Claude for Excel Is Live. Will It Cite You?

On May 5, 2026, Anthropic held an invite-only briefing in New York and shipped three things in one event. Ten pre-built financial-services agents. Eight new data partnerships, including a Moody's native app and Verisk MCP connectors. And general availability for the Claude add-ins inside Microsoft Excel, PowerPoint, and Word.

The financial-services agents got most of the press coverage. The Microsoft 365 GA got the smaller headlines. That ordering is backwards for anyone running a B2B GEO program.

Until this week, "where does Claude cite me" was a question about claude.ai and a handful of partner integrations. Starting May 5, Claude is querying the customer's own Excel workbook, the slide deck open on their screen, and the Word memo their analyst is drafting. The citation pool inside that environment is dominated by files the customer already owns. Public-web GEO does not reach into a customer's tenant.

What Anthropic actually shipped on May 5

Three components, all GA the same day, all aimed at enterprise Microsoft 365 customers.

Excel, PowerPoint, and Word add-ins moved to GA

Claude add-ins for the three most-used Microsoft 365 productivity apps moved out of beta and are now generally available for enterprise customers. Bloomberg's coverage and Fortune's coverage of the NYC event both confirmed the timing. Microsoft is now distributing a competing frontier model inside its own flagship apps. That is not a small thing.

Ten pre-built financial-services agent templates

The agent templates are framed as "deployable in days, not months." They cover pitchbooks, earnings analysis, credit memos, underwriting, KYC screening, month-end close, statement audits, insurance claims, and analogous middle and back-office workflows. Jamie Dimon attended the event in person, which is the cleanest possible signal that JPMorgan Chase is in the early-customer pool.

Eight new data-platform partnerships

Verisk shipped standardized MCP connectors for Claude the same day. Moody's became a native app inside Claude with credit-rating and risk data on roughly 600 million companies. Dun & Bradstreet, Experian, GLG, Guidepoint, IBISWorld, Third Bridge, and Fiscal AI joined the partnership list. These sit alongside earlier integrations with LSEG, S&P Capital IQ, Morningstar, and PitchBook.

The product surface here is not "Claude with internet access." It is Claude with the customer's own data plus a stack of premium B2B data sources, running inside Excel.

Why the citation surface moves inside the customer's tenant

When a user asks Claude inside claude.ai, Claude retrieves from its training pool and increasingly from web search results. The citation rules for that environment are well documented. We covered them in how AI platforms choose which sources to cite.

Inside Excel, Word, or PowerPoint, the retrieval order is different. Claude looks at the open document first. Then any files the user has attached, uploaded, or referenced in a Microsoft 365 connector. Then the customer's tenant content the connector exposes. Then training-pool and web sources, in that order.

That re-orders the citation pool from the model out. The customer's own files are the dominant retrieval source for the kinds of questions Claude is asked inside Office. Vendor evaluations. Internal policy lookups. Comparative analyses based on the case studies a sales rep dropped into the customer's SharePoint nine months ago.

The citation surface a B2B SaaS brand needs to win in 2026 is no longer "the public web." It is "the customer's own document repository." That is a strategic shift that has not landed in most GEO programs yet.

Five reasons your sales collateral is suddenly a citation problem

If you sell B2B SaaS into a Microsoft 365 enterprise, five things change the moment Claude is GA inside their Office apps.

Reason #1: Your case studies are now the primary citation source. When a buyer's analyst asks Claude to compare vendors inside Excel, Claude cites whatever evidence is in the workbook or the customer's connected files. If your three best case studies live only on your website and were never sent as PDFs to your champion, they are not in the answer. The competitor whose collateral was uploaded to the customer's SharePoint two quarters ago is in the answer.

Reason #2: Sales reps now have a job they did not have before. Distribution of marketing-produced PDFs into customer file repositories used to be a courtesy. It is now a citation strategy. Reps who routinely email PDFs of case studies, ROI calculators, security overviews, and technical architecture diagrams to champions are seeding the customer's tenant with evidence Claude will cite later.

Reason #3: ROI calculators stop being lead magnets and become citation primitives. A standalone ROI spreadsheet that lives behind a form on your website is invisible to Claude inside the customer's Excel session. The same calculator, sent as an attached workbook the champion saved into a deal folder, becomes a structured artifact Claude can read, cite, and quote. The format that worked for marketing automation is the wrong format for citation inside the buyer's Office.

Reason #4: Stale collateral becomes a liability, not a quiet wallflower. Marketing has historically tolerated a long tail of outdated case studies and old comparison sheets sitting on Drive. Claude does not weight by recency the way a salesperson would. If the customer's SharePoint contains your 2024 architecture diagram and you have shipped two replatforms since, Claude will cite the old artifact and you will lose the technical evaluation on facts you fixed eighteen months ago.

Reason #5: Internal-doc citation is invisible to your existing GEO tools. Profound, Peec, Otterly, and Evertune all monitor Claude through API calls or claude.ai sessions. None of them simulate "Claude inside an Excel workbook the customer opened." That blind spot will close eventually, but for the next six to twelve months your visibility inside Office is unmeasured. We surveyed the available GEO tools recently and this is the cleanest gap in the category.

The five reasons compound. Each one alone is a tactical problem. Stacked together they describe a citation surface most B2B SaaS marketing leads have not allocated a single hour of budget to in 2026.

Traditional GEO vs internal-doc GEO

The difference is sharper if you write it out side by side.

Traditional GEO asks:

• •Where does my brand show up in the public LLM answer?
• •Which Reddit threads, G2 reviews, and YouTube videos drive my citations?
• •How do my competitors structure their pages to get cited?
• •What schema and HTML do I need to make my passages easy to extract?

Internal-doc GEO asks:

• •Which of my customers has my latest case study in their SharePoint or Drive?
• •When the customer's analyst asks Claude inside Excel for a vendor comparison, what files in their tenant does Claude pull from?
• •Are my ROI calculators sitting in the deal folder as attached workbooks, or stuck behind a form on my marketing site?
• •When my customer's compliance team queries Claude for evidence in a vendor risk review, do they have my SOC 2 report, my pen test summary, and my security architecture diagram?

Both questions matter. The first one defines your top-of-funnel exposure. The second one defines whether you survive the late-stage technical and procurement evaluation. Most B2B SaaS marketing programs have an answer for the first list and no plan at all for the second.

A four-step playbook to ship sales assets into customer environments

This is the operational change. The brands that will benefit from Claude in Office are the ones that treat customer-tenant distribution as a campaign, not a courtesy.

Step 1: Inventory the assets that should live inside customer tenants

Pick the ten artifacts a buyer's analyst would actually re-read during a vendor evaluation. Your three strongest case studies, your ROI workbook in Excel, your security architecture deck in PowerPoint, your implementation timeline as a Word doc, your pricing matrix, your reference list, your SOC 2 summary, and your two best comparison teardowns. Format each one in the native Office format. PDFs are a fallback. Native Excel and PowerPoint files are the strongest.

Step 2: Map every active deal to a tenant-distribution checklist

Walk your top twenty open opportunities and ask one question per artifact: is this file actually inside the customer's deal folder, SharePoint, or Drive? If not, who is responsible for putting it there, and by when? Sales operations should run this audit once per quarter at minimum. The CRM should track tenant-distribution status as a deal-stage requirement, not a nice-to-have.

Step 3: Refresh on a 90-day cycle and ask champions to replace stale files

Stale collateral inside a customer tenant is worse than no collateral, because Claude will cite it. Build a 90-day refresh cycle. When a case study, architecture diagram, or pricing sheet changes materially, have the AE send the updated file to the champion with a one-line note: "Replace the version in your shared folder, the architecture changed in Q1." Most champions will do it. The ones who will not are not actually champions.

Step 4: Pair internal-doc distribution with named off-site citation work

The customer's analyst does not always start inside Excel. Sometimes the path begins with a public Claude session, a Reddit thread, or a G2 review. Internal-doc distribution does not replace the external citation work we usually recommend. It layers on top of it. Brands that show up in both pools win the comparison most of the time. Brands that show up in only one pool win when the buyer's path matches that pool. Brands that show up in neither pool lose the deal before the demo.

This four-step sequence is not exotic. Every step has been done by a strong AE somewhere. The change is making it a marketing-led, repeatable program instead of an artifact of which reps happen to be diligent.

What changes about your GEO model

Three working assumptions need an update.

Public-web GEO is necessary but no longer sufficient. A clean Reddit presence, a strong G2 footprint, and structured passages on your blog still get you into the public Claude answer. None of that work reaches the customer's Excel session. Add a tenant-distribution layer to your GEO program or accept that you are competing in only half the citation surface.

Sales operations now has a marketing dependency it did not have before. The deliverable from marketing to sales used to be the asset itself. Going forward, the deliverable is the asset plus a distribution checklist plus a refresh cadence. SalesOps and RevOps leaders should expect this conversation in the next planning cycle. Marketing leaders should start the conversation themselves rather than waiting for it to arrive in QBR feedback.

Measurement gets harder before it gets easier. As of May 6, 2026, no GEO platform measures internal-doc citation behavior at scale. The right interim move is a manual audit. Pick five live deals, ask the customer's champion what is in their deal folder, and treat the answer as your baseline. By the end of 2026, expect at least one of Profound, Peec, Otterly, and Evertune to ship a tenant-grounded retrieval simulator. The brands that build the playbook now will have data to feed those simulators when they arrive. We covered the same dynamic for Gmail and Drive AI Overviews earlier this month. The pattern is consistent across both productivity stacks.

What we are watching next

Three signals on the calendar that will tell us how fast this matters.

Microsoft Copilot's response. Microsoft is now distributing a competitor frontier model inside its own apps. The internal politics of that decision will surface in some form. Watch for either a Copilot upgrade that closes the capability gap with Claude or a quiet repositioning of the Claude add-ins as a niche customer choice. Either reaction tells you how aggressive Microsoft is willing to be against Anthropic specifically.

Microsoft 365 add-in adoption telemetry. Microsoft does not publish add-in installation numbers, but third-party trackers and Reddit threads will. The number to watch is the percentage of Office Enterprise tenants that install the Claude add-in within 90 days. Anything above 10% confirms broad adoption. Below 5% and the citation surface is real but narrow.

Anthropic's next vertical agent library. The May 5 financial-services agents are the first vertical pack. The intel signal that matters is which vertical ships next: legal, healthcare, retail, or operations. That sequencing tells you which B2B SaaS categories face the steepest adjacent-vertical pressure on internal-doc citation in Q3 2026.

The bigger frame is the same one we made when Anthropic announced the $1.5B services JV with Blackstone, Hellman & Friedman, and Goldman Sachs on May 4. Anthropic is converting model strength into distribution at a pace none of the other labs have matched in this cadence. May 5 added a second vector: distribution into the apps every enterprise user already lives in. Public-web GEO is half the work. Tenant distribution is the other half.

FAQ

What did Anthropic announce for Microsoft 365 on May 5, 2026?

Anthropic moved Claude add-ins for Microsoft Excel, PowerPoint, and Word from beta into general availability for enterprise customers. The announcement happened at an invite-only briefing in New York alongside ten pre-built financial-services agents and eight new data-platform partnerships. Microsoft is now distributing Claude inside its own Office apps directly. Bloomberg, Fortune, and The Next Web all confirmed the timing.

Why does Claude in Excel matter for AI visibility?

When Claude runs inside Excel, PowerPoint, or Word, the retrieval order starts with the customer's own document tenant. Open files, attached PDFs, and connected SharePoint or Drive content take priority over the public web. That means the customer's own files determine which brands and evidence Claude cites. Sales collateral sitting in a customer's deal folder becomes a citation primitive. Sales collateral that lives only on your marketing site does not.

How is internal-doc citation different from regular GEO?

Regular GEO targets public-web sources: blog posts, Reddit threads, G2 reviews, YouTube videos, schema markup, comparison pages. Internal-doc GEO targets what is inside the customer's tenant: case studies, ROI calculators, security overviews, technical architecture diagrams, pricing sheets. Both pools matter. Most B2B SaaS programs have a strategy for the first one and no strategy at all for the second one as of May 2026.

Do GEO tools track Claude inside Microsoft 365?

Not yet. As of May 6, 2026, Profound, Peec, Otterly, and Evertune all monitor Claude through API calls or simulated claude.ai sessions. None of them simulate "Claude inside an Excel workbook the customer opened." That gap will close. In the interim, the right approach is a manual audit of which sales artifacts actually sit inside live-deal customer tenants.

What should B2B SaaS marketing teams do this quarter?

Four things. Inventory the ten artifacts a buyer's analyst would re-read during a vendor evaluation, in native Office formats. Map every active deal to a tenant-distribution checklist and treat distribution as a stage requirement. Run a 90-day refresh cycle so stale collateral inside customer tenants gets replaced. Layer this on top of the public-web citation work you are already doing, not instead of it.

The shorter version

Anthropic shipped Claude in Excel, PowerPoint, and Word at general availability on May 5, 2026, alongside ten financial-services agent templates and eight new B2B data partnerships. Inside Office, Claude's citation pool starts with the customer's own document tenant before reaching the public web. That moves the citation surface from the public LLM answer into the customer's SharePoint and Drive. Sales-collateral distribution into customer tenants is now the new link-building, and almost no B2B SaaS team has audited it once. The brands that build a tenant-distribution playbook this quarter will be measuring against a clean baseline when the rest of the category figures out what shifted.