In short: Five systems account for nearly all AI answer traffic worth optimising for: ChatGPT, Google AI Overviews and Gemini, Perplexity, Microsoft Copilot, and Claude.
They are not five versions of one product. They differ in scale, in whether they search the live web by default, in how they cite, and in who uses them. Scale alone is a poor guide to where your effort belongs: the largest audience is rarely the one deciding on your category.
Read The Usage Numbers Carefully
Before the comparison, a warning about the figures in it. There is no independent audit of AI assistant usage. Every number below was published by the company that benefits from it, and the metrics deliberately are not comparable with each other.
Weekly users, monthly users and queries are three different things. A company reporting weekly active users has chosen the metric that flatters it most, as has a company reporting monthly queries.
Comparing the 800 million weekly users OpenAI reports against the 780 million monthly queries Perplexity reports is comparing two unlike quantities, and most articles that do it never say so. Use these numbers for order of magnitude and nothing finer.
The Five That Matter
ChatGPT
The largest by a wide margin. OpenAI reported roughly 800 million weekly active users during 2025. It is where most people who have used an AI assistant have used one, and where a general question about your category is most likely to be asked.
It searches the web when it judges that a question needs current information, and answers from what it already knows when it does not. This makes it the platform where your presence in training data matters most, because a large share of answers about established topics never trigger a search at all.
Google AI Overviews and Gemini
Two surfaces from one company, and they behave differently enough to be treated separately.
AI Overviews sit on top of ordinary Google results, which is the important structural fact: they draw on pages that are already visible in conventional search. This is the strongest single argument against abandoning SEO for GEO. On this surface, your classic ranking work is your GEO work.
Gemini is the standalone assistant. It cites more selectively than its peers, which raises the bar: on a surface that names two or three sources, being the fourth best answer is indistinguishable from being absent.
Perplexity
Perplexity reported around 780 million queries in a single month during May 2025. Smaller than ChatGPT, and disproportionately important, because of what it is built to do.
It searches the live web on essentially every query and shows its working, citing far more sources per answer than any competitor. For anyone starting GEO work, this makes it the most responsive testing surface available: publish something genuinely better, and Perplexity is where you are likeliest to see it reflected first.
Its audience also skews toward researchers and professionals, which matters more than raw scale if you sell to them.
Microsoft Copilot
Distributed rather than sought out. Copilot reaches people through Windows, Edge, Bing and Microsoft 365 rather than because they chose it, which gives it an enterprise footprint the others do not have.
It is grounded in Bing's index. That has a direct practical consequence which is easy to miss: if your site is well indexed in Google and poorly indexed in Bing, you are invisible on this surface for reasons that have nothing to do with your content.
Bing Webmaster Tools is free, and checking it is a twenty minute job that occasionally finds a large problem.
Claude
Anthropic reported roughly 30 million monthly active users as of the second quarter of 2025. The smallest of the five by consumer reach, and skewed heavily toward technical, professional and enterprise use.
Its citation behaviour leans toward established journalism and, notably, toward older journalism rather than the most recent coverage. Earned media outlasts publishing cadence here in a way it does not elsewhere. If your buyers are engineers, analysts or technical decision makers, its size understates its relevance considerably.
Side By Side
| Platform | Searches the live web | Citation style | Where your effort goes |
|---|---|---|---|
| ChatGPT | When the question calls for it | Few sources, often reference and community | Third-party mentions and entity clarity |
| AI Overviews | Always, on top of Google results | Linked, drawn from ranking pages | Conventional SEO, largely unchanged |
| Gemini | Yes | Selective, few sources | Being the single best answer, not a good one |
| Perplexity | Always | Many sources, prominent, primary preferred | Original documents and fast iteration |
| Copilot | Yes, via Bing | Linked, Bing-grounded | Bing indexing before anything else |
| Claude | Yes | Established journalism, less freshness-driven | Earned coverage and technical depth |
What this table does not tell you: how much these platforms overlap in what they cite. The answer is: much less than you would expect, and it is the single most consequential fact in the field. How Each Platform Works covers the evidence and what it means for measurement.
Scale Is The Wrong Way To Choose
The obvious move is to rank these by user count and start at the top. It is usually wrong, for a reason worth stating directly: your buyers are not distributed the way the general population is.
A firm selling infrastructure to engineering teams has more at stake on Claude and Perplexity, with a combined reach a fraction of ChatGPT's, than on the largest platform. A consumer brand has the opposite problem.
A business whose buyers work in large organisations on managed Windows machines should look at Copilot before anything else, because that may be the only assistant those buyers can actually reach.
Optimising for all five at once is the same as optimising for none. Pick based on who is asking, then let scale break the tie.
Find Out Which Two Matter To You
An afternoon of your own data beats any market share chart:
- In your analytics, filter referrals for chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com and bing.com. Record sessions and conversion rate for each over the last six months.
- Take your five most commercially important buyer questions and ask each on all five platforms, in fresh sessions.
- Score each platform on two axes: how visible you are today, and how much traffic or conversion it already sends you.
- Pick the two with the largest gap between commercial relevance and current visibility. That gap, not the user count, is where your next quarter belongs.
- Check Bing Webmaster Tools regardless of the outcome. It costs twenty minutes and Copilot depends on it entirely.
Key Takeaways
- No independent audit of assistant usage exists. Every headline figure is company-reported, and the metrics are not comparable with each other.
- ChatGPT leads on scale, Perplexity on citation volume and responsiveness, Copilot on enterprise distribution, Claude on technical audiences.
- AI Overviews are built on conventional Google results, which is why traditional SEO remains the foundation rather than a legacy concern.
- Copilot is grounded in Bing. Poor Bing indexing makes you invisible there no matter how good your content is.
- Choose platforms by who your buyers are, not by user count. Two platforms done properly beats five done thinly.
Check yourself
Before you move on
Not scored, not recorded, and not part of the certificate. Both answers are settled by a sentence in this lesson, and the reasoning appears whichever option you pick.
- 01
Your site is well indexed in Google and poorly indexed in Bing. Which surface does that hurt most?
- 02
Why is comparing OpenAI's 800 million weekly users with Perplexity's 780 million monthly queries misleading?