GEO Vendors Are Selling Levers That Don’t Connect

TL;DR: The GEO and AEO vendor layer is recycling SEO-era frameworks — schema, heading hierarchy, chunking checklists — and applying them to systems that don’t work that way. Large language models parse text by reading text, not by reading structured data. Operators paying for ‘Technical GEO’ audits are buying the appearance of control, not actual visibility.

The Claim That Doesn’t Hold Up

A recent Semrush infographic identified four pillars of what it calls “Technical GEO.” The fourth pillar — schema, structured data, clean architecture — comes with a specific promise: it “ensures AI engines can parse and connect your content.”

That word, ensures, is the tell. Large language models are transformer-based architectures trained on the public web in all its disorder: forums, Wikipedia stubs, scraped product copy, machine-translated junk, half-formed sentences. The model reads token sequences. There is no parser inside looking for schema tags. There is no preference for FAQ markup. The model reads the words. That is the mechanism, full stop.

Schema has legitimate uses. It feeds rich results in classical search. It supports entity disambiguation in knowledge graphs. Voice assistants pull structured fields from it. These are well-defined functions inside specific systems. None of them describe how an LLM processes your prose. When a vendor says structured data ensures AI comprehension, the parsing layer they’re imagining does not exist.

Semrush isn’t alone. AirOps published a graphic claiming schema markup increases citation likelihood by 13%, that sequential heading tags double your chances, and that short paragraphs make content 49% more likely to appear in AI answers. The methodology trail leads back to AirOps’s own 2026 State of AI Search Report. AirOps citing AirOps on whether AirOps’s prescriptions work is not a methodology — it’s circular marketing.

What the Actual Research Found

There is a peer-reviewed paper on generative engine optimization. Aggarwal and co-authors, KDD 2024 — the closest thing to a legitimate academic foundation the GEO vendor layer has. It tested nine optimization methods against a 10,000-query benchmark.

What produced the largest visibility lifts? Adding citations from credible sources. Including statistics. Improving fluency. Writing cleaner prose with more evidence. The paper’s summary, in plain language: write content with more evidence in cleaner prose.

What did the paper test and find did not work? Keyword stuffing — the closest analogue to the SEO-era playbook vendors are now repackaging. Result: below baseline.

What is not in the paper’s list of nine methods? Schema. Structured data. FAQ markup. Heading hierarchy. Machine-readable formats. None of them. Because none of them are the optimization surface the paper was studying. The paper studied content-level interventions — what you put in the words, not metadata layered around the words.

The SaaS layer borrowed the acronym GEO from this paper. The findings stayed inside it.

Chunking Is Not Yours to Control

The chunking advice circulating in GEO content sounds technical, sits neatly in a flowchart, and gives content teams something concrete to put on Monday’s task list. It is also incoherent.

Chunking happens at retrieval time inside the engine. Perplexity, ChatGPT, and Gemini each run a retriever over candidate documents, split them according to their own configurations — length, overlap, embedding model, sometimes semantic boundaries — and feed the top results into the model’s context. Those configurations belong to the engine. They get tuned on schedules no publisher has access to.

When a vendor says “optimize for chunk-level retrieval,” what they’re actually recommending is good writing. Short, self-contained paragraphs. Clear definitions near section openings. Logical internal structure. These are information architecture and technical writing — disciplines that have existed since long before the transformer was invented. They are not a new technical layer you need a dashboard to implement.

A more honest version of the pitch would be: hire a competent writer. That sentence doesn’t fit on a SaaS pricing page, which is why it doesn’t appear on one.

What This Means for Performance Marketing Operators

If you’re running budgets of $10K or more per month in forex acquisition, iGaming player acquisition, crypto lead generation, legal intake marketing, or CDL driver recruitment, the GEO vendor pitch is aimed directly at your CMO’s inbox. It arrives with percentage lifts, pillar frameworks, and a dashboard that makes optimization look like a scheduled activity.

Here’s what to pressure-test before buying in. Ask any vendor claiming schema improvements will lift AI citations to show you the counterfactual. What was the control? What was the reproducibility window? How does a citation lift get measured across a non-deterministic system that returns different outputs to different users for the same query? These aren’t hostile questions — they’re the basic methodology questions that separate billable activity from actual signal.

The operators who will hold ground in AI-influenced search are not the ones running schema audits. They are the ones who have product-driven content that actually contains evidence, citations, and useful information. That content gets read because LLMs were built to read whatever is there — and if what is there is genuinely good, the architecture doesn’t care that you didn’t add FAQ markup.

A full marketing audit should be asking whether your content strategy is built around quality signals the model can actually read, not around metadata frameworks the model doesn’t process. Pair that with honest paid performance management that doesn’t depend on organic AI citation as a primary acquisition lever, and you’re operating with a more defensible model than the GEO dashboard can offer.

The Industry Is Selling Comfort, Not Control

The honest version of the current AI visibility situation goes something like this: we are operating with reduced visibility into a system that does not expose its mechanics, returns different outputs to different people for the same query, and is changing month by month. We can keep doing what has always been the work — writing well, being useful, building authority. The deterministic dashboard we had before is not coming back.

That sentence is unsayable in a marketing meeting. It admits the lever isn’t connected. It tells leadership the budget line they approved doesn’t have a corresponding action with a traceable output. So the SaaS layer fills the gap. It manufactures levers. Pillars, schema audits, chunking checklists, citation likelihood scores. Reportable activity. Defensible expenditure. Something to say in the meeting.

The GEO cycle will follow the same shape as spinning, programmatic doorway pages, and every other controllable-input-dressed-as-discipline the SEO industry has run through. A brief period where the prescriptions appear to work because some underlying quality existed. Then degradation as volume scales, qualitative review collapses, and engines adjust. Then the next vendor selling the cleanup.

Tools like precision audience targeting and AI-assisted lead qualification operate on systems with observable, attributable outputs. That is a different category of tool than a dashboard measuring citation likelihood against a non-deterministic engine. Operators should be clear on which category they’re buying.

What Actually Accrues an Advantage

The KDD 2024 paper is clear on where the content-level gains come from: evidence, citations from credible sources, statistics, fluency. Write content with more signal in cleaner prose. That is the finding. The advantage will go to operators who produce content worth reading — not content templated around chunking heuristics that don’t exist inside the model.

The architecture was built to handle messy, unstructured input. It was trained on the public web because the public web is unstructured and still contains enormous amounts of useful language. The mess was never the problem GEO vendors are selling a solution to. Operators who stay focused on substance over metadata compliance, and who question vendor claims that don’t survive basic methodology review, are better positioned than those who subscribe to the dashboard and let it quietly replace the actual problem with the part of the problem it can measure.

Originally reported by Search Engine Journal, May 2026.