For twenty years, the same SEO tools worked for a dentist and a database company.

Ahrefs, SEMrush, Moz – none of them needed an industry-specific version. A keyword research workflow built for plumbers worked, mechanically, for a fintech startup. You changed the keywords. The machinery stayed identical. The entire SEO industry consolidated into a handful of horizontal platforms because the optimization surface was the same regardless of what you sold.

Generative Engine Optimization will not consolidate that way. And understanding why requires looking at what actually changed underneath the optimization problem when buyers moved from Google to ChatGPT, Perplexity, and Google AI Overviews.

I have spent the past few years building a GEO platform specifically for cybersecurity and B2B SaaS. The deeper we got into the problem, the clearer it became that horizontal GEO is not a smaller version of the right answer. It is a structurally different answer to a problem that has a different shape than SEO ever did.

This article makes the architectural case for why GEO has to be vertical. Not as a positioning preference. As a property of how AI search actually works.

Why SEO Could Be Horizontal

To understand why GEO is different, you have to understand why SEO was able to commoditize into horizontal tooling in the first place.

SEO worked on a uniform substrate. Three things stayed constant across every industry:

One ranking algorithm. Google’s algorithm was the same algorithm whether you sold software or shoes. The 200-some ranking factors didn’t change by industry. The weighting shifted slightly for some query types, but the fundamental machinery – crawl, index, rank by relevance and authority – was universal.

One signal set. The inputs that mattered were the same everywhere: backlinks, keyword relevance, page speed, mobile-friendliness, content depth, internal linking. A tool that measured these signals worked for any website because every website was being judged on the same signals.

One results format. The output was always ten blue links. Whether you searched for “enterprise CIAM platform” or “best pizza near me,” you got a ranked list of pages. The optimization target was identical: be higher on that list.

Because the substrate was uniform, the optimization tooling could be uniform. Industry was a variable you plugged into the machine – the keywords, the competitors, the content topics. The machine itself never changed.

This is why one Ahrefs could serve every vertical. The horizontal model wasn’t laziness or a market accident. It was the correct architecture for a problem with a uniform substrate.

Why GEO’s Substrate Is Not Uniform

GEO breaks every one of those constants. The substrate underneath AI search is high-dimensional in a way that the substrate underneath traditional search never was.

There is no single algorithm. There are many, and they disagree.

ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews each select and cite sources through different pipelines. This is not a minor variation. Research across multiple independent studies in 2026 found that only about 11% of domains receive citations from both ChatGPT and Perplexity. The same query, run on two different models, produces almost entirely different sources.

The differences are structural. ChatGPT favors Wikipedia heavily. Perplexity pulls a far higher share from Reddit and community content. ChatGPT and Google AI Mode pull most of their LinkedIn citations from individual member posts, while the weighting differs across platforms. Claude relies more on training data with a knowledge cutoff; Perplexity crawls the live web at query time. Google AI Overviews correlates more closely with traditional search rankings than the others.

Optimizing for “AI visibility” as a single thing is like optimizing for “weather.” There is no aggregate weather. There is weather in specific places. Optimizing for an average that no real buyer experiences produces content that performs mediocre everywhere and excellent nowhere.

The answer depends on user context, not just the query.

Traditional search treated a query as a query. Two people typing “best CIAM platform” got the same ten links. The query was the whole input.

AI search incorporates context. The role of the person asking, the framing of their prompt, the conversation that preceded it, the way they describe their situation – all of it shapes the answer. A CISO asking ChatGPT to compare identity platforms for a regulated healthcare environment gets a fundamentally different answer than a startup founder asking the same nominal question with different framing. The models reason about the context, not just match the keywords.

This means the unit of optimization is not a keyword. It is a buyer-intent-in-context. And buyer intent in context is, by definition, specific to who the buyer is – which is specific to the vertical.

The retrieval depends on query fan-out, which is domain-specific.

When a buyer asks an AI system a question, the model doesn’t run one search. It decomposes the question into multiple sub-queries, retrieves sources for each, and synthesizes an answer. This is called query fan-out.

The sub-queries a model generates from “compare zero-trust vendors” are nothing like the sub-queries it generates from “compare email marketing tools.” The fan-out for a security question pulls in compliance frameworks, CVE databases, architecture patterns, and analyst coverage. The fan-out for a marketing question pulls in review sites, pricing pages, and feature comparisons. To optimize for the fan-out, you have to know what the fan-out looks like in your domain. There is no universal fan-out to optimize for.

What Authority Means Changes by Domain

The deepest reason GEO has to be vertical is that LLMs make domain-specific judgments about what counts as authoritative. And those judgments are exactly where horizontal tools are blind.

In traditional SEO, authority was largely domain-agnostic. A high domain rating, a strong backlink profile, topical relevance – these were measured the same way everywhere. A backlink from a high-authority site helped you regardless of industry.

LLMs evaluate authority differently, and the evaluation is contextual to the subject matter. Research in 2026 found that the average domain age of ChatGPT-cited sources is around 17 years, indicating established entities get preferential treatment. Domains with G2, Capterra, and Trustpilot profiles show roughly 3x higher citation probability for software queries. But which third-party validators matter depends entirely on the category. G2 matters for SaaS. It is irrelevant for industrial manufacturing, where ThomasNet and GlobalSpec carry the authority signal instead.

Here is the part horizontal tools miss. When an LLM answers a cybersecurity question, the sources it trusts are not the sources it trusts for a marketing question. For security, the model weights primary research, CVE writeups, framework documentation, vendor security disclosures, and analyst coverage. A generic GEO tool optimizing for “authoritative content” cannot tell a cybersecurity vendor that their problem is not their FAQ schema – it is that no model trusts them on zero-trust architecture because they have never published primary research that security practitioners cite.

That diagnosis requires knowing what authority looks like in security specifically. A horizontal tool applies a generic authority rubric and produces generic advice. The advice is not wrong, exactly. It is just optimizing for the average, which means it systematically misses the domain-specific signals that actually determine citation in any given vertical.

The Three Things a Vertical GEO Engine Has to Do Differently

If the substrate is high-dimensional and domain-specific, a GEO solution that actually works has to be vertical in three specific places. These are not features. They are the architecture.

1. Prompt generation has to be vertical.

The starting point of GEO is knowing which prompts matter – the actual questions buyers ask AI systems when researching solutions in your category. A horizontal keyword tool will not surface these, because they are not keywords. They are how a specific buyer, in a specific role, interrogates an AI assistant.

A cybersecurity buyer does not ask “best identity platform.” They ask something like “SOC 2 compliant CIAM with SCIM provisioning for a healthcare SaaS” or “passwordless authentication vendor that supports FIDO2 and has FedRAMP authorization.” Those prompts come from understanding how CISOs and security engineers actually think, not from keyword volume data. Generating the right seed prompts requires domain knowledge encoded into the system.

2. Response and data quality scoring has to be vertical.

Once you know the prompts, you have to mine the responses for quality – which sources the models cite, whether those citations are favorable, and what content is driving them. But judging whether a citation is good requires knowing what good means in the domain.

For a security buyer, a citation in a SANS Institute resource or a reference in a CVE analysis is a high-quality signal. A mention in a generic listicle is noise. A horizontal tool that counts all citations equally gives you a number that does not correlate with actual buying influence. A vertical engine weights the data by what carries authority with the specific audience – which means it needs a quality filter calibrated to the industry’s trust signals.

3. Content audit has to be vertical.

The final piece is auditing existing content against the right rubric. This is where the wrong approach does the most damage, because it produces confident, specific, wrong advice.

A horizontal audit tells a cybersecurity vendor to add FAQ schema, shorten paragraphs, and include comparison tables. All reasonable generic advice. None of it addresses the real gap, which might be that the vendor has no primary research, no named technical experts publishing under their own bylines, and no presence in the communities where security practitioners actually validate vendors. Auditing against a security-specific rubric surfaces the real problems. Auditing against a generic rubric surfaces generic problems and misses the ones that matter.

The Honest Tradeoff

Verticalization is not a free win. It is a tradeoff, and pretending otherwise would undermine the whole argument.

A vertical GEO engine is, by design, narrow. The system we built for cybersecurity and B2B SaaS is strong precisely because it encodes deep assumptions about those buyers – and those same assumptions make it the wrong tool for a direct-to-consumer retail brand or a hospitality business. The prompt generation, the authority weighting, the audit rubric are all calibrated to technical B2B buyers. Point that engine at an e-commerce catalog and the calibration works against you.

This is the real reason horizontal tools exist and will continue to exist. They serve the long tail of businesses that need basic AI visibility monitoring across every industry. For broad, shallow coverage, horizontal is the right architecture. For deep performance in a specific category where the buyers are sophisticated and the stakes per deal are high, vertical is the right architecture.

The mistake is assuming one of these is simply better. They solve different problems. The question for any given company is which problem they actually have. A company selling six-figure enterprise contracts to CISOs has a fundamentally different GEO problem than a company selling consumer subscriptions, and the tool that fits one will misfit the other.

What This Means If You’re Choosing a GEO Solution

The practical takeaway is a different evaluation question than most buyers are asking.

The common question is “which GEO tool has the best monitoring across the most AI platforms?” That is a reasonable question for broad visibility tracking. But if you sell into a specific, sophisticated vertical, it is the wrong primary question.

The better question is: does this solution understand my buyers well enough to generate the prompts they actually ask, judge citations by what carries authority with them, and audit my content against the standard my industry’s AI answers are actually being held to?

A tool can monitor ten AI platforms beautifully and still give you a generic optimization roadmap that moves you toward the average. For a horizontal business, the average is fine. For a vertical business selling to demanding buyers, the average is exactly what loses you the deal, because your competitors who got the domain-specific signals right are the ones the models cite.

SEO could be horizontal because the substrate was uniform. GEO’s substrate is not uniform – it varies by model, by context, by buyer, and by what authority means in each domain. The optimization has to match the shape of the problem. And the shape of the GEO problem is vertical.

Part two of this series examines what this looks like in practice for one vertical: how cybersecurity buyers actually use AI to research vendors, the prompt patterns that matter, and which sources LLMs trust in security. Part three provides a checklist for evaluating whether a GEO solution fits your vertical.

Related reading

• Top 5 GEO Tools of 2026 Compared: the current GEO platform landscape
• Future of AI 2026: where AI search is heading
• MCP, RAG, and ACP Comparison: the retrieval infrastructure underneath AI answers
• Grok AI Explained: understanding the AI model landscape
• AI Tokens Guide: how LLMs process and weight information

The post Why GEO Has to Be Vertical (When SEO Never Was) appeared first on Deepak Gupta's notebook.

*** This is a Security Bloggers Network syndicated blog from Deepak Gupta's notebook authored by Deepak Gupta. Read the original post at: https://guptadeepak.com/why-geo-has-to-be-vertical-when-seo-never-was/