ChatGPT handles a fifth of Google's searches and sends 0.36% of visits. Both numbers hold up.

On Monday night in New York, at a dinner OpenAI hosted for reporters, ChatGPT's consumer product lead Adam Fry gave Alex Kantrowitz of Big Technology a number the search industry had been guessing at for two years. "ChatGPT's volume of search-like queries is now roughly one-fifth the volume of queries of Google," he said.

The next day, Ahrefs refreshed its public AI vs Search Traffic Analysis, which tracks where visits come from across the same 118,671 websites every month. In October, ChatGPT sent 0.36% of those sites' visits. Google sent 24.74%.

Put side by side, those figures look like a contradiction. One says ChatGPT is a serious search engine. The other says it barely registers as a traffic source. We think both are roughly right, and the gap between them is the most useful thing a CPG founder can take from this week's news.

What OpenAI actually claimed

The details matter, because "a fifth of Google" will get repeated without them.

Fry's counting rule is generous. Asked whether each question in a conversation counts separately, he told Kantrowitz, as quoted by Search Engine Journal: "Every time you ask a question and go search the web, that's a search." The queries that qualify, he said, are the ones that need current information: looking up a person, a place or a thing, and finance and health questions. "It's basically anything that a user would have traditionally gone to a traditional search engine for, but is now deciding to go to ChatGPT for," he said.

What he did not say matters as much. SEJ notes that Fry did not reveal which Google figure he compared against, over what period, or which regions are covered. The Verge also flagged the missing Google baseline.

He also gave a growth figure: ChatGPT went from 1 billion to 1.2 billion weekly active users in about 60 days.

SEJ adds useful context. In February, Ahrefs estimated ChatGPT's search volume at around 12% of Google's, using a different definition: it sorted prompts by the kind of request rather than counting web searches. So OpenAI's own figure is higher than that outside estimate, and the two are not measuring quite the same thing.

None of this makes the claim false. It makes it OpenAI's number, measured OpenAI's way, without an outside audit. Treat it as a direction of travel, not a market share.

What Ahrefs measured

Ahrefs' figure is a different kind of number. It comes from sites running Ahrefs' free Web Analytics, and it counts visits that arrived at those sites, labeled by where they came from. Because Ahrefs keeps the set of sites fixed, month-to-month changes reflect real shifts rather than new sites joining the panel.

Show the data

Source	Value
Google	24.74%
Bing	1.19%
Yahoo!	0.75%
ChatGPT	0.36%
Gemini	0.02%
Perplexity	0.01%

Source: Ahrefs, AI vs Search Traffic Analysis (updated Nov 3, 2026). Ahrefs Web Analytics data for the same 118,671 sites each month. All AI assistants combined: 0.46%. Traditional search engines combined: 27.55%.

Two things stand out. ChatGPT is by far the biggest AI source, sending most of the 0.46% that all AI assistants produced together. And it still sends fewer visits than Yahoo.

The channel view puts it in proportion. Ahrefs groups sources into categories that overlap, so they add up to more than 100%, but the order of magnitude is clear.

Channel	Visits	Share of tracked visits	Change vs September
Direct	374.2M	31.08%	-8.21 pp
Search engines	331.6M	27.55%	+0.73 pp
Unknown	229.2M	19.03%	+5.41 pp
Paid	202.0M	16.77%	+1.29 pp
Social	93.7M	7.78%	+0.36 pp
Email	13.7M	1.14%	+0.10 pp
AI assistants	5.4M	0.46%	+0.03 pp
Referral	3.3M	0.27%	-0.01 pp

Source: Ahrefs. "pp" means percentage points.

Why both numbers can be right

A search engine and an answer engine do different jobs with the same question.

When someone searches Google for "best mineral sunscreen for oily skin," Google's business has historically depended on sending them somewhere: an ad, a retailer, a review site. Even with AI Overviews, a lot of that journey still ends in a click.

When someone asks ChatGPT the same thing, ChatGPT runs a search, reads the pages, and writes the answer. The shopper gets three brand names and a reason for each. Many stop there. Some open a retailer app and buy. A few click a citation to check it. Only that last group shows up in anyone's referral report.

So the two numbers are measuring different stages. OpenAI's figure counts questions. Ahrefs counts visits that followed. If you put them together loosely, the ratio between them is the click rate of an answer engine.

Here is that rough arithmetic, with a warning attached. Ahrefs' panel shows Google sending about 69 visits for every one from ChatGPT (24.74 divided by 0.36, our calculation from the reported shares). If ChatGPT really handles a fifth of Google's queries, then each ChatGPT search would produce something like a fourteenth of the site visits a Google search produces. That combines two sources that count different things over different populations, so read it as an order of magnitude, not a finding. But the order of magnitude is the point: the question is being asked, and the visit mostly isn't happening.

The question is being asked. The visit mostly isn't happening.

The trend is slower than the headlines

The growth story is also more uneven in Ahrefs' panel than in OpenAI's user numbers.

Period	Google	ChatGPT	Bing
Jul to Aug 2026	-7.2%	+33.2%	+0.8%
Aug to Sep 2026	-2.7%	+14.1%	+2.1%
Sep to Oct 2026	-2.3%	+0.4%	+0.3%

Source: Ahrefs.

ChatGPT's referrals grew fast through the late summer and then went flat in October, while Google's kept slipping. Ahrefs' average across its whole series is a 5.2% monthly rise for ChatGPT and a 3.0% monthly fall for Google. Even at those rates, the gap in the chart above would take years to close. Meanwhile ChatGPT added 200 million weekly users in two months, by OpenAI's count. More people asking, about the same number clicking: that is what an answer engine doing its job looks like.

What this means for a CPG brand

For brands that sell mostly through retailers, the referral number was always going to undersell ChatGPT. A shopper who asks for a protein bar recommendation and then buys at Target or Amazon never touches your site. We made that argument at length in our piece on why small AI traffic is the wrong question. This week's numbers add a scale to it.

Four practical consequences:

1. Stop using referral traffic as your AI visibility metric

It is a lagging, partial signal. It tells you that some people clicked, not how often you were named, what was said, or whether it was right. Keep it in your reporting (our GA4 guide covers setting it up), but report it next to a measure of the answers themselves.

2. Measure the answers directly

Pick the 20 to 30 questions your shoppers actually ask, run them in ChatGPT on a schedule, and record whether you are named, in what position, and whether the facts are correct. A spreadsheet works. So do tools built for it: XENA Intelligence asks assistants shopper-style questions and scores how products show up, and its SearchPanel product does this per product. What matters is consistency: the same questions, the same way, every month.

3. Treat the clicks you do get as verification

The visits that arrive from an AI answer are often people checking what they were just told, a pattern Fractl documented: 56% of US adults in its survey said they read an AI answer and then click a link to double-check it. We covered that in the quiet click. The page they land on should confirm the facts quickly: ingredients, sizes, claims, price, where to buy.

4. Don't move budget on either number alone

"A fifth of Google" is not a reason to shift search spend to ChatGPT ads; we recently looked at a 1,898-shop study where a paid ChatGPT visit earned $0.89 against $2.45 for a paid Google visit. And "0.36% of traffic" is not a reason to ignore what ChatGPT says about your products. The first number is about attention. The second is about clicks. Brands that sell through retailers mostly need the attention.

The caveats, in one place

• OpenAI's figure is self-reported, its counting method is generous by search standards, and the source of its Google comparison was not given.
• Ahrefs' panel is sites that installed its free analytics tool. It is large and stable, but it is not a random sample of the web, and Ahrefs does not break out CPG brand sites.
• Ahrefs' "Unknown" channel rose 5.41 points in October. Some AI visits arrive without a referrer and land there or in Direct, so AI's true share is likely somewhat higher than 0.46%. It is very unlikely to be anywhere near Google's.
• Our ratio calculation mixes the two sources and is only good for scale.

The short version: more of your category's questions are being answered by ChatGPT every month, and almost none of that shows up as traffic. Plan your measurement around the answers.