One in Five ChatGPT Clicks Go to Google: What the New Study Means for SEO

If you have been watching “AI search” narratives lately, it probably feels like we are supposed to accept a simple swap.

People will ask ChatGPT. Then ChatGPT will replace Google. And all of our organic traffic plans will need a full rewrite.

The Semrush data that Search Engine Land just covered does not really support that clean replacement story. It shows something messier. More realistic, honestly.

Here’s the headline that matters.

A meaningful chunk of ChatGPT referral clicks go straight to Google. About 21.6% per the Semrush study, as reported by Search Engine Land. You can read that coverage here if you want the exact framing: Search Engine Land’s writeup of the Semrush study. Semrush also published their own breakdown here: Semrush: ChatGPT search insights.

So yes, ChatGPT can refer traffic outward. But a lot of that “outward” still funnels into the same place it always has.

Google.

And there is a second detail that matters just as much. ChatGPT appears to be triggering live web search on a smaller share of queries than it did in late 2024 (again, per the Semrush study coverage). Which hints at a behavioral split.

Some users want answers in the interface and never leave. Some users want confirmation and go to the open web. And when they do, a lot of them still choose the familiar door.

That has consequences for how we plan content, attribution, and what “visibility” even means now.

What the study actually suggests (without the hype layer)

Let’s simplify the implications:

1. AI search is not replacing Google cleanly. It is more like another layer on top of the same demand engine.
2. Referral traffic from AI tools is concentrated. A small set of destinations gets a disproportionate share of the clicks.
3. Citations are not the same as visits. You can “show up” inside AI answers and still see basically no sessions.

That is the core of it. Everything else is tactics.

Referral concentration: why AI traffic is not evenly distributed

This is the part a lot of teams miss when they glance at charts.

Even if AI assistants send referral traffic, it does not mean they send you referral traffic.

In most markets, AI referrals behave like winner take most. A few destinations absorb the bulk of the outbound clicks. The Semrush reporting showing Google as a top destination is one example of this concentration effect, but you see it elsewhere too. Wikipedia, YouTube, big publishers, big forums, major retailers. Entities that are already “default trusted endpoints” on the web.

There are a couple reasons for that:

• User intent is often navigational downstream. People ask ChatGPT something informational, then decide they want to “go do the thing” somewhere else. Google is the easiest generic next step.
• AI interfaces reduce exploratory clicking. When the answer is summarized, fewer users feel the need to open ten tabs. That means fewer total clicks to distribute.
• The model’s safest links skew mainstream. If the assistant is unsure, it tends to cite sources with broad authority signals. Big brands, established domains, well known documentation.

So if your AI traffic plan is “get cited, watch sessions roll in”, you are going to be disappointed.

This is also why it is worth internalizing a slightly uncomfortable truth: modern SERPs and AI layers are already stealing clicks that used to be yours even when you ranked. If you have not read it, this is a good primer on that reality: how modern SERPs steal clicks and how to optimize for clicks, not just ranks.

AI citations do not equal traffic (and what they do equal)

A citation inside an AI answer is closer to an impression than a click.

Sometimes it leads to a visit. Often it does not. And in a lot of product categories, the citation is doing a different job:

• Validating the assistant’s answer (credibility)
• Offering a “proof link” for skeptical users
• Acting as an escape hatch when the answer is incomplete

That means citations are still valuable, just not in a direct response equals click way.

They can drive:

• Brand recall (people search you later, usually on Google)
• Assisted conversions (someone sees you cited, then trusts you when your ad or organic result shows up later)
• Sales enablement (prospects quote the AI answer internally, and your brand is attached to it)

This is why you should treat AI visibility as part of the top and mid funnel measurement stack, not purely last click acquisition.

If you want a deeper playbook on how to structure content for being referenced in AI answers, this is a practical one: a GEO playbook for getting cited in AI answers.

Why Google still captures downstream demand

The study detail that one in five ChatGPT referral clicks go to Google is not random. It is structural.

Google still owns a few things that assistants do not fully replace:

1. Query refinement is easier in a search engine

When users need to compare options, scan varied sources, or find a specific page, they often want a results page. Not a single synthesized response. Even if they start in ChatGPT, they end up in Google for the messy middle.

2. Transactional intent lives on the open web

Shopping, demos, pricing comparisons, “best X for Y”, local services. These often end in search because search is still the best map for commercial exploration.

3. Trust is distributed, not centralized

Users might enjoy an AI summary, but when stakes rise they go verify. Verification behavior usually means Google, brand search, reading reviews, checking documentation.

Also, Google is now adding its own AI layers, which makes the ecosystem even more circular. If you are tracking this shift, it is worth reading: Google AI Mode citing behavior and what it means for SEO.

So in practice, many journeys look like this:

ChatGPT for a fast explanation → Google to validate and compare → a few brand sites for final evaluation → conversion.

And that puts classic SEO right back in the center. Just with new constraints.

What operators should change in response (SEO, content, growth)

Stop thinking channel replacement. Start thinking journey ownership.

Instead of “AI search versus Google”, plan around the full path:

• Do we appear as a trusted entity in AI answers?
• Do we win the follow up search when the user goes to Google?
• Do we provide the best landing page when they finally click?

If any of those three are missing, the funnel breaks.

Re prioritize content that captures follow up intent

If AI tools answer the basic question, then the clicks that remain will skew toward:

• comparisons
• templates
• calculators
• detailed workflows
• case studies
• “how to choose” pages
• integration docs and setup guides
• pricing and alternatives pages

You still need the definitional content. But you need it to lead somewhere that AI cannot compress into three sentences.

A good internal framework to keep your on page work consistent is here: SEO content writing framework.

Make your pages easier to cite and easier to click

The weird thing now is that you are optimizing for two readers at once.

• the human skimming a SERP or an AI answer
• the machine extracting a snippet and deciding if you are citation worthy

That changes formatting and information design.

Use:

• short definitions near the top
• direct answers, then deeper expansion
• clear headings that match likely questions
• tables for comparisons
• explicit numbers, constraints, steps
• named entities (brands, standards, tools, locations) where relevant

And yes, keep an eye on how Google rewrites your titles because that affects clickthrough a lot more than people admit. Relevant read: Google AI headline rewrites and the SEO impact.

Building content that earns AI mentions and direct clicks

Here’s what tends to work right now, across SaaS and publishing.

1. Write “quotable blocks”

AI systems love clean, self contained chunks:

• a definition
• a step by step sequence
• a checklist
• a warning box
• a mini framework with named steps

Not fluff. Not vibes. Actual usable units.

2. Add proof, not just claims

If you say “this improves conversion”, show the setup, show the result, show the constraints. If you say “Google prefers X”, cite the doc, show the test, show the SERP pattern.

This is also where a lot of AI content fails. It can generate a smooth page, but with no real-world grounding. And Google is clearly getting better at sniffing out low value scaled content, even if the exact signals are debated. Worth reading if you publish at scale: Google AI content detection signals and machine scaled content vs programmatic SEO.

3. Build the follow up page, not just the answer page

If you publish “What is X”, also publish:

• “X vs Y”
• “X pricing and cost breakdown”
• “Best X tools for Z”
• “How to implement X in 30 minutes”
• “X checklist”
• “X templates”
• “Common mistakes with X”

Because that is where the clicks go when people leave the AI interface.

4. Use internal linking like you mean it

If clicks become rarer, every click has to travel further.

This is not about SEO theory. It is conversion reality.

You might like this specific guide on not overdoing it while still building strong pathways: internal links per page and the SEO sweet spot.

5. Don’t ignore performance basics

If users click less often, they bounce faster when the page is slow or cluttered.

Keep Core Web Vitals and page speed in check. Here are practical fixes: page speed SEO fixes that improve rankings.

Practical checklist: what to do this week

Use this as a working list for your team. Not theory. Just execution.

Measure and diagnose

• In analytics, segment referral traffic from ChatGPT and other AI surfaces (where identifiable). Track landing pages, not just totals.
• Identify which pages get cited (brand mentions in AI answers) but do not get visits. That gap is the story.
• Map the top 20 queries where you show up in AI answers. Then map the follow up queries people likely search on Google.

Fix your content for citation + click

• Add a direct answer block near the top of key pages (2 to 5 sentences).
• Add a “how to” section with numbered steps and screenshots or real examples.
• Add a comparison table where relevant (alternatives, features, pricing, use cases).
• Improve titles and intros so they promise a payoff the AI answer does not fully deliver.
• Add internal links to next step pages (setup, templates, tool pages, pricing, case studies).

Build the downstream demand capture

• Create at least 3 follow up intent pages per core topic (vs, alternatives, implementation, checklist).
• Refresh pages that already rank but are losing clicks due to SERP features. Use this guide as a reference: how to fight back when Google AI summaries kill traffic.
• Standardize on page QA before publishing using a checklist like this: SEO content optimization checklist.

Operationalize it (so it scales)

• Build a repeatable workflow for research, writing, optimization, and publishing. If you need a structure, this is a solid one: AI SEO workflow: on page and off page steps.
• If you are launching or rebuilding, follow a disciplined first month plan so you are not guessing: new website SEO first 30 days strategy.

So what does “one in five clicks go to Google” really mean?

It means AI assistants are not a new web where Google disappears.

They are more like a new front door that often leads right back to the old mall.

For SEOs, publishers, and SaaS teams, the move is not to abandon Google SEO for “AI SEO” or chase citations as the only KPI. It is to build content that can be extracted and cited, yes, but also content that wins the next step when the user goes searching anyway.

Because they do. The Semrush data is basically proof.

A soft CTA (because you will want to track this properly)

If you are serious about competing in this mixed reality, you need to monitor both sides: rankings and clicks in Google, plus visibility and mentions across AI answers.

That is the direction we are building toward at SEO Software. It is an AI powered platform for researching, writing, optimizing, and publishing rank ready content at scale. And it is the kind of setup that makes it easier to keep pace when the SERP changes week to week.

If you want a starting point, explore the platform at seo.software and use it to tighten your workflow, ship more pages that deserve to rank, and keep an eye on how AI surfaces are affecting your real traffic.