“AI visibility is unmeasurable” and “AI visibility is the next SEO” are both wrong — and the industry’s November 2026 fight is about data quality, not hype.
On November 6, 2026, Search Engine Journal’s Garry Grant published a synthesis of four independent datasets and the AMEC (International Association for the Measurement and Evaluation of Communication) framework for defensible metrics. The piece lands one week after Obsurfable’s LLM measurement and branded search article — but Grant’s angle is vendor skepticism: most tools optimize for slides, not decisions.
This article extracts the four studies, what each can prove, where they contradict, and how marketing leaders can build an audit-ready measurement stack without waiting for perfect Google Analytics attribution.
The four datasets at a glance
| Source | Published | Sample | Headline finding |
|---|---|---|---|
| Kevin Indig, Growth Memo | Oct. 19, 2026 | 12 GA4 properties + 632 correlation tests on four brands | No reliable link between AI visibility scores and later direct traffic or branded search in short windows |
| Marie Haynes | Oct. 21, 2026 | 100 sites, 145M+ sessions, 12 months | ChatGPT = 0.62% of Google organic sessions; all LLMs 0.73%; ChatGPT converts higher per session but Google drives 96.5% of conversions |
| SEO Works | Sept. 2025–Aug. 2026 | 95 GA4 properties | Organic sent 39× more traffic than AI assistants; AI beat organic on conversion in only 40% of sites with enough volume |
| Ahrefs | May 2025 (still cited) | 75,000 brands | Branded search volume correlated 0.392 with AI Overview brand visibility — weaker than web mentions (0.664) or branded anchors (0.527) |
Grant’s thesis: AI influence is real (buyers use assistants), but most commercial visibility scores fail basic quality gates — especially when vendors imply they predict revenue through a single composite index.
Gate 1 — AMEC validity: what are you actually measuring?
AMEC’s Valid Metrics framework asks whether a KPI is valid, reliable, sensitive, and actionable. Grant applies that lens to GEO dashboards:
- Valid: Does the metric track a concept that matters (e.g., buyer recommendation share)?
- Reliable: Does repeating the same prompt yield stable enough samples (see citation volatility)?
- Sensitive: Can it detect change after a real intervention?
- Actionable: Does a movement tell you which workstream to fund?
Many tools report a single visibility score blending mentions, citations, sentiment, and rank-like positions — without separating entity recognition from linked evidence. HubSpot’s November 3–4, 2026 AEO content emphasizes the same split: mentions vs citations require different playbooks.
Passing AMEC scrutiny means publishing method cards: prompt lists, engine versions, sampling cadence, and confidence intervals — not only a week-over-week delta on a proprietary index.
Gate 2 — Kevin Indig: visibility scores do not move GA4 (yet)
Indig’s October 2026 Growth Memo work is the skeptic’s anchor. Across 12 properties and 632 statistical tests:
- Early spikes in “Direct” traffic after AI visibility campaigns often vanished when filtered to engaged sessions.
- On one property, GA4 logged 409,000 Direct sessions with only 198 engaged — classic dark traffic / tagging loss, not proof of AI lift.
- Correlations between vendor visibility scores and subsequent branded search or direct were not stable at conventional significance thresholds.
Indig does not claim AI is irrelevant. He claims most score vendors have not shown incremental GA4 outcomes — aligning with his earlier ghost citation work (ghost citations analysis).
Operator takeaway: treat vendor scores as exploratory. Validate against your prompt panel and branded search panels you control.
Gate 3 — Marie Haynes: tiny traffic, loud conversion stats
Haynes’ 100-site GA4 study is the one vendors quote in sales decks:
- ChatGPT referrals ≈ 0.62% of Google organic session volume.
- ChatGPT conversion rate per session ≈ 13.8% vs Google organic 3.02% — a 4.5× multiplier.
Grant (and Haynes in her own writeups) stresses the denominator problem:
- Google organic still delivered 27× more conversions than all LLMs combined in the same dataset.
- ChatGPT session growth flattened around August 2025 (GPT-5 era) — so YoY charts may overstate momentum.
Operator takeaway: report absolute conversions from AI referrals alongside rates. A high conversion rate on 0.6% volume does not fund a channel shift.
Gate 4 — SEO Works: conversion advantage is not universal
SEO Works’ 95-property panel finds:
- Pooled AI referral conversion 1.18× organic — but AI converted worse on 23 of 63 sites with sufficient data.
- B2B product businesses saw AI traffic convert at 0.62× organic — the opposite of the Haynes headline.
That heterogeneity matters for Obsurfable’s audience (B2B SaaS, fintech, infrastructure). Category and funnel stage moderate AI referral quality. Grant argues composite “AI visibility ROI” claims that ignore vertical mix fail AMEC’s sensitivity test.
Gate 5 — Ahrefs: branded search is a weak proxy for AI visibility
The May 2025 Ahrefs brand study (still the best public correlation anchor) ranked predictors of AI Overview brand appearance:
| Signal | Correlation with AI Overview brand visibility |
|---|---|
| Branded web mentions | 0.664 |
| Branded anchor text links | 0.527 |
| Branded search volume | 0.392 |
If AI visibility caused branded search, you would expect branded search to correlate more tightly. The weaker link supports a measurement hierarchy:
- Mentions and third-party entity footprint (what models “remember”).
- Citations (what models retrieve and attribute).
- Branded search & direct (lagging business outcomes).
Obsurfable’s brand coverage more stable than citations article argues similarly: mention panels stabilize faster than URL citation sets — use both, not either/or.
Building a November 2026 measurement stack that survives finance review
Layer A — Owned prompt panel (weekly)
- 30–100 buyer questions you did not invent for SEO (support transcripts, sales notes).
- Log mention, citation, position, competitors — per engine.
- Store distributions, not single scrapes (volatility article).
Layer B — Attribution triad (monthly)
| Metric | Role |
|---|---|
| AI referral sessions (GA4) | Lower bound on linked traffic |
| Branded search (GSC) | Lagging demand capture |
| Self-reported source (forms, sales) | Captures zero-click journeys |
Layer C — Quality audit (quarterly)
Run AMEC-style checks Grant outlines:
- Transparency — publish prompts and engine versions.
- Neutral benchmarks — include competitor prompts, not only brand keywords.
- Outcome linkage — pre-register which KPI should move if a tactic works (citation share vs mention share).
Layer D — Business guardrails
- Finance sees absolute AI-sourced pipeline, not only conversion rates.
- Product marketing sees mention share on comparison prompts (comparison prompts cite most varies by engine — measure yours).
What vendors get right anyway
Grant’s piece is critical, not nihilistic. AI visibility is measurable when:
- Samples are large enough for stochastic answers.
- Metrics map to decisions (rewrite vs PR vs docs vs robots.txt).
- Teams accept non-determinism (Giordano measurement guide, cited in Obsurfable’s Nov. 2 article).
Composite scores are fine for monitoring — dangerous for commissioning without layer A–C validation.
How Obsurfable fits
Obsurfable is built on observations, not proprietary black-box indices: public Explorer corpora, repeatable prompts, and company-matched mention/citation rates in the App. Export prompt panels for AMEC documentation; use Tasks when a visibility score moves but citations do not — the gap Indig warns about.
FAQ
Is AI visibility “a sham”?
No. Unaudited visibility scores are the sham — not the underlying shift in buyer behavior.
Should we stop buying GEO tools?
Buy tools that expose raw observations and methods. Reject tools that only sell a number tied to undisclosed prompts.
Which single KPI should the CMO own?
Category mention share on a fixed prompt panel — supplemented by owned-domain citation rate when referrals matter.
Does branded search still matter if correlation is only 0.392?
Yes as a lagging outcome, not a leading visibility proxy. Pair it with mention/citation panels.
Bottom line
November 2026’s best measurement writing agrees on a uncomfortable truth: AI assistants influence buyers, but referrals stay tiny, visibility scores often fail outcome tests, and branded search is an imperfect echo. Leaders who survive budget scrutiny will measure mentions, citations, and business outcomes in separate layers — with AMEC-grade transparency — instead of one green arrow on a vendor dashboard.