LLM Optimisation
ChatGPT Is Already a Real Search Engine. Your Brand Just Can’t See It Yet
For a while the safe line has been: “ChatGPT is tiny. Maybe half a percent of Google. Interesting, but not a real channel yet.”
5 December 2025
8 min read
Peec just tore that up.
Their analysis suggests that once you look at how people actually use AI assistants, ChatGPT isn’t 0.6% of Google, it’s more like 4–12% of Google’s search volume.
That moves it out of “toy” territory and into “mid-sized search engine you can’t ignore.”
Why click data makes ChatGPT look small
Most market share charts compare referral traffic:
- How many clicks does a website get from Google?
- How many from ChatGPT?
On that basis, Google sends thousands of clicks for every couple of dozen from ChatGPT. You end up with neat numbers like “0.6% market share.”
The problem: ChatGPT isn’t designed to send clicks.
Google’s UI is built around blue links.
ChatGPT is built around answers. Most users never need to leave the interface.
So if you use “clicks to websites” as your proxy for “search volume,” you’re only counting the fraction of AI journeys where the user needed more than the answer they already got.
That’s useful for measuring traffic. It’s terrible for measuring how often people are asking AI questions that your brand could be in.
Two better ways to think about ChatGPT’s size
Peec basically re-asks the question: if clicks are the wrong lens, what’s a better one?
1. Look at usage, not referrals
OpenAI’s own research shows billions of prompts per day. A big chunk of those are “information seeking” – people asking questions, comparing options, looking for recommendations.
If you apply a reasonable share of “search-like” prompts to OpenAI’s daily volume, you land in the region of:
~600 million “search” queries per day in ChatGPT
vs ~14 billion searches per day on Google
That’s about 4–5% of Google’s volume, on usage alone.
2. Adjust clicks for very different CTRs
We know three things:
Websites get far more clicks from Google than ChatGPT.
A large share of Google searches still result in a click to another site.
Only a small share of AI answers do. Most journeys are zero-click by design.
If Google sends, say, 4,000 clicks at a ~40% CTR, that implies around 10,000 searches.
If ChatGPT sends 24 clicks at a ~2–5% CTR, that implies hundreds to over a thousand searches.
Once you do that correction, you land in the 4–12% of Google band.
Different data, different method, similar answer.
Why this matters more than the exact percentage
You don’t need to memorise the maths. The strategic takeaway is simple:
ChatGPT is operating at real search-engine scale.
A huge chunk of its value is zero-click influence inside the interface.
Your current analytics stack barely sees any of it.
If you only value channels by “clicks into GA4,” you’ll systematically underestimate AI – exactly because it is doing its job well: answering the consumer before they ever reach your site.
For some categories (finance, software, health, DTC), the effective impact is even bigger, because early adopters skew towards higher-value decision makers.
What this means for AI shelf-space
For LMO7, this is the punchline:
If ChatGPT is already at 4–12% of Google’s scale, then AI shelf-space is no longer experimental. It’s just under-measured.
That has a few implications:
“Share of model” belongs next to “share of search”
You need to know how often your brand appears (or doesn’t) when people ask AI about your category, jobs-to-be-done and competitors.
Category language is now a core asset
LLMs reason over concepts: “socks for standing on concrete all day,” “sunscreen for ultra-distance cyclists,” “gummies to feel calm without alcohol.”
If your content doesn’t clearly sit inside those concepts, you’re invisible.
AI, Amazon and D2C content must line up
Rufus, ChatGPT, Gemini and retail media are converging on the same product universe. Your PDPs, FAQs, schema and Amazon detail pages should all tell one clear, structured story about who you are for and what you’re best for.
Measurement needs to evolve
You won’t be able to justify AI work purely on last-click ROAS. You will need AI visibility metrics, experiments and brand-lift style tests.
How LMO7 is building around this
This is how we’re baking Peec-style thinking into our work with consumer brands:
Audit AI shelf-space
Use tools like Peec, Share of Model and our own LMO7 visibility analyser to see where you appear (and don’t) in ChatGPT, Claude, Gemini, Perplexity and Amazon Rufus.
Fix the category story
Tighten how you describe your products so they map to real prompts: use-cases, occasions, symptoms, benefits – in plain, human language.
Align content for humans, bots and models
Update Amazon listings, D2C pages and structured data so AI systems can reliably lift your brand into answers.
Add “AI share of model” to your dashboards
Treat it as a leading indicator for both organic and paid performance over the next 12–24 months.
Because if ChatGPT really is already a mid-sized search engine in disguise, the brands that win are the ones that start treating AI answers as seriously as search results – before everyone else catches up.
Their analysis suggests that once you look at how people actually use AI assistants, ChatGPT isn’t 0.6% of Google, it’s more like 4–12% of Google’s search volume.
That moves it out of “toy” territory and into “mid-sized search engine you can’t ignore.”
Why click data makes ChatGPT look small
Most market share charts compare referral traffic:
- How many clicks does a website get from Google?
- How many from ChatGPT?
On that basis, Google sends thousands of clicks for every couple of dozen from ChatGPT. You end up with neat numbers like “0.6% market share.”
The problem: ChatGPT isn’t designed to send clicks.
Google’s UI is built around blue links.
ChatGPT is built around answers. Most users never need to leave the interface.
So if you use “clicks to websites” as your proxy for “search volume,” you’re only counting the fraction of AI journeys where the user needed more than the answer they already got.
That’s useful for measuring traffic. It’s terrible for measuring how often people are asking AI questions that your brand could be in.
Two better ways to think about ChatGPT’s size
Peec basically re-asks the question: if clicks are the wrong lens, what’s a better one?
1. Look at usage, not referrals
OpenAI’s own research shows billions of prompts per day. A big chunk of those are “information seeking” – people asking questions, comparing options, looking for recommendations.
If you apply a reasonable share of “search-like” prompts to OpenAI’s daily volume, you land in the region of:
~600 million “search” queries per day in ChatGPT
vs ~14 billion searches per day on Google
That’s about 4–5% of Google’s volume, on usage alone.
2. Adjust clicks for very different CTRs
We know three things:
Websites get far more clicks from Google than ChatGPT.
A large share of Google searches still result in a click to another site.
Only a small share of AI answers do. Most journeys are zero-click by design.
If Google sends, say, 4,000 clicks at a ~40% CTR, that implies around 10,000 searches.
If ChatGPT sends 24 clicks at a ~2–5% CTR, that implies hundreds to over a thousand searches.
Once you do that correction, you land in the 4–12% of Google band.
Different data, different method, similar answer.
Why this matters more than the exact percentage
You don’t need to memorise the maths. The strategic takeaway is simple:
ChatGPT is operating at real search-engine scale.
A huge chunk of its value is zero-click influence inside the interface.
Your current analytics stack barely sees any of it.
If you only value channels by “clicks into GA4,” you’ll systematically underestimate AI – exactly because it is doing its job well: answering the consumer before they ever reach your site.
For some categories (finance, software, health, DTC), the effective impact is even bigger, because early adopters skew towards higher-value decision makers.
What this means for AI shelf-space
For LMO7, this is the punchline:
If ChatGPT is already at 4–12% of Google’s scale, then AI shelf-space is no longer experimental. It’s just under-measured.
That has a few implications:
“Share of model” belongs next to “share of search”
You need to know how often your brand appears (or doesn’t) when people ask AI about your category, jobs-to-be-done and competitors.
Category language is now a core asset
LLMs reason over concepts: “socks for standing on concrete all day,” “sunscreen for ultra-distance cyclists,” “gummies to feel calm without alcohol.”
If your content doesn’t clearly sit inside those concepts, you’re invisible.
AI, Amazon and D2C content must line up
Rufus, ChatGPT, Gemini and retail media are converging on the same product universe. Your PDPs, FAQs, schema and Amazon detail pages should all tell one clear, structured story about who you are for and what you’re best for.
Measurement needs to evolve
You won’t be able to justify AI work purely on last-click ROAS. You will need AI visibility metrics, experiments and brand-lift style tests.
How LMO7 is building around this
This is how we’re baking Peec-style thinking into our work with consumer brands:
Audit AI shelf-space
Use tools like Peec, Share of Model and our own LMO7 visibility analyser to see where you appear (and don’t) in ChatGPT, Claude, Gemini, Perplexity and Amazon Rufus.
Fix the category story
Tighten how you describe your products so they map to real prompts: use-cases, occasions, symptoms, benefits – in plain, human language.
Align content for humans, bots and models
Update Amazon listings, D2C pages and structured data so AI systems can reliably lift your brand into answers.
Add “AI share of model” to your dashboards
Treat it as a leading indicator for both organic and paid performance over the next 12–24 months.
Because if ChatGPT really is already a mid-sized search engine in disguise, the brands that win are the ones that start treating AI answers as seriously as search results – before everyone else catches up.