A content workflow is like a kitchen line. One cook who does everything, from prep to plating to dishwashing, can only serve so many tables a night. Add stations, give each one a clear job, and the same kitchen serves triple the covers without hiring triple the staff. Agencies scaling content face the identical problem: the bottleneck usually isn’t talent, it’s structure.
By the end of this, you’ll have a working plan for producing client content at volume without adding a single writer to headcount: what to automate, what to keep human, and how to stop AI drafts from reading like AI drafts.
Prerequisites:
- At least one subject matter expert per client account (product, ops, or customer-facing staff willing to do short interviews)
- A documented brand voice guide for each client, even a rough one
- An AI writing or agent platform your team can configure without engineering support
- A review process with a named owner, not “whoever’s free”
1. Decide whether to scale without hiring at all
Start with the uncomfortable number: 71% of marketers already use AI for content ideas, but 68% say return on investment only improves when expert insight gets folded into the process, according to data reported by Eyeful Media. Read that twice. The tool alone isn’t the lift. The lift comes from pairing it with someone who actually knows the client’s business.
This matters for the decision you’re about to make. An agency that swaps writers for software and stops there usually sees output go up and quality flatten out. An agency that swaps writers for software And builds in a structured way to capture expert knowledge tends to see both go up together.
The trade-off is real, though. Removing writers removes a layer of people who used to catch inconsistency, chase down facts, and push back when a brief didn’t make sense. Software doesn’t do that unprompted. Eyeful Media’s research also notes that content with deeper analysis or expert commentary tends to hold attention longer than generic copy, which is the whole argument for not treating AI as a replacement for expertise, only as a replacement for typing.
So the real question isn’t “AI or writers.” It’s whether your agency can build a system where the AI does the drafting and someone with client knowledge still touches every piece before it ships. If you can build that, cutting the writer roster makes sense. If you can’t, you’re just trading slow, expensive content for fast, forgettable content.
One thing to flag early: this isn’t a one-time setup. Client businesses change, products get updated, and a workflow that worked at launch needs revisiting every few months. Treat it as a system you maintain, not a project you finish.
2. Turn client knowledge into automated drafting playbooks
Here’s the contrast that should shape your next six months: a generic AI prompt produces a page that looks like every other page on that keyword. A page built from a structured SME interview produces something competitors can’t easily copy, because it contains facts only that client’s team knows.
This isn’t a small difference. Flying Cat Marketing’s analysis found that product marketers, solutions engineers, and customer success leads hold exactly the positioning and product detail that bottom-funnel comparison pages need, and that primarily AI-generated articles now make up roughly half of all new web content. When everyone’s drawing from the same public sources, a page built on a client’s internal experts stands out by default, not by effort.
The agency D2L worked with took this approach for competitive and comparison pages, pulling verified positioning directly from internal product experts rather than guessing at it from public competitor sites. The result, per Flying Cat’s case study, was a 73% increase in organic search and LLM demo requests in the first year. That’s not a claim about AI writing faster. It’s a claim about what happens when the input to the AI is better.
What changes for your operation? Writer’s agent Skills creator and Playbook builder are built around exactly this idea: a non-technical team member describes a workflow, the specialized expertise involved, and the outcome they want, in plain language, and the system turns that into a reusable capability other accounts can use too. You’re not writing prompts from scratch for every client every time. You’re building a library.
That library is the actual asset here, more than any single article. The interview process stays roughly the same: find the person who talks to customers or builds the product, ask them what competitors get wrong, ask for specific numbers or examples, and feed the transcript into the playbook rather than a blank prompt box.
It’s slower by a day, maybe two. But a draft built on nothing produces exactly what Flying Cat describes as the forgettable, 10%-citation version. A draft built on a 20-minute call with someone who actually knows the product produces the 76% version, the one more likely to rank and get cited. Don’t cut the step that creates the differentiation.
If you’re mapping out how this fits into a broader production pipeline, it helps to look at a 5-phase content workflow structure before you build your own, so the interview step has a defined place rather than getting bolted on as an afterthought.
3. Review drafts to prevent scaled content abuse and voice drift
Google’s own standards for human search quality raters give agencies a useful warning label here, even though these raters don’t directly set rankings. The guidelines Google publishes cover two specific failure modes: section 4.6.5 addresses scaled content abuse, and section 4.6.6 covers main content built with little effort, little originality, and little added value. Neither section is about AI specifically. Both are about output that reads like it was produced to fill a quota rather than answer a question.
That’s the actual risk of removing writers without replacing their function. A writer, even a mediocre one, tends to notice when a paragraph says nothing. An unreviewed AI draft doesn’t notice anything. It just finishes.
The fix isn’t “hire an editor” in the old sense. It’s building review into the same system that does the drafting, with someone checking specifically for the things Google’s raters are trained to flag: does this page add anything a reader couldn’t get from the next ten results, and does it sound like it was written for a human reader or generated to occupy a URL.
There’s a second failure mode that’s quieter but just as damaging to client relationships: voice drift. Glean’s research on this is blunt. Citing the Content Marketing Institute’s 2024 B2B report, 64% of the most successful content marketers have documented brand voice guidelines.
Only 23% actually feed those guidelines into their AI tools when prompting or reviewing. That gap is where client complaints come from. The guide exists in a shared drive somewhere, nobody references it when generating drafts, and three months in, the client notices their content doesn’t sound like them anymore.
McKinsey’s 2024 research on generative AI adoption backs this up from a different angle: most organizations use AI for content now, but fewer than a third report consistent quality in the output. Speed was never the hard part. Consistency is.
What should an agency actually do with this? Treat the brand voice document as an input to every single draft, not a reference that lives in onboarding materials and gets ignored after week one. If a client has documented tone guidelines, a reviewer, human or system-assisted, checks every draft against them before it ships. If a client doesn’t have documented guidelines, that’s a gap worth closing in month one, because you can’t automate consistency you’ve never written down.
The agencies managing this well treat review as a gate, not a formality. A draft doesn’t go out because it was generated. It goes out because someone confirmed it adds something a reader couldn’t already find and sounds like the client, not like a tool. Skip that gate and you’re not scaling content, you’re scaling the exact failure mode Google’s raters are trained to spot. For a fuller breakdown of where review fits across the production pipeline, the AI content writing workflow for 2026 lays out best practices by stage.
Troubleshooting
Drafts sound generic even after SME interviews. The interview probably stayed too high-level. Ask for specific numbers, named objections customers raise, or a product detail competitors often get wrong. Vague interviews produce vague drafts regardless of the writing tool behind them.
Voice guidelines exist but drafts still sound off. Check whether the guidelines are actually loaded into the prompt or playbook, not just stored in a shared doc. Glean’s research found this exact gap between documentation and use is where most drift starts.
Review is becoming the new bottleneck. This usually means review has no clear owner or no clear checklist. One named reviewer per account, checking against the brand guide and against Google’s originality standard, moves faster than a rotating group with no defined criteria.
Frequently Asked Questions
Can an agency really replace a full writing team with AI?
Not entirely, and that’s not really the goal. The agencies seeing ROI gains are the ones pairing AI drafting with expert input and a review step, not removing the human element altogether. What goes away is the writer role as a dedicated full-time hire; what stays is expert interviews and editorial judgment, applied by fewer people across more accounts.
How much does subject matter expert input actually matter if the AI is good?
It matters more as the AI gets better, not less. Flying Cat’s research found pages built on verified SME positioning reached far higher AI citation rates than generic pages targeting the same keywords. The tool can write fluent sentences from nothing; it can’t invent facts specific to a client’s product or customers.
Will Google penalize agencies for publishing AI-assisted content at scale?
Google’s own documentation on gen AI content doesn’t set a penalty for using AI itself. Its quality raters are trained to flag scaled abuse and low-effort, low-originality output, which is a different standard: volume isn’t the problem, lack of added value is. Content built on real expert input and reviewed for originality fits comfortably inside that standard.
What’s the single biggest mistake agencies make when cutting writers from the workflow?
Skipping the interview step to save time. It feels like the slow part of the process, but it’s the part that produces the differentiation a reader or an AI assistant has a reason to cite. Cutting it doesn’t speed up good content, it just speeds up forgettable content.
Do agencies need documented brand voice guidelines for every client before automating drafts?
Yes, and most don’t have them. CMI’s 2024 research found 64% of top-performing content marketers have these documented, but only 23% actually feed them into AI tools. Building the guide and connecting it to the drafting process closes a gap most agencies haven’t noticed yet.
A system that drafts fast but never checks for added value or voice match isn’t a leaner version of your old process, it’s a faster way to publish content that quietly damages client trust and search performance at the same time. Build the review gate before you build the volume.
If you want help setting up a workflow that pairs automated drafting with the editorial checks that actually protect rankings and client relationships, get in touch with our team at Keyworth and we’ll walk you through how it fits your accounts.
About the author
Abayomi Adenekan
SEO Specialist, 10+ years
Abayomi “Yomi” Adenekan built Keyworth. He has spent over ten years in growth marketing, SEO, and product. He began as a web developer and email marketing specialist, moved into paid media managing $13M a year in ad spend, and now leads growth at a podcast incubator with 1,300+ creators. He also runs Dynareach, his own SEO, web development, and branding agency. He holds an M.Sc. in Data-Driven Marketing from Universidade NOVA de Lisboa and lives in Lisbon.
Leave a Reply