Module 95: The LLM Traffic Reality Check — The 0.62% Math That Should Rebalance Your GEO Budget
The Signal — What Changed
Marie Haynes published a data study on Search Engine Journal on October 23, 2026: "GEO Is A Distraction: Here's Where The Real Opportunity Is." She analyzed 12 months of GA4 data across 100 websites — over 145 million sessions — and pulled the actual numbers behind the AI-search hype:
- ChatGPT sessions were just 0.62% of the volume driven by Google Organic. Including every other LLM (Perplexity, Gemini, Copilot, and the rest), the figure rose only to 0.73%.
- The conversion-quality story is real: ChatGPT visitors converted at 13.8% vs 3.02% for Google organic — a 4.5x conversion-rate edge.
- But volume dominates rate: Google still drove 96.5% of the total conversions across all channels — roughly 27x more conversions than all LLMs combined.
This lands a week after the same publication and SEL covered AI Overviews expanding over commercial keywords and AI brand recommendations swinging by platform and week. The industry narrative says "optimize for AI search or die." The measurement says: most sites get a rounding error of traffic from LLMs, but each visitor is worth 4–5 Google visitors. Both facts are true, and the Get-Found doctrine says you allocate budget with numbers, not with vibes.
Why It Works
Three principles make this the most important analytics module of the quarter:
- Conversion-weighted allocation beats channel hype. A channel with 0.62% of volume but 4.5x conversion rate deserves some investment — but not a headline budget line. When you multiply sessions × conversion rate, Google's 27x conversion advantage is the number that pays salaries. Budget should follow conversions, and today almost every SMB's conversions still come from Google organic + local pack.
- The 4.5x edge has a structural reason you can serve. A ChatGPT visitor arrives mid-decision — they've already asked the model to research options. High intent means: answer their one remaining question fast (pricing, coverage, availability), make the phone number unmissable, and never make them re-verify your credibility. That's a landing-page and entity-consistency job, not a separate "GEO strategy."
- Measurement kills both failure modes. Without GA4 segments, agencies over-sell GEO (chasing citations nobody converts from) or ignore it entirely (missing the 13.8% visitors hiding in "referral" traffic). One hour of segment setup per property replaces both arguments with a monthly number.
The compounding angle still matters — as Edward Sturm put it this week, "the best way to get covered in AI is SEO": LLM citation presence is downstream of the same entity signals, structured answers, and review credibility that classic SEO already builds. You are not choosing between Google and AI search. You are doing SEO properly, then measuring where it pays.
How to Run It
Step 1 — Build the LLM segment in GA4 (15 minutes, per property)
- In GA4: Explore → Free form → Segments → Create new segment → Session segment.
- Name it
LLM Referrals. Condition:Session source / mediummatches regex:chatgpt\.com|perplexity\.ai|gemini\.google|copilot\.microsoft|claude\.ai|you\.com. - Create a comparison segment for
google / organicsessions alongside it. - Add metrics: Sessions, Engaged sessions, Key events (conversions), Conversion rate. Set the date range to trailing 12 months.
Step 2 — Compute the two numbers that matter (10 minutes)
For each property, calculate:
- Volume share: LLM sessions ÷ Google organic sessions (Haynes' dataset: 0.62%–0.73% — anything within an order of magnitude of that is "normal").
- Conversion-weighted share: (LLM conversions ÷ total conversions) vs (Google conversions ÷ total). This is the budget-decision number. If LLM conversion share exceeds ~2% of total, GEO gets real hours; below that, GEO stays inside the normal SEO retainer (entity consistency, answer blocks, reviews).
Step 3 — Apply the decision rules (the actual playbook)
- If LLM volume share is under 1% and conversion share is under 2%: keep 90%+ of hours on Google organic and local. Maintain GEO hygiene — consistent NAP, review velocity, clear answer blocks — inside the standard retainer. Do not sell a separate GEO package on these numbers.
- If LLM conversion share is 2–10%: add a monthly "AI-surface check" — ask the major models the client's top 5 money questions, log whether the client is cited and whether the cited facts are correct, and fix wrong answers at the source (the client's own pages and the profile data the models read).
- If LLM conversion share is above 10%: build dedicated high-intent landing treatment for LLM referrals: one-screen answer to the money question, click-to-call above the fold, proof (reviews, licenses) visible without scrolling. Measure whether the 4.5x rate holds on your traffic.
For local service clients (HVAC, roofing, security, appliance repair)
- Run the segment on the property; most will show sub-1% volume — confirm it and move on.
- Local conversions live in the GBP call and direction data, not web forms — pull GBP Performance calls alongside GA4 so the report doesn't undercount Google's lead contribution.
- Spend the saved GEO hours on the things that actually move the local pack: review velocity, GBP posts/photos, one keyword per city page.
For insurance/YMYL clients (xactinsure.net, floridahealthinsuranceweb.com)
- Run the segment, then check which pages LLM referrals land on. Insurance LLM visitors are typically quote-and-eligibility stage — if they land on a blog post instead of the quote path, add internal links from the cited page to the quote page.
- Verify every LLM-visible fact on money pages (plan years, enrollment dates, carrier names) — a wrong date cited by a model is a compliance problem, not just a conversion problem. YMYL accuracy standards apply to what AI says about you, not just what your site says.
- Keep the people-first/E-E-A-T program as the priority: models cite the sites that quality systems already trust.
Real Example
growwithfootprint.com runs this as the internal case study: build the segment, pull 12 months, and produce the one-page scorecard — volume share, conversion share, and the decision-rule verdict. The scorecard becomes the client deliverable for every managed property: it replaces "we do GEO now?" conversations with a monthly number, and it protects retainer scope by proving where the leads actually come from.
xactinsure.net gets the YMYL version: insurance intent is exactly the high-conversion LLM visitor Haynes measured (13.8% conversion rate), so the volume share will look small while the conversion share over-performs. The action is landing-path hygiene — LLM referrals reaching quote pages — and fact-accuracy verification on any page an AI model cites, because a model repeating a stale enrollment date is an E-E-A-T and compliance issue at once.
Checklist
- [ ] GA4 session segment
LLM Referralscreated with the 6-domain regex - [ ] Google / organic comparison segment added, trailing 12-month window
- [ ] Volume share computed: LLM sessions ÷ Google organic sessions
- [ ] Conversion share computed: LLM key events ÷ total key events
- [ ] Decision rule applied (under 1% / 2–10% / above 10% thresholds) and logged
- [ ] Local clients: GBP call data pulled alongside GA4 so Google's contribution isn't undercounted
- [ ] YMYL clients: LLM landing paths checked — citations reach quote/conversion pages
- [ ] YMYL clients: facts on AI-visible money pages verified current (dates, carriers, plan years)
- [ ] Scorecard saved per property; next review date set (monthly)
Quiz
Q1: Across Haynes' 100-site dataset, what share of Google organic volume did all LLMs combined represent, and why doesn't that end the GEO conversation?
0.73% (ChatGPT alone was 0.62%). It doesn't end the conversation because ChatGPT visitors converted at 13.8% vs Google's 3.02% — a 4.5x edge — so each LLM visitor is worth 4–5 Google visitors. Tiny volume, real quality: fund hygiene, not a separate GEO package.
Q2: A local roofing client's GA4 shows LLM conversion share of 1.2%. What do you do with the GEO budget conversation?
Apply the decision rules: under 2% conversion share means GEO stays inside the standard SEO retainer — entity consistency, answer blocks, review velocity. Do not sell a separate GEO package. Also pull GBP call/direction data before reporting, because web-form conversions undercount Google's real local contribution.
Q3: An insurance client's LLM referrals are landing on blog posts instead of the quote page, and one cited page shows last year's enrollment dates. What are the two fixes?
(1) Landing-path fix: add internal links from the cited pages to the quote/conversion page so high-intent visitors reach it. (2) Fact-accuracy fix: update the stale dates on-page — for YMYL content, what AI models repeat about you must meet the same accuracy standard as your own pages; a wrong cited date is both a conversion and a compliance problem.
Try This Today
Pick one managed property — start with growwithfootprint.com — and complete Steps 1–2 of the playbook in under 30 minutes: create the LLM Referrals segment, run the trailing-12-month comparison against Google organic, and write down the two numbers (volume share, conversion share). Then apply the decision rule and log the verdict. Repeat across properties this week; the first scorecard you hand a client replaces an hour of GEO debate with one line of math.
Sources
- Search Engine Journal — Marie Haynes, "GEO Is A Distraction: Here's Where The Real Opportunity Is" (Oct 23, 2026) — the 100-site / 145M-session dataset: article
- Search Engine Land — "Content strategy in 2026: What actually changed" — traffic and visibility decoupling context: guide
- Google Analytics 4 — segments and comparisons (the measurement tool this module installs): GA4 segments documentation
- Edward Sturm — "the best way to get covered in AI is SEO" (Oct 21, 2026): edwardsturm.com
- SEOFOMO by Aleyda Solis — weekly AI-search tracking and the State of AI Search Optimization survey: seofomo.co