AI assistants send almost no traffic. That's the wrong reason to ignore them.

Every few months the search industry has an argument about whether AI assistants matter, and this week it had a big one. Three studies landed within days of each other, rounded up by Search Engine Roundtable, and they point the same way. ChatGPT, Perplexity, Gemini and the rest send a small fraction of the visitors Google sends. Not a third. Not a tenth. Closer to a fortieth.

We read all three closely, because we run a site that tells CPG founders how to show up in AI answers, and if the honest reading is "don't bother," we should say so. It isn't. But the reason it isn't has very little to do with traffic, and the studies are worth understanding before anyone uses them to win an internal budget fight in either direction.

What the three studies found

Marie Haynes looked at 12 months of Google Analytics data across 100 sites and more than 145 million sessions. ChatGPT sessions came to 0.62% of Google organic volume. Add every LLM together and it was 0.73%. On the 55 sites that tracked conversions, ChatGPT visitors converted at 13.8% against 3.02% for Google organic, more than four and a half times better per visit. Google still drove 96.5% of conversions across the channels she analyzed, and Google organic alone produced 27 times more conversions than all LLMs combined. She also found ChatGPT referrals stopped growing around August 2025 and have fallen since. Her piece is titled GEO Is A Distraction.

The SEO Works, a UK agency, published an AI referral traffic benchmark across 95 sites from September 2025 to August 2026. Organic search sent 2,825,594 sessions; AI platforms sent 72,095. That's 39 organic visits for every AI visit. AI converted at 7.18% against 6.08% for organic, a smaller edge than Haynes found, and AI did worse than organic on 23 of the 63 sites where the comparison was possible.

Kevin Indig at Growth Memo went after a different claim: that AI traffic is undercounted because people read an answer, then type your URL or search your name later. Across 12 analytics properties, total direct traffic rose 140.9%, but engaged direct sessions moved by about minus 1%. The growth was bots. Sessions with identifiable AI traces made up 0.003% of direct. In 632 tests across 4 brands he found no reliable link between AI visibility and later branded search or direct visits.

Study	Sample	AI traffic vs. Google organic	AI conversion vs. organic
Marie Haynes	100 sites, 145M+ sessions, 12 months	0.62% (ChatGPT), 0.73% (all LLMs)	13.8% vs 3.02% (55 sites)
The SEO Works	95 sites, Sep 2025 to Aug 2026	2.6% (1 AI visit per 39 organic)	7.18% vs 6.08%
Kevin Indig	12 GA4 properties, 632 tests	No measurable hidden lift in direct or branded	Not measured

These are careful pieces of work, and the authors are mostly sensible about what they mean. Haynes isn't saying AI doesn't matter; her argument is that the opportunity is in agentic shopping and Google's own surfaces rather than chatbot citations. Indig isn't saying visibility is worthless; he says measure it with three signals (where you appear, what customers say when asked how they found you, and commercial outcomes) instead of hoping direct traffic proves it.

The per-visit edge is real and also not enough

The conversion numbers get quoted a lot, so it's worth being clear about what they can and can't do.

The SEO Works did the useful arithmetic. Across the panel, AI produced 5,001 conversions and organic produced 155,605. That's 3.2 AI conversions for every 100 from organic. They then modeled what a typical site in the panel would gain each month from different moves.

Show the data

Scenario	Extra conversions per month
Double AI traffic	6.6
Triple AI traffic	13.2
Organic +5%	10.3
Organic +10%	20.6

Source: The SEO Works, AI Referral Traffic Benchmark. Doubling organic traffic was modeled at +205.8 a month and is left off so the other bars stay readable.

Tripling AI traffic is worth less than growing organic by 10%. If your own site is the main place you sell, that chart is the whole argument, and it's a fair one.

The conversion edge also depends on what you compare it to. Against paid search, The SEO Works found AI converted at 0.67 times the rate. Against direct traffic it was 2.48 times. The "AI converts better" headline is partly a statement about which baseline you pick.

Show the data

Compared with	AI conversion ratio
Paid search	0.67x
Organic search	1.18x
All sessions	1.37x
Direct	2.48x

Source: The SEO Works, AI Referral Traffic Benchmark

One detail in their split matters for our readers. B2C sites saw AI convert at 1.51 times organic, the best of any group. B2B product sites saw 0.62 times. Consumer brands are on the better end of a small number.

Where the studies stop being about you

Here's the part none of the three set out to measure, and the reason we don't think a CPG founder should read them as permission to look away.

Most CPG sales don't happen on the brand's website. A website-analytics study can only count people who land on your site. If an assistant tells someone your protein bar is the one to try and they buy it at Target, Walmart or their grocery delivery app, your Google Analytics never hears about it. For a DTC apparel brand, the referral is the sale. For a snack brand in 4,000 stores, the referral is a rounding error and the recommendation is the point. None of these studies are wrong about that; it's just outside what they counted.

The biggest AI answer is filed under Google. The SEO Works says plainly that clicks from Google's AI Overviews are counted as organic, so the AI share in its data is a floor. Haynes makes the same point from the other direction: the AI search that sends people is Google's. Meanwhile Semrush's new study found AI Overviews on 88% of the commercial keywords it tracks, up from 45.75% a year earlier, and on 85.11% of ecommerce ones. "Best X for Y" searches had an Overview 96.12% of the time. So the most-read AI answer about your category is sitting on top of the channel these studies treat as the safe, non-AI choice. We wrote up those numbers in a separate post.

The trend line moved this year. The SEO Works' monthly data shows AI holding around 2% of organic through April 2026, then stepping up after early May, when, by their account, ChatGPT started linking brand names inline in its answers.

Show the data

Month	AI as % of organic
Sep 25	2.25%
Oct	2.47%
Nov	2.76%
Dec	2.57%
Jan 26	2.12%
Feb	2.01%
Mar	2.12%
Apr	2.14%
May	3.06%
Jun	2.76%
Jul	2.62%
Aug	3.86%

Source: The SEO Works, AI Referral Traffic Benchmark

That's still small. It's also a reminder that these numbers depend on product decisions made in San Francisco, and they can change in a month.

A website-analytics study can only count people who land on your site. For a brand in 4,000 stores, the recommendation is the point.

The work is mostly the same work

The most useful finding across all three studies might be the least quoted one. The SEO Works checked whether sites that grew AI traffic did so at the expense of organic. They found the opposite: AI growth and organic growth moved together, with a correlation of +0.557 (+0.534 after controls). Sites that did well in one tended to do well in the other.

That fits what we see when we look at why an assistant names one brand over another. The inputs are things a good search and retail team already cares about:

• product pages that state plainly what the product is, who it's for, and the facts a shopper compares on, which we covered in tip 3
• the same facts on Amazon, Walmart, Target and your own site, so no source contradicts another
• reviews that mention specific uses, on the sites that get cited
• mentions in the publications and forums assistants pull from, which is what people mean by GEO when they aren't selling something

None of that needs a separate "AI budget." It also doesn't need you to opt out of anything. Today's tip 34 is about the Search Console setting that removes a site from AI Overviews, and why we'd leave it alone.

So what should a founder do with these studies?

Use them to kill two bad ideas and keep one good one.

The first bad idea is that chatbot referrals will replace Google traffic any time soon. They won't, and anyone pitching a plan built on that should be asked for their numbers.

The second bad idea is the "hidden traffic" story: that AI sends far more people than analytics shows, through direct visits and branded searches. Indig's 632 tests are the best evidence we've seen on this, and they don't support it. If you want to know whether AI answers are sending you customers, ask them. A "how did you hear about us" field at checkout, or a one-question post-purchase survey, beats any attribution model. Indig cites Ramp, which found 30% to 40% of its direct traffic came from answer engines through exactly this kind of self-reporting.

The good idea worth keeping is that what an assistant says about your product is now a shelf, and you should know what's on it. That means checking the answers, in Google and the chat assistants, for the questions your shoppers actually ask, and fixing what's wrong at the source. Tools have appeared for this; XENA Intelligence, for one, built SearchPanel to score products by what the assistants say about them. A spreadsheet and an afternoon a month gets a small brand most of the way, as we laid out in our post on measuring share of answers.

The studies answer the question "how many visitors will ChatGPT send my website?" The answer is: not many, for now. The question a CPG brand should be asking is "when someone asks a machine what to buy in my category, does it name me, and does it get the facts right?" Nobody has measured that across 100 sites yet. You can measure it for one.