There is a version of AI-assisted content creation that produces generic, undifferentiated output and another version that produces something a human editor would actually be proud to publish. The difference, according to Steven Coufal – a growth marketing leader who has spent 15 years in SEO, content, and digital product – is not the model you use but what you feed it before you ask it to write anything. In a conversation with Thibaut de Lataillade, Coufal walked through the specific, step-by-step workflow he has built at Retired.com that combines Claude, SEMrush, Ahrefs, and proprietary thought leadership into a content system that compresses research from days into hours without sacrificing depth or originality.
The Competitive Research Layer
The workflow starts not with writing but with looking. Coufal picks a target keyword – he uses the example of “self-directed IRA account” – and goes to the SERPs. He captures full-page screenshots of the top five ranking results using Chrome’s inspect tool or a full-page capture plugin. For a website redesign project, he might do this for two or three related keywords, ending up with seven to twelve competitor pages in his collection.
These screenshots go into Claude with a prompt asking for a component inventory: what elements does each page contain, and how frequently does each component appear across the competitor set? The six components that every site has are table stakes. The components that only one or two sites include are potential differentiators. The components that nobody has are gaps.
“Maybe we would’ve taken a day to do that project before and now you can do it in a couple hours.”
The critical point is that Coufal is not asking Claude to make design decisions. He is asking it to do the tedious cataloguing work that a human would do by opening twelve tabs and toggling between them with a spreadsheet open. The strategic interpretation – which components to adopt, which to skip, which gaps to exploit – remains human. The machine handles the inventory. The human handles the judgment.
Building a Content Brief That Actually Works
For blog content, the process deepens. The starting point is the same – screenshots of the top five ranking results – but Coufal layers on additional data sources before Claude sees any of it. He includes the AI overview results for the target keyword, because if Google is surfacing a direct answer, your content needs to account for it. He pulls keyword data from SEMrush: primary keywords, secondary keywords, People Also Ask queries. Everything goes into a single Claude conversation as context before the first creative prompt is issued.
The prompt itself is not “write me a blog post.” It is “here is everything I know about this competitive landscape. Build me an outline that is comprehensive relative to what’s ranking and identifies topical gaps that the top sites are not covering.” The distinction matters. The first prompt produces generic content. The second produces a structural blueprint informed by the actual competitive reality.
From the outline, Claude can fill in sections, and the rough draft becomes what Coufal calls “B-plus style content” – algorithmically informed, structurally sound, but not yet distinctive. This is where most AI content workflows stop, and it is precisely the stage where Coufal’s approach diverges.
The Special Sauce: Proprietary Thought Leadership
The step that transforms B-plus content into something worth publishing is the injection of proprietary expertise. Retired.com has an internal thought leader who has recorded hours of video interviews. Coufal loads those YouTube transcripts into Claude and asks it to find places in the draft where genuine expert quotes can be inserted – moments where the internal perspective adds something that no competitor’s content contains.
He also feeds in Ahrefs backlink data for relevant sources, asking Claude to cite them appropriately. And internal links get woven in during this same pass.
“It’s not only are you able to do something that’s comprehensive compared to competitors, but you’re able to already get those niches in there.”
The result is content that satisfies three criteria simultaneously. It is structurally competitive with whatever currently ranks – because it was literally built from an analysis of those pages. It covers topical gaps the competitors missed – because Claude was explicitly asked to identify them. And it contains unique, proprietary perspectives that no other site can replicate – because those perspectives came from the company’s own thought leaders, not from the model’s training data.
The Editing Pass
The final stage runs the draft through SEMrush’s writing tools. Coufal uses them to identify overly complex sentences, wordy passages, and potential originality concerns. This is the quality control layer, and it serves a specific purpose: catching the places where Claude’s natural verbosity or pattern-matching has produced something that reads like AI wrote it.
The workflow, from start to finish, looks like this: SEMrush and Ahrefs provide the data. Google SERPs provide the competitive landscape. Claude compresses the research, builds the structure, and produces the draft. Internal thought leadership provides the differentiation. SEMrush’s editing tools provide the final polish.
What Coufal does not do is ask Claude to generate content from nothing. Every creative step is informed by specific, curated inputs. The model is not deciding what to write. It is being told exactly what the competitive landscape looks like, what gaps exist, what the company’s unique perspective is, and then asked to synthesize all of that into a coherent document. The human decides the strategy. The machine executes the assembly. The human edits the result. That division of labor is what makes the output consistently publishable rather than consistently generic.
Beyond Content: Data Visualization and Dashboards
Coufal extends the same philosophy to performance tracking. Using a data connector called Windsor AI, he pipes Google Analytics, AdWords, and Search Console data directly into Claude, enabling real-time dashboard creation and recommendations. For quick analyses, he simply dumps Excel exports of backlink audits into Claude and asks it to build interactive HTML charts comparing sites across different factors.
The pattern is consistent: traditional SEO tools generate the data, and Claude handles the synthesis, visualization, and pattern recognition that would otherwise consume hours of manual Excel work. The strategic decisions – where to invest, what to prioritize, which signals matter – remain with the human who understands the business context that no model can infer from a spreadsheet alone.
For the full interview breakdown, see our complete Expert Insight with Steven Coufal.
Tools Mentioned in the Interview
The following tools and platforms were referenced during this conversation.
Related Content
From Link Building to AI Workflows: Expert Insights with Steven Coufal
Steven Coufal has spent over 15 years scaling organic growth from compliance content to international SEO empires to AI-powered workflows. He breaks down how Claude and SEMrush are replacing day-long research sprints, why chasing Google updates is a losing game, and what LLM citation tracking actually reveals about your brand.
Tracking AI Search Citations: What SEMrush Reveals About LLM Visibility
Steven Coufal explains how SEMrush's AI citation tracking uncovered surprising visibility gaps -- from competitors ranking via LinkedIn posts to outdated doctor profiles being cited by ChatGPT -- and why fewer clicks from AI search might actually be worth more.
Stop Chasing Google Updates: Why a User-First SEO Strategy Wins
Steven Coufal explains why obsessing over Google algorithm updates is a recipe for professional burnout, how an SEO consultant helped him shift to a user-first strategy, and why building for users instead of algorithms is the only approach that survives the fracturing search landscape.