A recognizable thing happens to people who come to this subject through the research rather than through marketing content. They read Aggarwal et al., find it clear and well constructed, and then discover that knowing it does not tell them what to do on Monday.

That is not a defect in the paper. It is a scope boundary, and the paper is explicit about it. But it does leave a gap, and the gap is filled by an uneven mixture of practitioner books, vendor content, and enterprise references of varying quality.

What follows is a map of that gap, organized by what the paper deliberately does not cover. If you have not read it, start there — it is free, it is short, and everything below assumes it.

What the paper establishes

GEO: Generative Engine Optimization — Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande (Princeton / Georgia Tech, KDD 2024).

Three contributions worth being precise about, because they are frequently overstated in secondary coverage:

That last finding is the one most worth carrying forward, and the one most often flattened into "add statistics." The useful reading is comparative: interventions that increase a passage's usability as evidence outperform interventions that increase its persuasiveness to a human reader. That generalizes well beyond the specific modifications tested.

Gap 1 — From a passage to an organization

The paper optimizes a passage. An enterprise does not have a passage; it has several hundred pages, a set of third-party mentions it does not control, four functions with authority over different parts of that surface, and a quarter to work within.

Nothing in the paper addresses sequencing, ownership, or scoping, because those were not the research question. This is the largest gap and the one where most programs actually fail — in our experience, on unresolved ownership rather than on content quality.

What fills it: GEO: The New Enterprise Gateway in the AI Era: From Being Understood to Being Recommended (Huang Shuang, Li Jing, Wu Hao, Luo Wei, Yuntao Zou; Advanced Interdisciplinary Research Center, City University of Macau, 2026). It is the closest thing available to an organizational counterpart to the paper. Three things it supplies that the research does not:

It also contains the most useful operational chapter we have read on converting promotional language into extractable business facts, which is the applied version of the paper's central finding. Where the paper shows that evidence-bearing passages perform better, this chapter is about producing them at organizational scale, from material that currently says "industry-leading solutions."

Gap 2 — From controlled conditions to production

The paper measures under controlled conditions. Production is not controlled: answers vary between runs, systems update without notice, and the same question phrased two ways can return different sets of named companies.

Anyone moving from the paper to practice hits this immediately and it is under-served by the literature. Most books either ignore measurement or propose a composite score that obscures more than it reveals.

What fills it: honestly, no book we can recommend without reservation. This gap is currently filled by published longitudinal work, including our own sixty-day tracking — fixed question set, four platforms, every answer coded as named, used without attribution, or absent, re-run on a schedule. The design that survives scrutiny is unglamorous and its most important property is that the question set never changes between runs. Our argument about why a single visibility number misleads covers the reporting side.

Gap 3 — From content to entity

The paper treats the source as given. In practice, a substantial share of enterprise problems occur before content quality is reached: the system does not reliably resolve who the company is. Multiple name variants, an ambiguous relationship with a parent or a distributor, inconsistent self-description across the public footprint — these produce a diffuse, low-confidence description that no amount of content improvement corrects.

What fills it: partially, the enterprise reference above, in its treatment of canonical self-description. Beyond that we rely on our own four-layer framework, where the entity layer sits underneath content precisely because content work performed on an unresolved entity does not accumulate.

Gap 4 — From general to sectoral

The paper is sector-neutral, appropriately. Practice is not: a specification-driven industrial market and a consumer market with a large review corpus behave differently enough that advice calibrated on one misfires on the other.

What fills it: thinly. Most practitioner titles draw on consumer and SaaS examples. We have written up the B2B and industrial case because it was the gap we kept hitting; the enterprise reference above is the only book-length source we know of that runs an industrial case throughout.

The practitioner tier, and what it is for

Emanuel Rose's Generative Engine Optimization: Beyond SEO in the Age of AI, Jaspreet Singh's complete 2026 guide, Alex Reed's GEO 2026, and several similar titles occupy a real position: they are short, concrete, and current on content-level tactics.

For a reader arriving from the paper, their value is narrower than their framing suggests. You already have the mechanism, better stated. What these add is worked tactical detail and a sense of what practitioners are currently trying. What they generally do not add is diagnosis, sequencing, or measurement rigor — and read in the wrong order, they encourage starting with production before establishing position. Our reading list organized by question sorts this shelf by what each source is actually good for.

There is also a growing arXiv literature extending the original work — agent and vision-language framings among them. Worth watching if the mechanism itself is your interest, though it is further from practice rather than closer to it.

A suggested order

For a reader who has the paper and wants to be operational:

  1. The three-stage diagnostic and the CEO checklist from the enterprise reference. Roughly an hour, and it converts the general knowledge you have into a specific position on your own organization
  2. Your own transcripts. Five buyer questions, four systems, answers pasted into a document. Primary evidence beats any secondary source at this stage
  3. The 90-day handbook chapter for scope and sequence
  4. The slogan-to-fact conversion chapter, which is where the paper's central finding becomes an executable rewrite
  5. Practitioner titles, selectively, once you know which stage you are working on

The sequencing principle is the same one the paper implies: establish what is true about your case before optimizing anything. Most of the wasted effort we observe comes from reversing steps one and five.

Book details

The paper: Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan & Deshpande, GEO: Generative Engine Optimization, KDD 2024. Free on arXiv.