AI Search

8 min read8 September 2026Nathan Mzumara

Claude vs ChatGPT: What the Traffic Share Data Actually Shows

In eleven months, ChatGPT's share of global assistant traffic fell from 76.4% to 52.7%. That is a drop of 23.7 points, or 31.0% in relative terms. Over the same period, Claude went from 1.6% to 9.2% – a 5.8-fold rise, and the fastest of any platform tracked. Those two lines are the honest starting point for any Claude vs ChatGPT comparison.

Most Claude vs ChatGPT comparisons are written by someone who used both for a fortnight. That is a fine way to pick a tool for yourself. It is a poor way to make a decision for a company, because your fortnight is not a market.

So here is the other kind of answer: what the traffic data says, what it does not say, and what actually matters when you choose.

This is the single most misread number in AI coverage, so it is worth being clear.

The assistant category grew fast over this period. It grew fast enough that ChatGPT could give up 23.7 points of share and still gain users in absolute terms. Falling share and falling usage are different things.

What share does tell you is the shape of the market, and the shape has genuinely changed. It is no longer one product with everyone on it. Two platforms now hold 80.0% of tracked share between them, and the rest divides what is left. A year earlier, one platform held three quarters on its own.

So the old shortcut – "everyone uses ChatGPT, just go with that" – was defensible in mid-2025 and is not defensible now. Roughly half the category is somewhere else.

Claude's growth is a small number moving fast

A 5.8-fold rise sounds enormous, and in relative terms it is. In absolute terms Claude went from 1.6% to 9.2%. That is still the smallest base of the western frontier labs.

Both readings are true and you should hold both. Fastest growth in the category. Smallest starting point.

One thing that does strengthen the finding: brand search demand moved the same way over the same window. Two independent measurements agreeing is worth more than either on its own, and it is rare in this dataset.

What the data cannot tell you is why people chose either one. Share reflects availability, price, habit and what an employer already pays for, at least as much as it reflects how good the model is at your job.

The bigger issue: most of the market never chose

Here is the finding that should reframe the whole comparison.

Across the products in the dataset with a headline user number, 58.3% of those numbers describe people who were given an assistant, sold it inside a bundle, or counted through a product they already used. Not people who picked one. Some 41.2% of the stacked headline user total sits on products nobody actively selected.

And of 15 adoption figures in the workbook, two are audited.

So when a comparison article tells you a platform has hundreds of millions of users, that number is probably (a) not audited and (b) counting exposure rather than use. Google's AI Mode reports a billion monthly users; in panel data, only 0.34% of searches actually moved into it.

OpenAI's own published analysis of how people use ChatGPT, based on 1.5 million conversations and released alongside an NBER working paper, is one of the few first-party sources that describes actual usage rather than reach. It is worth reading precisely because it is a vendor being specific rather than promotional.

The practical version: if you are choosing an assistant for a team, market share matters far less than you would think, because a large part of that market did not choose either. Pick on your own task performance, your data rules and your price ceiling.

If your brand is what gets compared, this changes your job

There is a second audience for a Claude vs ChatGPT search, and it is people who sell software rather than people buying an assistant.

If buyers now ask an assistant to compare products, then which assistant they ask decides which answer they get – because the platforms do not read the same web.

A citation is what you get when an assistant quotes your page and links to it. Count every citation a site receives across six major assistants, and you can see which assistant is doing the quoting.

The split is stark. Of all the times the review site G2 is quoted, only 11.3% of those quotes come from ChatGPT. Wikipedia gets 37.5% of its from Perplexity, more than from any other assistant. YouTube gets 41.7% of its from Google's AI Overviews. On one site in the probe the range ran from 1.2% to 54.2%, depending purely on which assistant was asked.

That has three consequences worth acting on:

A visibility report tracking one platform is measuring that platform's house preferences and calling it the market. The distinction between being quoted and being visited is covered properly in why citation share is not traffic.
Reputation matters less than format. In a software-category probe, 43 comparison sites nobody tracks out-cited G2, Capterra and TrustRadius combined.
The list of surfaces keeps growing. Grok now reads X posts in real time, which makes social content something an answer can quote.

What the release data says about the two labs

Share is one view. The release history is another, and it is less flattering to the "one of them is winning" framing than either camp would like.

Across 261 releases from 15 labs, neither of these two has the most models currently available. That is Mistral AI with nine, because it ships many small specialised models rather than a short ladder of large ones.

Both labs now also ship models that were never generally released. Seven models on the current shelf are restricted access, and seven carry a category naming cyber capability. That is new to this dataset.

And whichever you pick, the ground moves: 75.5% of every model ever announced is already superseded or retired, and the median current model is 60 days old. Neither of these is a stable choice in the way a database is a stable choice. Write the swap clause.

How to decide between Claude vs ChatGPT

Ignore the share figures for the decision itself. Use them to understand the landscape, then do this instead.

Pick ten real tasks. Not sample prompts. The things your team actually does, with your actual data.

Run both, blind if you can. Have someone score the outputs without knowing which tool produced them. This removes brand preference, which is stronger than most people admit.

Score on the things that break deals, not just quality: data residency, retention policy, rate limits, price per million tokens, and what happens when the underlying model is deprecated.

Re-run it each quarter. At a release every 2.98 days, a decision made once is a decision made on stale evidence within months.

Keep the test. The harness is the durable asset. The verdict is not.

Questions people ask about Claude vs ChatGPT

These are the questions searched most often alongside Claude vs ChatGPT, answered from the release and traffic data rather than from opinion.

Which is better, Claude or ChatGPT? The data in this piece cannot answer that, and neither can anyone else's share chart. It can tell you the market is no longer one-sided: ChatGPT holds a bit over half of tracked assistant traffic, down from three quarters a year earlier. Run both against your own ten tasks.

Is ChatGPT losing users? Not necessarily. It lost 23.7 points of share in eleven months, but the category grew over the same period, so share can fall while absolute usage rises. These are different measurements.

How fast is Claude growing? Its share of assistant web traffic went from 1.6% to 9.2% between June 2025 and May 2026, a 5.8-fold increase and the fastest in the tracked set. It remains the smallest base among the western frontier labs.

Can I trust the user numbers in comparison articles? Treat them carefully. In this dataset, 58.3% of headline user figures describe people who were given an assistant rather than people who chose one, and only two of 15 adoption figures are audited.

Does it matter which assistant my customers use? Yes, more than the tool choice does. The six assistants quote very different sources: on one site in the probe, the share of its citations coming from a single assistant ranged from 1.2% to 54.2%. So which assistant a buyer asks materially changes which brands appear in the answer.

Traffic share figures come from the assistant traffic share column of a workbook of 261 model releases across 15 labs, June 2025 to May 2026. Citation figures come from a live Ahrefs index pull across six answer platforms on 7 September 2026. Every figure was computed once and independently re-derived before publication. Full method in the report, How AI Changed the Way People Search.