Why Vertical AI Now Gets Cited More Than Horizontal

On May 12, 2026, Anthropic launched Claude For Legal. The launch had 12 practice-area plugins, a Microsoft 365 cross-app context-carrying agent, and named Big Law customers including Freshfields, Quinn Emanuel, Holland and Knight, and Crosby Legal.

Buried inside the announcement was a number that matters more than any of the practice-area plugins.

Anthropic disclosed that legal had become the #1 power-user job function inside Claude Cowork, with over 3x the usage of any other function. That is the first per-vertical usage breakdown Anthropic has ever shared from its enterprise product. It is also the strongest single argument we have for why vertical AI visibility now wins more citations than horizontal AI visibility.

If you are running a B2B SaaS marketing program in 2026 and your GEO plan is "get cited across every industry we serve," the next two quarters are going to feel rough. Generic horizontal GEO is being commoditized. Vertical GEO is where the citation share is now compounding.

What Anthropic actually said

The press coverage of Claude For Legal focused on the practice-area plugins. The more useful signal sits inside the Anthropic announcement itself, corroborated by the Law.com coverage and Fortune's writeup.

Three points to memorize.

Legal usage inside Claude Cowork is 3x any other power-user function. Anthropic is bundling 20+ legal tech integrations (Harvey, Westlaw, LexisNexis, Legora, DocuSign, Box, Free Law Project, Thomson Reuters) into Claude, not building a separate vertical product. Pricing is unchanged: Claude For Legal is available to every paying Claude customer at the same Business and Enterprise tier rates.

Read those three points together and the implication is clear. Anthropic is not betting that horizontal Claude wins the legal vertical by accident. It is shipping a vertical-shaped surface inside the horizontal product because that is where the highest-density usage actually concentrates.

If the AI labs themselves are building vertical density, the citation pools they retrieve from are going to weight vertical sources more heavily. That is the part that flows directly into B2B GEO strategy.

Why horizontal GEO is being commoditized

The horizontal GEO playbook, written in 2024, looked like this. Build an answer-engine-friendly site. Add FAQ schema. Get cited on Reddit, LinkedIn, G2, Wikipedia. Track citation share across ChatGPT, Claude, Perplexity, and Gemini. Repeat.

It worked when AI search was new and the citation pool was narrow. It is now table stakes.

Three things changed in the past nine months.

Reason #1: The lab products themselves went vertical

Anthropic shipped Claude For Legal in May 2026, Claude for Wall Street earlier in the same month, and Anthropic Finance Agents in late April. OpenAI launched the Deployment Company with McKinsey, Bain, and Capgemini as equity investors on May 11. SAP and Anthropic announced at SAP Sapphire that Claude becomes the primary reasoning layer across hundreds of thousands of SAP customers in finance, HR, procurement, and supply chain. None of these are horizontal pitches. Every one is a vertical wedge.

When the lab itself organizes around verticals, the retrieval pipeline starts to favor vertical-aligned sources for vertical-aligned queries. A model that knows "legal" is its top usage cohort is going to retrieve more aggressively from legal-tech publications, bar-association journals, and legal vendor comparison sites when answering legal questions.

Reason #2: B2B buyers ask vertical-shaped questions

The horizontal query "what is the best CRM" still happens, but it is now a small minority of category prompts. The dominant query shape in 2026 is vertical: "what is the best CRM for a mid-market legal services firm," "what CRM does a 200-person life sciences sales team typically use," "best procurement platform for a SAP-shop manufacturer."

The vertical shape changes the retrieval pool. AI engines hunt for sources that explicitly match the qualifier. A horizontal landing page about your CRM is one match. A case study with a Big Law firm, a comparison page tagged for life sciences, or a SAP-integration documentation page beats the horizontal landing page on retrieval relevance every time.

Reason #3: Vertical trust signals stack

In legal, the third-party trust stack is bar associations, AmLaw 100 mentions, Above the Law coverage, and integrations with Westlaw or LexisNexis. In life sciences, it is FDA submissions, journal citations, and partnerships with named teaching hospitals. In financial services, it is regulatory filings, AICPA bodies, and named buy-side firms.

These stacks are unrelated to the horizontal trust stack (Reddit, G2, Wikipedia, LinkedIn). A brand can be perfectly optimized on the horizontal stack and still invisible inside the vertical stack. AI engines have started weighting vertical stacks heavily for vertical queries because the Yext source-controllability research and Muck Rack's Generative Pulse findings both converge on the same point: earned media drives 84% of citations, and vertical earned media is what gets retrieved for vertical questions.

Reason #4: Horizontal citation share is now a thin signal

We see this in every B2B SaaS portfolio audit. A client that has been working on GEO for a year often hits 30 to 45% horizontal citation share for its category prompts. That number plateaus. Pushing it from 40% to 50% costs three to five times what pushing it from 10% to 30% cost. Diminishing returns hit early.

The vertical citation share for the same client is usually under 5% when we start. Pushing it from 5% to 30% on a single vertical takes one quarter of focused work and produces more pipeline impact than the same investment in horizontal optimization. The math has flipped.

Reason #5: Procurement is increasingly vertical-aware

Inside SAP-shop accounts, procurement queries now route through Joule and Claude. Inside Microsoft 365 shops, Copilot pulls from Word, Outlook, Excel, and PowerPoint context plus the public web. Inside law firms running Claude with Westlaw, the retrieval window is vertical-shaped from the first prompt.

That is a new procurement surface. It did not exist eighteen months ago. It rewards vendors who show up in vertical-aligned third-party sources and punishes vendors who only show up in horizontal ones.

Generic GEO got you cited in the categories you serve. Vertical GEO gets you cited in the procurement queries that close deals.

The honest contrast: horizontal vs vertical thinking

The cleanest way to see the gap is to write the two playbooks side by side.

Horizontal GEO asks:

• •Are we cited in our category on ChatGPT, Claude, Gemini, Perplexity?
• •Do we have G2 reviews above 4.0?
• •Are we on the Wikipedia page for our category?
• •Did we publish a "best of" post for the category this quarter?

Vertical GEO asks:

• •Are we cited when a buyer adds an industry qualifier (legal, life sciences, finance, manufacturing, SAP-shop)?
• •Do we have a named case study with at least one logo from the vertical's reference list?
• •Are we mentioned in the vertical's trade publications (Law.com, AmLaw 100, BioPharma Dive, AICPA outlets)?
• •Do we integrate with the vertical's mandatory stack (Westlaw, EHR, ERP, SAP, Salesforce Industries)?

A horizontal-only program can pass the first list and fail every item on the second. The buyer who matters in 2026 is asking the second list, not the first.

What this means inside Claude Cowork specifically

The Claude Cowork disclosure matters because Cowork is the surface where B2B usage actually concentrates. Per Anthropic's own breakdown, legal hits 3x the next-highest function inside Cowork. The next-highest functions, by implication, are consulting, finance, and software engineering, in some order.

For a B2B SaaS targeting any of those four cohorts, the implication is direct. Your buyer is inside Cowork. The retrieval pipeline weights vertical sources. Generic content competes for a thin slice of citations; vertical content competes for the dominant slice.

We have written about adjacent dynamics in why Claude cites older content than ChatGPT and the B2B SaaS citation concentration data. The vertical density angle is the missing piece. Citation concentration plus vertical weighting plus per-Cowork-function density compound into a single insight: the brands that win 2026 GEO will be vertical-shaped at every layer.

The five-play vertical GEO pipeline

If the diagnostic is "your GEO program is horizontal in a world where buyers ask vertical questions," the prescription is a five-play pipeline that converts horizontal content programs into vertical citation share.

This is the playbook we run for B2B SaaS clients in legal-tech, fintech, healthtech, and SAP-adjacent verticals. Five plays, each with a clear output.

Play #1: Re-map your prompt set from horizontal to vertical

Most clients audit citation share against 20 to 50 category-level prompts. Replace half of them with vertical-qualified versions. Instead of "best [category] software," track "best [category] software for [industry]" and "[category] platform that integrates with [vertical-mandatory stack]." For most B2B SaaS, twelve to fifteen vertical prompts per priority industry is the right baseline.

Play #2: Build a vertical evidence stack on the site

Add a per-vertical section to your site that does three things in plain text. Names the industry. Lists the integrations that matter inside that industry (Westlaw, Epic, SAP, Salesforce Industries, NetSuite). Shows at least one customer logo from the industry's reference set. This becomes the page AI retrieves when a buyer adds an industry qualifier.

Play #3: Win the vertical's third-party sources

Identify the three to five trade publications, association directories, and review aggregators that the vertical actually trusts. For legal, that is Law.com, ALM Vanguard, ABA Journal, and Above the Law. For finance, it is American Banker, Risk.net, and Finextra. Pitch one to two earned media placements per quarter in those outlets. The vertical-shaped citation share moves on that input.

Play #4: Cite vertical benchmarks inside your own content

Most horizontal GEO content cites horizontal benchmarks (5W, Profound, Yext). Vertical GEO content cites vertical benchmarks (industry analyst reports, regulator filings, named association studies). When you write for a vertical, ground every claim in a source the vertical recognizes. A piece of legal-tech content citing the ABA Journal will outrank the same piece citing Yext for legal queries, because the retrieval pipeline scores source-relevance to query.

Play #5: Track per-vertical citation share weekly

Stop reporting a single GEO dashboard. Run one per priority vertical. Most B2B SaaS will have three to five priority verticals; the dashboard count is small. The weekly cadence is what matters. Movement on a vertical dashboard inside four weeks is the leading indicator the pipeline is working. A flat dashboard at week eight is the signal to reallocate.

The full pipeline takes 10 to 14 weeks to set up and runs as a continuous operation after that. The output is a defensible 20 to 35 percentage point lift in vertical citation share, which is roughly the same range our portfolio data shows for clients running the pipeline through one full cycle.

Where this fits in the broader 2026 GEO mix

Vertical GEO does not replace horizontal GEO. Both still matter. The reallocation is about where the marginal dollar goes.

The Trustpilot study published May 12, 2026 showed brands with 80+ active reviews get cited in 75% of AI answers versus 1% for brands with no review presence. That is a horizontal lever and it is still the largest single citation lift number of the year.

The Muck Rack May 2026 Generative Pulse study showed earned media drives 84% of AI citations across three quarterly cuts. Also horizontal.

The 5W May 2026 70% to 20% overlap data showed Google rankings no longer predict AI citation share. Horizontal again.

What none of those studies do is segment by industry. The horizontal data is the floor. The vertical work is what wins the buyer who adds an industry qualifier to their prompt. Both layers compound. Skipping the vertical layer in 2026 is the equivalent of skipping the off-site distribution layer in 2024.

For a longer treatment of the underlying citation framework see our B2B AI visibility audit playbook and the how to measure GEO primer.

FAQ

What did Anthropic disclose about vertical usage inside Claude Cowork?

In its May 12, 2026 Claude For Legal announcement, Anthropic disclosed that legal had become the #1 power-user job function inside Claude Cowork, with over 3x the usage of any other function. This is the first per-vertical usage breakdown Anthropic has ever shared from the Cowork product. It points at a concentration pattern where one vertical drives more than triple the engagement of any other inside an otherwise horizontal product.

Why does vertical AI visibility now beat horizontal?

Three forces stack. The AI labs themselves are shipping vertical products (Claude For Legal, Claude for Wall Street, SAP Joule integration), which pushes the retrieval pipelines to favor vertical sources. B2B buyers increasingly add industry qualifiers to their prompts, which changes which sources get retrieved. Vertical trust signals (trade publications, mandatory integrations, named industry logos) sit on a different stack from horizontal trust signals (Reddit, G2, Wikipedia), and the vertical stack is where the procurement queries land.

How do I know if my GEO program is too horizontal?

Three quick tests. Audit your top 20 category prompts. If fewer than 30% include an industry qualifier, your prompt set is too horizontal. Check whether your site has per-vertical evidence pages with named integrations, named customer logos, and at least one named case study per vertical. Check whether your earned media plan for the quarter includes at least one placement in a vertical trade publication for each priority industry. If any of the three answers is no, the program is horizontal.

Does vertical GEO replace horizontal GEO?

No. Horizontal levers (review platforms, off-site distribution, refresh cadence, passage extraction) still drive the citation floor. Vertical GEO is what wins industry-qualified procurement queries on top of that floor. The right allocation in Q3 2026 is roughly 60 to 70% horizontal maintenance and 30 to 40% vertical pipeline build for most B2B SaaS portfolios.

Which verticals should B2B SaaS prioritize first?

Pick the two to three verticals where your existing pipeline already concentrates. Most B2B SaaS portfolios have a long tail of industries but 70% of ARR concentrated in three or four. Run the vertical playbook on those first. Adding a fourth or fifth vertical pipeline is a Q4 decision once the first two are measurable.

The shorter version

Anthropic's May 12 disclosure that legal is the #1 power-user function inside Claude Cowork at 3x any other function is the first per-vertical usage breakdown the company has shared. It is also the cleanest single argument that vertical AI visibility now beats horizontal. The AI labs are shipping vertical products. B2B buyers ask vertical-qualified questions. Vertical trust stacks are unrelated to horizontal trust stacks.

The five-play vertical pipeline (vertical prompt set, per-vertical evidence stack, vertical earned media, vertical-cited benchmarks, per-vertical dashboards) takes 10 to 14 weeks to set up and produces a 20 to 35 percentage point lift in vertical citation share inside one cycle. Horizontal GEO still matters for the citation floor. Vertical GEO is where the marginal dollar wins the buyer who adds an industry qualifier to their prompt. That is roughly half of B2B SaaS procurement queries in 2026.