AI sends a sliver of your traffic. That's a reason to do the work in the right order, not to skip it.

Cloud AI cites who it can see. This week, three people who measure websites for a living asked a fair follow-up question: fine, but does anyone click?

Their answer, from their own client data: not many. Search Engine Roundtable put the three side by side today, October 22, under the headline 3 New Studies Suggest AI Visibility & Search Not Worth Investing In (HTTP 200 this fire).

I read all three. I think the numbers are mostly right, and I think the headline takes the wrong lesson from them. Here's the data first, then what I'd actually do with it.

The short version

• AI referrals are small next to Google organic. In one 100-site dataset, ChatGPT sessions were 0.62% of Google organic sessions. In another 95-site dataset, organic sent 39 times more traffic than AI assistants.
• AI visitors can convert well, but the volume is tiny. Google organic still drove 96.5% of conversions in the first dataset. In the second, AI delivered 3.2 conversions for every 100 from organic.
• AI and organic grow together. In the 95-site dataset, the sites gaining AI referral traffic were the sites whose organic search was improving.
• The "people see you in AI and come back later" theory didn't show up in analytics. Two of the three couldn't find a reliable link between AI visibility and later direct or branded traffic.
• GA4 can't see most of AI search. Google's AI Overviews arrive as ordinary organic visits. Answers that never produce a click don't show up at all.
• So: do the work in order. Be the clearest source about your business on your own pages. That work serves Google first and AI engines second. Measure AI discovery by asking buyers, not by staring at referral rows.
• Monitoring is not execution. A dashboard of AI mentions doesn't move either number. Fixing the source does.

What the three datasets actually say

Finding	Source (opened this fire)
12 months of GA4 data, 100 websites, over 145 million sessions. ChatGPT sessions = 0.62% of Google organic volume; all LLMs together = 0.73%	Marie Haynes, "GEO is a distraction," Oct 21, 2026
55 of the 100 sites tracked conversions. Google organic converted at 3.02%, ChatGPT at 13.8%. Google organic still drove 96.5% of total conversions; 27x more than all LLMs combined	Haynes
45 of the 100 sites "were not tracking any key events"	Haynes
Haynes: ChatGPT sessions across her 100 properties stopped growing around August 2025 and have declined since	Haynes
95 websites, Sep 2025–Aug 2026: 2,825,594 organic sessions vs 72,095 AI referral sessions. Organic sent 39x more; AI = 2.6% of organic	The SEO Works, "How Much Traffic Do AI Assistants Actually Send, And Does It Convert Better?" (published Sep 7, updated Oct 14, 2026)
AI share of organic sat between 2.0% and 2.8% until April 2026, then rose to 3.1% in May and 3.9% by August. SEO Works ties the timing to ChatGPT making brand names clickable inline around May 7, 2026, and calls that "a working hypothesis rather than an established cause"	SEO Works
AI referrals converted at 7.18% vs 6.08% organic (1.18x, n=63 sites). AI converted worse than organic on 23 of 63 sites; B2B product sites at 0.62x organic	SEO Works
Versus other channels: 0.67x paid search, 1.37x total sessions, 2.48x direct. "Anyone quoting a conversion multiple for AI traffic without stating the comparison channel has published a number that cannot be checked or reproduced."	SEO Works
5,001 AI conversions vs 155,605 organic: 3.2 AI conversions per 100 organic. A 5% organic lift = 10.3 extra conversions per site per month; doubling AI referrals = 6.6	SEO Works
AI referral growth vs organic growth across 61 sites: correlation +0.557; +0.534 after controlling for growth in the rest of the site (p < 0.0001). AI growth vs direct growth, controlled: −0.230 (p = 0.075)	SEO Works
AI Overviews are excluded: those clicks "arrive as ordinary organic sessions and cannot be separated," so "the measured AI contribution should be read as a floor"	SEO Works
12 GA4 properties, Aug–Sep 2026. Direct sessions with identifiable AI traces = 0.003% of direct traffic. Much of the direct growth was bots: one site logged 409,000 direct sessions, 198 engaged	Kevin Indig, "Does AI drive more direct traffic?", Growth Memo, Oct 19, 2026
Across 632 tests on four brands, Indig "couldn't establish a reliable link between AI visibility and subsequent branded search or Direct traffic. An effect could still exist. I just couldn't isolate it."	Indig

Where I agree with the headline

If your plan was "ignore Google, chase ChatGPT mentions, and wait for the traffic," the data says that plan is wrong. Google organic is still the channel that sends most of the people and most of the sales for these sites. Haynes says it plainly: "The AI search that is actually sending people to businesses today isn't ChatGPT. It's Google."

I'd add one thing she also says: a lot of that Google traffic is already AI search. AI Overviews, AI Mode, and ranking systems driven by machine learning are all inside Google. If you're asking "should I do AI visibility or SEO," you've drawn the line in the wrong place.

Where the headline goes further than the data

1. The work is mostly the same work. The strongest finding in the SEO Works dataset isn't the 39x. It's that AI growth and organic growth moved together, even after controlling for sites that were simply getting bigger. Their reading: "the same underlying work, being a clearly defined business with consistent information and genuine authority, drives both." That's not a reason to skip AI visibility. It's a reason to stop treating it as a separate project with a separate budget line.

2. Clicks are the wrong unit for a lot of AI search. All three datasets count what lands in GA4. None of them can count an answer that names you and never sends a click, or an AI Overview visit that looks like plain organic. Indig's comment section has a B2B buyer describing exactly that: the research happens inside ChatGPT and "never touches your site at all." That's an anecdote, not data. It's also why "AI sends almost no traffic" and "AI shapes which businesses get considered" can both be true.

3. The "billboard effect" is unproven, not disproven. Neither SEO Works nor Indig could find AI visibility pushing later direct or branded visits. That's an honest negative result. It also means anyone selling AI visibility on "they'll remember you and come back" doesn't have analytics proof yet. Indig names the one place he's seen it measured: Ramp, using self-reported attribution, in a podcast interview he relays. We didn't open that interview.

4. "Converts 4x better" was always a fragile number. Haynes found 13.8% vs 3.02%. SEO Works found 1.18x pooled, and worse-than-organic on more than a third of sites. SEO Works notes that the higher multiples circulating elsewhere come from single-company or publisher datasets. Different samples, different denominators. Don't plan on a multiple.

What I'd actually do with this

1. Track conversions before you argue about channels. 45 of Haynes's 100 sites weren't tracking any key events. If you're one of them, no AI-vs-Google debate matters yet. Set up the lead, call, booking, or purchase as a key event in GA4 first.

2. Put the true answer on your own pages first. Services, service area, hours, pricing approach, credentials, who you're for. One clear page per fact, kept current. This is the work that showed up in both channels in the SEO Works data. Google's AI optimization guide (HTTP 200 this fire) says content people find "unique, compelling, and useful will likely influence your website's presence in generative AI search in the long run more than any of the other suggestions in this guide."

3. Treat AI visibility as a check on that work, not a separate campaign. Run the 10 questions your buyers actually ask in Google (including AI Mode), ChatGPT, Gemini, Perplexity, and Claude. Note who gets named and which URLs get cited. If you're absent or wrong, that's a gap in your source pages, and fixing it helps Google too. This is how New Reward runs it. It's a method, not a figure from the studies.

4. Read AI referrals as a floor. Use a referral-source rule (SEO Works used a regex on chatgpt, openai, claude, copilot, gemini, perplexity, grok, and similar). Compare against organic, not total sessions, and always say which channel you compared against. Remember AI Overviews hide inside organic, and ChatGPT links have recently been arriving without tags (see our ChatGPT UTM post).

5. Ask how people found you. Add a "How did you hear about us?" field to every form and intake call, free text, and keep the AI tool name if they give one. Indig recommends combining that with prompt tracking and revenue: "Where your brand appears. How people discovered you. What happens commercially." That's the only method in these three pieces that has caught AI discovery the analytics missed.

6. Don't buy a dashboard and call it a strategy. Haynes is blunt about money raised by mention trackers. She has a point. Knowing you were mentioned 40 times this month changes nothing unless someone fixes the page that was missing or wrong. Monitoring is not execution.

That's the proof loop: measure conversions, fix your own source pages, check where AI and Google name you, read AI referrals as a floor, and ask buyers.

What this is not (honesty block)

• Not official data. None of this comes from OpenAI, Google, Anthropic, or Perplexity. No AI company has published referral volumes for these sites. OpenAI's newsroom and Perplexity's changelog returned 403 this fire.
• Not a random sample of the web. Each dataset is the author's own client or partner properties: 100 sites, 95 sites, 12 sites. Industry mix differs. Indig's skews B2B.
• Not peer-reviewed. Practitioner analyses. SEO Works publishes its aggregated dataset; we didn't re-run it.
• Not complete AI search. GA4 can't separate AI Overviews or AI Mode clicks from organic, and can't see zero-click answers. SEO Works calls its AI number "a floor."
• Not agreed on trend. Haynes sees ChatGPT sessions declining since August 2025 in her data. SEO Works sees AI referrals rising as a share of organic from May 2026. Different samples; we don't know which is more representative.
• Not a GA4 product statement. SEO Works says Google's "AI Assistant" traffic category only exists from June 2026. We didn't open a Google Analytics page confirming that.
• Not proof the billboard effect is false. Two authors couldn't detect it. Indig: "An effect could still exist."
• Not a forecast. SEO Works: "the arithmetic changes if that trend continues." Four months of post-May data is short.

Related New Reward Field Guides

• ChatGPT UTM tags are missing. Direct traffic is not a visibility crash.
• GEO measurement is noisy, so build a proof loop
• SEO and GEO move together
• Cloudflare is helping ChatGPT find more pages. Found is not cited.
• Bing now shows the AI queries you're missing. That's a gap list, not a fix.
• ChatGPT's share of AI prompts fell. Visibility work still has to cover Gemini and Claude.

Want the fast read first?

I'm not going to pretend a scan is a strategy. The free AI-visibility score is a quick check: email only, no credit card, a 0–100 read across the engines we track. It won't tell you how much AI traffic you get. It will tell you whether AI can see your business clearly, and that's the same foundation Google rewards.