Claude for Marketing: How We Trained It to Work Like a World-Class Marketer

Most teams using Claude for marketing get a capable chatbot that writes decent drafts. What they don’t get is an AI that understands the business, works out the highest-value tasks, does them, and reports back.

That gap is what Tim Cameron-Kitchen, Founder of Exposure Ninja, set out to close by building Opal: a framework that turns Claude into an AI agent that thinks and acts like an experienced marketer.

Why Vanilla Claude Produces Content That Doesn’t Sell

To show the difference, Tim Cameron-Kitchen gave the same prompt to two versions of the same model (Claude Opus 5.5 on max thinking in Claude Cowork). One was Claude straight out of the box. The other was Claude running on Opal.

The prompt asked for a blog post comparing heat pump running costs with and without a home battery, plus tips on sizing a battery.

“Vanilla” Claude did a solid job on paper. It sourced energy tariff data, linked its references, and added a chart and a comparison table. As Tim Cameron-Kitchen puts it, “a couple of years ago, this would have been a really high-quality blog.”

But he explains why that’s not enough from a commercial point of view: “Yes, it’s a detailed and factual blog with plenty of data in, but there’s nothing about this blog that sells the business that would have been publishing this article, right? So, as a marketer, it’s not going to generate any leads.”

That’s the problem for marketing leaders. Generic content is “just a good piece of content that matches what anybody could ask their AI for.” If your competitors can produce the same article with the same prompt, it gives you no competitive advantage and no pipeline.

What Claude With Opal Produced Instead

The Opal version was published on the website (after approval), with calls to action (CTAs), the brand’s tone of voice and visual guidelines, images from the brand’s own library, and a relevant embedded video.

The biggest difference came from what it knew about the business. Tim Cameron-Kitchen explains where that knowledge came from:

“Because this version actually knows the business’s customers in depth, because it’s got access to the CRM and it’s been through those customer conversations, about 5 and 1/2 thousand customer conversations through email threads and phone transcripts, it understands the customers, and it also understands the brand’s tone of voice.”

That customer insight shaped the content. It added a calculator because it had tested calculators before and seen good engagement, and the calculator used “this company’s charges rather than just generic charges that might not work.”

Tim Cameron-Kitchen is clear that it isn’t perfect. Some layout elements needed tidying, and every article still needs editing and fact-checking. But as a marketing asset, it was in a different league.

Vanilla Claude would just give you a draft that you have to then publish. Claude in Opal not only published it on the website, it optimised the page title and meta descriptions and even submitted it to Google via Search Console.

The Results: 3x Quote Requests and Top AI Search Recommendations

For Elite Renewables, one of the businesses we own, Opal delivered three times the quote requests from search, a doubling of total weekly enquiries, and “a bunch of position one rankings that actually Opal has picked up itself.”

Search “best heat pump installers London” or “best MVHR installers London” and Elite Renewables is the top recommendation in AI Search, cited as a source, and ranking first in the organic results.

In Tim Cameron-Kitchen’s words, Opal “does actual things. It gets results.”

The Four Gaps That Stop Claude From Doing Real Marketing

So why does vanilla Claude fall short? Tim Cameron-Kitchen identifies four gaps.

1. It can’t see. Without your rankings and company data, vanilla Claude has no way to tie its work back to your goals.
2. It waits to be told. As he puts it, “that little blank box at the bottom waiting for your instruction can be really difficult for marketers.”
3. It stops at a draft. Opal is trained to publish, follow up, and submit to Search Console, so it’s “taking things off your to-do list rather than putting things on your to-do list.”
4. It forgets. Opal has a proper memory, so “once you correct it once, you don’t have to correct it on that thing again.”

How the Opal Framework Works: Observe, Plan, Act, Learn

Opal stands for observe, plan, act and learn. It runs this loop against quarterly targets, then monthly, weekly and daily.

Observe: Connect It to Real Business Data

Opal starts with a context pack about the business, then connects to its data. For Elite Renewables, that includes Google Search Console, Google Analytics 4 (GA4), Semrush (for rankings and AI Search visibility), the CRM, the sales sheet, customer survey data, Google Drive, government websites and relevant industry sources.

It also connects to tools where it does the work: WordPress, ad campaigns, browsers, YouTube, OpusClip, Descript, Slack, and Microsoft Outlook.

Plan: Work Back From a Commercial Goal

Elite Renewables gave Opal a sales goal of £700k per month. It surpassed that in month one, then worked out the levers: more large enquiries, more leads reaching a survey through email follow-up, and higher value per sale.

In the first half of October, it generated almost £2 million in enquiries from the larger jobs it targeted. Tim Cameron-Kitchen explains that everything runs through this goal: “If you ask it to do something, it will do that thing, but it’ll do it through the filter of getting to that goal.”

Act: Use Proven Skills Within Clear Rules

Opal uses skills built from Exposure Ninja’s experience with thousands of clients over 12 years. Without them, Claude “makes up the process to do that task each time, right? Because it’s non-deterministic.”

It also works within a constitution. If you tell it “never draft an email” or “never access our CRM,” it won’t.

Learn: Build a Memory That Improves Every Run

After every task, Opal updates a memory file. When it gets stuck, it “leaves guidance for future agents to follow so that they don’t get stuck in the same way.” Each morning, it sends a status update (via Slack, in Exposure Ninja’s case) covering what it did, what needs approval and where it’s stuck.

“Because it’s running this cycle repeatedly, it gets better and better and better. So, as you use it over time, it learns you and it learns the business more and more,” Tim Cameron-Kitchen, Founder of Exposure Ninja.

Why Experienced Marketers Still Matter

Tim Cameron-Kitchen estimates that Claude with Opal “gets you about 70% of the way.” The human piece is small, but it’s high leverage:

“It’s kind of like you’ve got a really super intelligent employee that you need to give feedback to, but if you give great feedback to it, it learns from that feedback forever and doesn’t make those same mistakes,” Tim Cameron-Kitchen, Founder of Exposure Ninja.

The catch is that “you do have to know what really good looks like.” That’s why Exposure Ninja’s team calibrates and scores Opal for each business.

After running Opal on our own businesses for most of this year, we’re now building it for clients. If you want AI that does your marketing rather than adding to your to-do list, request a demo to see whether Opal could work for your business.