Windsor.ai published twelve “Connect to Claude” posts in two weeks. Every one follows the same formula: pick a platform, OAuth it, install Windsor’s native Claude app, and ask questions.
Taboola. Braze. Microsoft Dynamics 365. Spotify Ads. ConvertKit. MailerLite. MediaGo. Amazon Vendor Central. CoinMarketCap. Yahoo Finance. Vibe.co. Airtable. Twelve platforms, twelve posts, one sentence repeated twelve ways: connect your data, then ask Claude.
That isn’t a criticism — it’s the smartest SEO flywheel in the analytics space right now. Every connector becomes a landing page, every landing page captures a long-tail search for “how to connect X to Claude,” and the full library is already deeper than the twelve that shipped this fortnight. At this velocity, Windsor will own a “connect to Claude” page for every platform on the internet before the quarter ends.
But here’s what’s missing. Of the twelve posts — of the whole library — not one is the post a regulated agency actually needs. Windsor taught Claude to read twelve platforms’ data. Nobody has taught an AI to run a regulated campaign and prove it was compliant.
That’s the thirteenth post. It’s the one Windsor can’t write, because it isn’t a content gap. It’s an architecture gap.
The Blitz, Quantified
Let’s be precise about what Windsor is doing, because it deserves to be understood before it’s countered.
Every “How to Connect [Platform] to Claude” post follows the same two-step template: connect the source (no code, OAuth), install Windsor’s native Claude app, then ask. “What did we spend on Spotify Ads last month?” “Compare Taboola CPMs across campaigns.” “Which Braze segment drove the most conversions?” Claude queries the live data through Windsor’s pipeline and answers in plain English.
That’s the product promise in one line — connect any data source, analyze it anywhere, get insights with AI — and it’s priced to spread: MCP is included in every plan from $19 a month, working with Claude, ChatGPT, Copilot, Gemini, and Perplexity.
And the content strategy is worth studying, because it’s the rare SEO flywheel where the content is the product documentation. Each connector post is a landing page. Each landing page targets a real search — people genuinely type “how to connect Taboola to Claude” into Google. Windsor captures that demand one platform at a time, and the “how to” format means every post also earns the AI Overview citations and People Also Ask placements a product page never would. Twelve posts in two weeks is the cadence; a library that deep is the moat being built.
Fair play. This is teaching Claude to read — at scale.
What Every One of the Twelve Posts Leaves Out
Read any of the twelve and notice what none of them contains. Not one mentions:
- who changed a campaign,
- who approved it,
- when it went live,
- whether the disclaimer was attached,
- which jurisdiction’s rules it had to satisfy.
The “connect → ask” model has one verb: read. Windsor’s Claude can tell you what happened to your data. It cannot tell you who did what to your campaigns — because it doesn’t record that. It moves data; it doesn’t record accountability.
For a marketing analyst pulling a weekly report, that’s completely fine. For a regulated agency running financial promotions, it’s the difference between an insight and evidence. And the FCA is now asking for evidence.
The Post Windsor Needed to Write
Here’s the one they should have published instead of “How to Connect Taboola to Claude”:
“How to Prove Your Claude-Connected Campaigns Were Compliant.”
It doesn’t exist — and not for lack of writers, because Windsor shipped six posts in a single day. It doesn’t exist because you can’t write it with a data connector. To prove a Claude-connected campaign was compliant, you need three things a connector fundamentally doesn’t provide.
An immutable, actor-attributed history. A Claude chat log is not an audit trail. It’s mutable, it isn’t tied to an approval chain, and it doesn’t export as a structured compliance document. When the FCA asks “who changed this FX CFD campaign in March, and under what sign-off?”, “ask Claude about the March data” is not an answer. The FCA isn’t asking what the campaign spent. It’s asking who touched it, when, and with what authorisation.
Jurisdiction context. The connectors read platforms. None of them knows which regulator governs the campaign behind the data. A forex promotion that’s compliant in the UK isn’t automatically compliant under MiCA. A crypto creative that passes review in one market can violate another’s promotion rules. Reading the spend tells you nothing about whether the promotion itself was legal where it ran.
Operations, not just queries. Connecting data to Claude answers questions. Operating regulated PPC means catching an anomaly before the regulator finds it, enforcing budget controls with a timestamped ledger, and failing over to a backup Business Manager when Meta bans your primary. A connector reads. Compliance requires acting — and recording the act.
Reading vs. Operating
There’s a clean way to state the difference, because it’s the whole argument.
Windsor taught Claude to read marketing data. Ott taught AI to operate regulated campaigns — and left an immutable record of every operation.
Both use MCP. Both market “AI for ads.” That’s where the similarity ends, because they answer different questions:
- Windsor’s Claude answers: “What did we spend?”
- Ott’s AI answers: “Who changed this campaign, when, under what sign-off, and can we prove it to the FCA?”
A data connector reads ad spend. A compliance-aware operations layer reads regulatory exposure. Different verbs. Different buyers. Different categories.
| Capability | Windsor + Claude | Ott |
|---|---|---|
| Ask AI about ad data | ✅ Dozens of connectors, MCP from $19 | ✅ Operations MCP (26 tools) |
| Record who changed what, when, with what approval | ❌ Chat log (not evidence) | ✅ Immutable Activity Logging |
| Prove compliance in ~90 seconds | ❌ No export | ✅ Regulator-ready audit trail export |
| Jurisdiction context (FCA, MiCA, ESMA, Alberta) | ❌ None | ✅ 3-level hierarchy with jurisdiction tags |
| Catch anomalies before the regulator |
The table makes the structural argument visible. Windsor built the read layer — and built it well. Ott built the proof layer. You can bolt another sixty connectors onto the read layer and never touch the proof layer, because proof isn’t a connector. It’s an architecture.
Why the Timing Makes This the Only Question That Matters
This distinction stopped being a nice-to-have about six weeks ago.
On June 24, FCA Chief Executive Nikhil Rathi told a techUK audience that “legislation will never keep up” with AI — and that 80%+ of financial services firms are already adopting it. His conclusion: “Accountability for regulated activities and outcomes must remain clear.”
Translation: the FCA isn’t going to write specific rules for AI advertising. It’s going to demand that regulated firms prove their AI operations followed the rules that already exist.
Then it acted on that position, fast. A trial date for illegal FX CFD social-media promotions (July 30). Information requests to roughly 900 firms, with registration applications “closely scrutinised” (August 7). A machine-readable Handbook API (August 6). The warning phase ended; the enforcement phase began.
Against that backdrop, “connect your data to Claude” is a productivity feature. “Prove your Claude-connected campaign was compliant” is a regulatory requirement. Windsor is winning the first race — brilliantly. Nobody, Windsor included, is running the second. That’s the gap, and it’s the entire regulated vertical.
What Windsor Got Right
None of this is a knock on Windsor’s execution. Twelve posts in two weeks is real velocity. MCP in every plan from $19 is the most aggressive accessibility play in the market. UBS as an enterprise customer means Windsor has cleared security reviews most horizontal platforms never face. And the connector-for-SEO flywheel is a genuinely smart growth strategy — the live blog index confirms it’s working.
Windsor is a capable platform executing a coherent plan. The question isn’t whether Windsor is good at what it does. It is. The question is whether “what Windsor does” — teaching Claude to read data — is what a regulated agency needs when the regulator comes asking for proof.
It isn’t. And the library proves it, not because the posts are bad, but because they’re all the same shape — and that shape has no room for accountability.
The Self-Assessment
Forget platform comparisons for a second. Three questions for the tool you’re running regulated campaigns through right now:
- Can it produce an immutable, timestamped, actor-attributed record of every change to every financial promotion you’ve run?
- Can it organise that record by jurisdiction — FCA, MiCA, ESMA, or whichever regulator comes calling?
- Can it export that record, regulator-ready, in under two minutes?
If the answer to any of these is no, you’re running the FCA’s enforcement escalation on a platform built to answer “what did we spend?” — not “can we prove we were compliant?”
Connecting your data to Claude tells you what happened to your spend. Connecting your operations to compliance-aware infrastructure tells you whether you can prove it was legal. Both are useful. Only one is mandatory — and the FCA just reminded everyone which one.
See how Ott’s compliance-aware operations work → or Start your free trial →