Don't trust generic AI output for your emails

We surveyed a ton of email professionals this year, and one pair of numbers stuck out. 64% use AI somewhere in their workflow. Only 1.1% actually start a new email from AI. Ouch. Almost everyone still kicks off a campaign the old way: they open the last email that worked and copy it.

If you ask me, that gap is the interesting part. People trust AI to draft a line or spark an idea, but they don't trust it with the thing that ships.

We've watched this happen a hundred times, and we've done it ourselves. You ask an AI for an email, you get a decent one in ten seconds. Then you spend the afternoon fixing a blue that isn't your blue, rewriting the parts that sound like every other brand, and moving it into the tool that actually sends. The generating process is quick, but everything after it is the job that you were really looking to solve.

In other words, the problem in email AI was never generation because the important part is what happens next.

Why generic AI stalls in email

There are a few things that get in the way. The RGE annual survey named two of them (below); the third shows up the moment an email hits a real inbox.

The first is a lack of trust in the output. The fears people cited most were brand voice landing wrong and the AI inventing something. In email those aren't small risks. We all know that a bad send doesn't reach ten people, it reaches your whole list. So when we asked what would make AI genuinely useful, the top answers were confidence that the output is safe to send (53.5%) and AI that fits the workflow they already use (40%), not speed or intelligence.

The second is the code itself. Ask a general model to write the HTML and you'll get something that looks fine in preview and breaks in the wild (e.g., cramped on mobile, unreadable in dark mode, different from one inbox to the next). Email HTML is a fussy, unforgiving craft, and a model freehanding it doesn't clear that bar reliably. As seen in the tests, a draft can read or display perfectly in ChatGPT and still be unsendable or break in an inbox.

The third is plumbing. As one email maker put it: "AI tools aren't natively integrated into our ESP, so everything requires manual transfer." That's the copy-paste tax that feels minor each time but eats the time you thought you saved. Generic output is disconnected by design becuase it doesn't know your brand, doesn't remember what you've approved, and has no route to the system that sends.

Strip it down and a sendable email needs three things a generic draft doesn't have:

1. It has to be on-brand, built from your real approved work rather than a guess.

2. It has to be approved, with a real, human person who signed off.

3. And it has to be exportable, wired to the ESP that delivers it.

None of those are creation problems. They live after the draft exists, which is the part generic AI skips.

What the RGE Studio MCP does

That's the gap we built RGE Studio MCP to close. It's in beta and usable today.

Quick definition, since the acronym does some lifting. MCP (Model Context Protocol) is an open standard that lets an AI assistant use an outside tool directly. The RGE Studio MCP surfaces the things you already do inside RGE Studio (like browsing your work, opening it up, building and saving an email, exporting it) as tools any assistant can call. So Claude, ChatGPT, or your own agent can work with your real designs without you leaving the AI tool you're in.

Concretely, it can:

• Walk your actual workspace. RGE Studio organizes work by customer, brand, and project (like your accounts, workspaces, and folders). The MCP reads that structure so you can ask for a design in plain language, and drop a new build into the exact folder it belongs in. For a team running dozens of brand workspaces, that alone removes a lot of clicking.
• Build from work you've approved. Point it at an existing on-brand email and it generates a new one in the same structure and style. New work starts from a design you've already signed off on and not a blank canvas.
• Export code that actually renders. Because the build happens in RGE Studio, the HTML you export is the builder's tested, responsive output (not code the model freehanded and hoped would hold up on mobile or Outlook). From there you can push your email to your ESP to do the merging and sending part.

One honest limitation while we're here. The connector can't pull local image files into a build yet, so it works in styled placeholders and you drop the final images in the builder. The same goes for brands. The RGE Studio MCP doesn't force your brand onto whatever the AI produces. It lets you build from designs and styles you've already approved, and it keeps a human sign-off between the drafting process and sending process. Consistency comes from starting on approved material and catching issues at the gate.

Why this one is different

Plenty of tools have an MCP now, so having one isn't the point. What it lets the AI do is. Most fall into three buckets: generation-first ones that make you an untested email, data-first ones for querying campaign and CRM data (useful for analysts, not about the design), and export-only ones that let an agent download existing files but stop there.

The RGE Studio MCP sits at the workspace layer instead: find, build from approved work, act on it, export, reachable from wherever you already create. That layer matters because a starting point drawn from work you've already approved beats one invented from nothing, every time. The MCP is how that starting point reaches the AI tool you're actually working in. You don't even have to leave your platform to do it.

It's also tool-agnostic on purpose. No forced connector, no per-platform lock-in. The same server works with Claude, ChatGPT, n8n, or whatever you pick up next. Claude is the most tested and the one we'd point to first today. RGE Studio becomes the layer underneath your workflow rather than another place you detour into.

Keeping people in the loop

It would be easy to frame this as handing the work to the machine, but the data says otherwise.

Having AI imposed on you by your team was the worst outcome in the survey, worse than rejecting it outright. People who were against AI but whose teams rolled it out anyway reported the lowest satisfaction and the longest production cycles (9.5 days), slower than people using no AI at all. Satisfaction rose with deeper integration and, most of all, with agency. Choosing how AI fit the work mattered more than using it.

Worth noting: our own research found practitioners treat speed as table stakes, not the goal. Quality, confidence, and a starting point that already feels like theirs rank higher. That's why the tool handles scale while you keep judgment, direction, and the sign-off. Connecting AI to a real production layer takes the copy-pasting and color-fixing off your plate, so your time goes to the part only you can do. It doesn't remove you from the loop.

What it looks like in practice

The designer who lives in Figma. The most common professional email workflow runs RGE for inspiration, Figma for design, a builder for the build, and an ESP for the send, with a manual, lossy handoff at every seam. The RGE Studio MCP helps the design process. Design in Figma, hand the screenshot to Claude, and have it rebuild the layout in Studio against your saved styles, so the fonts and hex values are yours from the first pass. You tidy the layout in the builder and export. The screenshot-to-brief-to-agency detour goes away.

The agency juggling a dozen brands. When you run email for ten clients, half the job is not losing track of which folder belongs to whom. The RGE Studio MCP reads your workspace structure in the client, brand, and folder locations so from your assistant you can ask it to build a launch email for a specific client off their approved spring template. It starts from that client's real work and saves the result back into that client's folder. Nothing bleeds across brands, and you never went hunting through dozens of workspaces to find the right starting point.

The one person who does email. Plenty of teams have exactly one person running email (we see you) and they are often defending email's value while doing everything themselves. For them, the builder is usually where things get managed and they dread all the other places they have to go to wrangle assets, copy, approvals, and coordination. With the RGE Studio MCP they stay in the platform they already use: duplicate last quarter's best performer, spin up two or three variations with fresh copy, approve each step, and export. That's a lot of production without opening the builder until they want to or need to.

Keeping the brand from drifting. Large marketing orgs have the opposite problem: a dozen teams, uneven design skill, and a brand that drifts a little further every time someone starts from a blank page. The RGE Studio MCP points all of that build activity at the same approved source. A division building through their own AI assistant starts from the templates and styles the central team already signed off on, saved in the right workspace. And every result still runs through a human approval step before it sends. It doesn't lock the AI down; it makes the on-brand path the path of least resistance.

How to connect it

It's a remote server, so there's nothing to install and no key to manage. Any MCP-capable assistant connects with the server URL. Sign in once with your RGE Studio account (standard OAuth, in the browser) and your tool remembers you. 

See how to connect it.

The short version

• 64% of email pros use AI, but only 1.1% start a new email from it. Trust, not adoption, is the gap.
• The bottleneck was never generation. It's getting output on-brand, rendering cleanly across inboxes, approved, and connected to your ESP.
• What people want most: confidence before sending (53.5%) and integration into the workflow they already use (40%).
• The RGE Studio MCP (beta, usable now) works at the workspace layer. Walk your folders, build from approved work, export where you need it.
• It's tool-agnostic, and the human keeps the sign-off.

FAQ

What is the RGE Studio MCP?
A remote MCP server that lets any AI assistant (Claude, ChatGPT, custom agents, etc) work directly with your RGE Studio designs: browse your workspace, build from approved work, and export, without leaving the AI tool.

Does it generate emails for me?
‍It can build a new design from your approved work and saved styles, right inside RGE Studio's builder. Unlike generic AI, it starts from your real material and routes the result through RGE Studio's approval step instead of handing you a disconnected draft.

Does it enforce my brand automatically?
No. It lets you build from approved designs and styles and keeps a human sign-off between draft and send. Consistency comes from starting on-brand and catching issues at approval.

Can it use my own images?
Not from local files yet. It builds with styled placeholders, and you drop the final images in the builder. Everything else (structure, styles, layout) it handles.

Do I need to install anything?
No. It's a remote server. You need the server URL and a one-time browser sign-in with your RGE Studio account.

Which tools does it work with?
Any MCP-capable assistant. Claude is the most tested and the one we recommend today. ChatGPT (paid plan), Claude Code, n8n, and custom agents all run off the same server.

‍

If your team spends more time cleaning up AI output than thinking about the email, that's the thing worth fixing. Book a demo and see how RGE Studio fits your workflow.

‍