AI didn't kill the writing job. It made everyone an editor.
A content operation that once needed 20 writers can now run on one person and a well-trained AI system, according to a conversation between C2 Media's Enricko Lukman and StoryMint's Miftahul Khoir. The catch: someone still has to know what good looks like.
Ten years ago, Enricko Lukman ran a content team of about 20 people, each writing two or three articles a day. Today, he says one person can put out roughly 100 articles a month, with AI doing most of the drafting and the human job shifting toward directing, training and checking the system.
That shift anchored a September 4 conversation on StoryMint's The Brief podcast, where two people who've each spent years scaling content operations compared notes on what AI search is actually changing.
Enricko is CEO of C2 Media and a former tech journalist at Tech in Asia; his agency has worked with hundreds of clients on their comms and marketing. Miftahul Khoir is co-founder of StoryMint and Head of Digital Marketing at Kalbe Group, where he previously grew KlikDokter's monthly pageviews to more than 30 million.
StoryMint grew out of that same tension between producing more content and knowing whether it's actually reaching anyone, an audience-first suite with persona tools, AI visibility tracking and Google Search Console-connected optimization for how brands show up in search and AI answers.
The conversation moved between two problems that are increasingly hard to separate: how marketers get discovered when AI systems answer questions directly, and how editorial teams should use AI without giving up judgment.
Table of contents
- AI search is changing what success looks like
- The old SEO playbook still matters
- AI writing moves humans up the stack
- Hallucination makes verification part of the workflow
- Persona-specific content may have an edge
- What marketers should know
AI search is changing what success looks like
Enricko's starting point is that search has not stopped evolving. Google moved from blue links to direct answers and featured snippets, then added AI Overviews, while tools such as ChatGPT, Claude and Perplexity made conversational answers part of everyday discovery.
The bigger change, he argued, is not simply the interface. It is the metric. A pageview used to be an obvious sign that search had done its job. In AI search, a brand can influence a decision without sending a visitor to the site at all.
Miftahul described seeing that gap in practice. In one example from his work, a customer said an AI recommendation had led them to a product, then completed the purchase elsewhere. In another, he said people arrived at a hospital already asking about a particular product after encountering it through Google or AI tools. The marketing team could see the downstream behavior, but not cleanly attribute it back to a specific AI answer.
Enricko said his own team has run into the same problem with inbound leads. When a prospect appears from an unusual market, the team now asks directly how they found the company. Sometimes the answer is ChatGPT rather than Google.
Google has started exposing more AI-specific visibility data in Search Console through its generative AI performance report, including impressions and page-level visibility. But that still does not solve the broader cross-platform attribution problem the two speakers described, especially when a recommendation happens inside one AI assistant and the eventual conversion occurs somewhere else.
The old SEO playbook still matters
Neither speaker treated GEO or AEO as a clean break from SEO.
Enricko said much of the foundational work remains familiar: a technically sound site, useful content, authority, trust and enough specificity to match the question being asked. Where AI search differs is in how many sources it can synthesize. A company's own website may be only one input alongside reviews, forums, media coverage, social posts and other third-party material.
That makes narrow positioning more valuable. Enricko gave a simple example from his own agency work. Competing for a broad prompt such as best PR agency is difficult. Narrowing it to best PR agency in Indonesia is still difficult. But best PR agency in Indonesia for tech companies creates a more specific context in which the company has a better chance of appearing.
The lesson is less about finding a magic AI-writing format than about giving the system enough consistent evidence to understand what a company is relevant for. StoryMint's own AI Mode documentation follows a similar logic by using Search Console queries to identify questions where a site already has some relevance but is not yet cited in AI answers.
Patrecia Meliana
AI writing moves humans up the stack
The most concrete part of the conversation was not about ranking at all. It was about what happens inside the content team.
Enricko said beginner-level writing work is increasingly easy to automate, especially for non-specialist content. His team has moved from a traditional chain of writers, editors and a chief editor toward a model in which people increasingly act as the editor of an AI system.
“Now everyone is the chief editor,” Enricko said. “They decide the direction, hire the AI agents, train them, and supervise their work.”
He described that progression in three stages. First, AI acts like an assistant: a human chooses the topic, reviews the draft and publishes it. Next, AI becomes the writer and uploads the article as a CMS draft for final human review. At the most mature stage, the system can write, upload and publish, while humans audit its work periodically and feed corrections back into the workflow.
Miftahul uses a more layered process for his own writing. He said he often uses ChatGPT for ideation, Perplexity to validate and structure the material against sources, and Claude to smooth the language. He has also built internal dashboards that combine marketing data and use different AI agents for analysis tasks.
“The tools are just weapons,” Miftahul said. “It comes down to the person behind the gun.”
For him, experience is what lets a marketer recognize when an AI output is weak, implausible or simply wrong. That is also why the lower barrier to entry cuts both ways: younger practitioners can reach technical competence faster, but inexperienced users can also accept mediocre output because they do not yet know what good looks like.
Hallucination makes verification part of the workflow
Both speakers returned repeatedly to the same operational risk: AI can produce fluent answers that are not reliable enough to publish without checks.
Enricko said his editorial systems are designed around that limitation. AI is instructed not to assume facts, to double-check claims and, in some workflows, to pull from databases the team has already reviewed. The point is not to pretend hallucination has been solved. It is to build a process in which unsupported claims are harder to pass through unnoticed.
Miftahul takes a similar approach, but uses the differences between tools as part of the workflow. He said he is comfortable using ChatGPT's tendency to range widely during ideation, then moves the material into a research-oriented tool for validation before using another model for style and readability.
That division of labor matters because AI-assisted can describe very different systems. One team may use a single model to draft copy. Another may use separate agents for research, source checking, writing, editing, CMS delivery and performance analysis. The common requirement is still human judgment over what the system is allowed to assert.
Patrecia Meliana
Persona-specific content may have an edge
The conversation closed on a point where StoryMint's product philosophy and Enricko's search argument converged: generic content is becoming harder to defend.
StoryMint's persona system is designed to make audience profiles a working input across content drafting, gap analysis and AI visibility checks. Miftahul said that if an output misses the mark, the problem may start with an inaccurate persona rather than the writing prompt itself.
Enricko connected that directly to AI search. As queries become longer and more contextual, content written for a specific audience has more useful signals to match against.
“If we make content for everyone, we'll definitely lose,” Enricko said. “We need to make content for a specific persona.”
For content teams, that suggests a practical split. Broad informational articles may still have a role, but the defensible work increasingly comes from material that is specific about the reader, the use case and the decision being made.
What marketers should know
The conversation did not produce a single new rule for GEO, and both speakers were explicit that the field is still changing too quickly for that kind of certainty.
What it did produce was a workable operating model. SEO fundamentals still matter. AI visibility needs to be tracked beyond clicks. Content has to exist across a wider source ecosystem. And AI can automate much of the production layer only if experienced people remain responsible for direction, verification and quality.
For teams deciding where to start, the sequence may be simpler than the terminology suggests: define who the content is for, create something specific enough to be useful, make sure the claims can be verified, then measure both the traffic you can see and the AI-driven outcomes that still require manual attribution.
