps the same spot repeatedly and fast. That is frustration.

Microsoft Clarity gives you this for free with no session cap: session recordings, five heatmap types, and automatic dead-click and rage-click detection as standard. The obstacle to layer two is not budget. It is not having set it up.

The third layer costs nothing at all. Just your phone. And it is the one everybody skips.

From the field: two minutes on a signup form

I am not naming the brand. One of Turkey’s largest e-commerce sites, the market leader in books, hundreds of thousands of visitors a day. A brand I worked with years ago.

What I did was simple: I opened the signup form at phone width (375 pixels), filled in nothing, and pressed Sign Up. 13 September 2026, clean browser session.

Finding 1: the cookie banner lands on top of the button

The page loaded, the form appeared, the Sign Up button sat near the bottom of the screen. I pressed it.

At that exact moment a privacy notice slid up from the bottom and covered the lower half of the screen. I did not press the button. I pressed the banner.

In session replay data this shows up as a dead click. In your analytics report it shows up as nothing at all.

Finding 2: an empty form produces no error message

I dismissed the banner, went back to the form, left every field empty and pressed again.

Nothing happened. No red text, no “this field is required,” no marker pointing at what is missing. The page just sits there.

I saw that with my eyes, then measured it: not one of the seven fields carries the attribute that turns on the browser’s own validation. So even the free “please fill out this field” prompt the browser would give you is switched off. There is not a single visible error message on the page.

What does a user do? Press again. Then again. That is how rage clicks are born.

Finding 3: a verification checkbox appears and shoves the layout down

After that first press, a “verify you are human” checkbox appeared directly above the button.

It takes up space and pushes the button down. So while your thumb is travelling toward the button, the button moves. Your thumb lands on empty space.

Three points of friction, one page, two minutes. All three run at once and all three feed each other.

Note this: none of the three shows up in a GA4 report. There you get one line: “signup page conversion rate is low.” The team reading that line decides to shorten the form. The form gets shorter, the problem stays, and nobody understands why.

Limiting my own claim

I am not claiming causation, and neither should you.

What I have: three observations, one browser, one session, one observer. I do not have access to this site’s conversion data, so I do not know how many people these three things cost them. Cookie banner behavior varies with prior visits. The verification checkbox may not appear for everyone.

The honest sentence is this: I have three cheap, concrete hypotheses worth testing. If I owned the site, I would go in this order: open the session replay tool and measure dead clicks on that button, ship the three fixes, then look at the same number two weeks later. If it drops, the hypothesis holds.

A two-minute look replaced what would have been weeks of “form design” debate.

The setup: build your own conversion desk

I wrote this section as a lesson. No technical background needed; just go in order.

1. Open a workspace and write the brief. I use Claude’s Cowork side, because it reaches a folder and my connections directly; I am not re-uploading files and re-explaining every time. I drop a one-page note in the folder: which conversion are we improving, which number measures its quality, what date range, which pages, what changed on the site recently, what we do not know.

That last item matters most. A single line like “the cookie banner changed at the end of August” can reverse the interpretation of the whole audit.

2. Actually define the goal in GA4. Marking a key event is half the job. Also write down what that event means for the business and which number shows the quality behind it. A form submission is a signal, not a sale; a shorter form produces more submissions but sales may accept a smaller share of them.

A note for anyone working outside the US: clicks that go to WhatsApp and other regional messaging channels. In Turkey, Brazil, India, Indonesia and plenty of other markets, the most valuable contact happens there, and on most sites it is not measured at all. What you do not measure is not in the report, and what is not in the report does not get budget.

3. Install Microsoft Clarity. Free, no session cap, one snippet. Wait a week and let data accumulate. That is your second layer.

4. Connect everything read-only. Hook up GA4 and Clarity with read access only. The AI does not need to change goals, build audiences or touch settings. Least privilege, one property, no personal data, and you review every query yourself.

5. Write the rules once, not every time. Here are my standing rules, and I keep them in the folder: separate observation from hypothesis, put the source next to every finding, give a confidence level, list alternative explanations, and if the evidence is thin, say “insufficient evidence” without dressing it up.

6. Give narrow tasks. Do not say “run an audit”; you will get a generic answer. Say “for this date range, pull dead-click counts on the signup page broken out by device, and list the most-clicked text.” One question at a time.

7. Then close the laptop and go to the page. On your own phone. Fill the form in. Fill it in wrong. Leave it empty and press. Skip this step and what you have left is a convincing story.

What does AI do at this desk, and what does it not do? It does: sort the data, pull the breakdowns, put three sources side by side, draft the findings table, turn half a day into ten minutes. It does not: establish causation, verify that the tracking is correct, or know the constraints of the business. The diagnosis stays with you.

One thing to do this week

Fill in your most valuable form wrong.

Pick up your phone, open your site’s signup, contact or checkout form. Then try three things:

• Press submit without typing anything. Does an error appear, where does it appear, can you read it?
• Leave the page as it first loads and look at what sits on top of the submit button. Cookie banner, live chat bubble, promo strip. Is any of them touching the button?
• After that first press, does anything new appear on the page? If so, does it move the button?

If you get stuck on any of the three, you have probably been losing money for months and it is not visible in any report.

It takes half an hour. If you have an agency, ask them for it too, with a screen recording.

Tell me what you find

I am publishing the long version of this setup separately: folder structure, the full text of the rules file, the steps for connecting GA4 and Clarity read-only, and the narrow-task templates I use.

If you run the three-step test on your own form, reply and tell me what broke. I would like to know which of the three shows up most often across different markets. I will report back with what comes in.