GEO Has a Measurement Problem, Not a Hack Problem

The pitch that outran the evidence

Every week brings another agency case study claiming it got a client cited in ChatGPT through some formatting trick, a magic word count, or a specific way of front-loading statistics. A new academic survey put a number on how much of that actually holds up, and the number is not encouraging for anyone selling tactics.

Reviewing 45 studies published between November 2023 and July 2026, a critical survey of generative engine optimization research posted to arXiv this month concludes that no reviewed technique shows a stable, cross-platform causal effect on organic discoverability, once you strip away demos that already assume a source is sitting inside the model's context window. That distinction, being already retrieved versus being found in the first place, is the whole story. Most of the widely cited early gains only apply once your content already made it into the pool an AI system is drawing from. Getting invited into that pool and winning attention once you're already in it are different problems, and much of the answer engine optimization industry has sold tactics for the second while implying they solve the first.

For US growth teams running visibility programs against ChatGPT, Perplexity, and Google AI Overviews, that confusion is expensive. Budgets get allocated to citation tricks and schema tweaks while the actual bottleneck, whether content gets crawled, indexed, and retrieved at all, goes completely unmeasured.

What the research actually supports

The survey's author doesn't dismiss the field, he reframes it. GEO isn't a single ranking task, it's a nine-stage pipeline: search activation, crawling and indexing, retrieval, reranking and context allocation, citation, prominence, factual absorption, fidelity, and finally user behavior. Winning one stage says nothing about the others, and most vendor case studies only ever measure the one stage they already happen to be winning at.

Two findings matter most for anyone setting a budget around this. First, topical relevance and where content sits within the retrieved context are the only levers with reproducible evidence behind them. The generic formatting heuristics that fill most GEO checklists, bullet structuring, adding statistics, writing in a Q&A format, transfer poorly across platforms and frequently fail to replicate. Second, competition erodes individual gains: once every competitor in a category adopts the same trick, the trick stops differentiating anyone, and in some cases citation-oriented rewrites actually hurt a page's odds of being retrieved in the first place.

Independent commercial audits cited in the survey add a fidelity problem on top of the discoverability problem. Even when content does get cited, the citation isn't always accurate.

The ownership question nobody wants to own

For growth leaders, the more urgent problem isn't tactics, it's who is accountable for any of this. A recent industry survey found that 73% of communications professionals consider generative engine optimization at least somewhat important to their strategy, yet 29% say no one at their organization owns it. Just 24% say PR or communications owns it, compared with only 11% for marketing leadership.

That split makes sense once you look at why marketing owns SEO in the first place: clicks, conversions, and sales are directly measurable, so the team that can prove ROI keeps the budget. AI-generated answers break that model, because a chat response or an AI Overview increasingly resolves the query before a link ever gets clicked. The team best positioned to defend a budget for ai search optimization work is also the team GEO makes hardest to measure, and that mismatch is why the mandate keeps landing in nobody's lap.

The industry-wide numbers back up the gap. Most marketing organizations report having a GEO plan in place, but a clear minority say they can actually measure whether it's working. That distance between plan and measurement is the single most important statistic in this space right now, because it previews exactly what happens to any team that inherits AI-search ownership without a way to defend the line item at the next planning cycle: they either lose the mandate, or they keep it and start reporting vanity numbers, mention counts and sentiment scores, to look busy instead of effective.

A framework that separates the four things GEO actually measures

Instead of another tactics list, the survey proposes something more useful for a growth team building a reporting model: a visibility vector that separates four distinct outcomes, so leadership stops treating them as one blended metric.

Discoverability asks whether an AI-generated answer gets triggered by, and pulls from, your content at all. Citation asks whether the system actually names and links you once your content is retrieved. Absorption asks whether the answer faithfully represents what your content said, rather than a confidently hallucinated paraphrase. Economic outcome asks whether any of the above moves an actual business number, a lead, a signed deal, a subscriber.

Zero-click search has made this separation urgent rather than academic. Zero-click Google searches rose sharply in the year following the rollout of AI Overviews, meaning fewer of your discoverability or citation wins ever show up as a session in your analytics at all. A high citation count paired with weak absorption isn't a win, it's misattributed authority that a competitor's product page is quietly borrowing.

Most US growth teams still don't have a clean way to track more than the first two of these four stages. This is one of the few places where tooling has caught up faster than the research: platforms built for traditional SEO have started rolling brand-mention and AI-citation tracking into their standard reporting, giving teams a way to sit AI-visibility numbers next to the organic and paid metrics they already report on, instead of relying on a screenshot from a ChatGPT session. Semrush is one option worth a look if your team is building this reporting layer from scratch.

Key takeaways

• GEO is a nine-stage pipeline, not a single ranking task. Winning at one stage doesn't guarantee gains anywhere else, and most case studies only measure the stage they're already winning.
• Generic formatting hacks transfer poorly across AI platforms and frequently fail to replicate. Topical relevance and context position are the only levers with reproducible evidence behind them.
• 54% of marketers have a GEO plan; only 23% can measure it. That 31-point gap is the real competitive risk, not a missed formatting trick.
• 29% of brands have no clear owner for GEO work. Assign one before assigning a budget, and pick the team that can actually be held to a number.
• Track discoverability, citation, absorption, and business outcome as four separate metrics. A blended 'AI visibility score' hides exactly which stage is broken.

Sources