If you run marketing for a small company — two people, maybe three — you already know that content demands are relentless and time is not. An ai content workflow built for a small team isn’t about doing everything AI can theoretically do. It’s about removing the four or five bottlenecks that eat your week, so the two of you can actually publish consistently without burning out. At Choco Media, we run exactly this kind of lean operation, and the pipeline described here reflects what we’ve refined over the past year of daily production.
This post is for founders, in-house marketers, and small agency teams who are done experimenting with AI tools in isolation and want a repeatable end-to-end system. We’ll walk through every stage — brief, draft, edit, publish — and name the specific decisions that make the workflow function without a dedicated content manager watching over it.
By the end, you’ll have a model you can adapt to your own stack in a day or two. No six-month transformation project. Just a working pipeline.
Why most small teams’ AI workflows stall before they start
The failure mode we see most often isn’t tool selection — it’s missing scaffolding. Teams buy access to Claude or ChatGPT, generate a few decent drafts, then drift back to writing from scratch because the AI output needs so much reworking it’s faster to do it manually.
The root cause is almost always the same: the brief going into the AI is too thin. When the brief is a two-sentence instruction, the model fills in the gaps with generic content. When the brief is a structured document with audience, intent, proof points, and voice notes, the output is 80% publishable on the first pass.
- Thin briefs produce generic drafts that require full rewrites.
- No review checklist means every editor applies different standards — inconsistency accumulates.
- No publish queue means finished content sits in a folder while the team decides what to do with it.
- No feedback loop means prompts never improve, so quality stays flat.
The system below addresses all four. It’s designed to be run by two people — one handling briefs and strategy, one handling editing and publication — though in practice one person can cover everything if the workflow is tight.
Stage 1: Brief — the document that does the real work
The brief is the highest-leverage document in the pipeline. Ten minutes writing a good brief saves forty minutes of editing later. We use an eight-field structure that’s become non-negotiable for us.
The eight-field brief structure
- Target audience — one specific person, not a demographic. “A founder running a 3-person SaaS team who handles their own marketing” beats “B2B marketers.”
- Search intent — what the reader typed and what they actually want. Informational, commercial, navigational.
- Target keyword — primary phrase with secondary variants. Include the exact phrasing, not a paraphrase.
- Angle / point of view — the specific opinion or frame that makes this piece ours. Not “AI is useful for content” but “AI is only useful after you fix the brief.”
- Proof points — three to five specific claims, data points, or examples the post must include. These prevent the model from inventing plausible-sounding but false statistics.
- Voice notes — two or three sentences from our brand voice guide. The model doesn’t read the full guide; it gets the relevant extract for this post.
- Internal links — two to three URLs we want referenced naturally in the body.
- Structure skeleton — four to six H2 headings as working titles. The model can rename them, but the structure gives it a spine.
We keep these briefs in a Notion database. Each row is one post. When the brief is complete, it gets a status tag of “ready to draft” and lands in the AI queue automatically.
“The brief is the brief for the AI and the brief for the editor. If a human writer couldn’t execute from it, the AI definitely can’t.”
Stage 2: Draft — what you give to the model and how you prompt it
We use Claude for long-form drafts and ChatGPT for shorter-form variation work, though either can do both. The prompt structure matters more than the model choice.
The draft prompt
We feed the model the full brief, then append a prompt block that includes: word count target, section-by-section structure (with H2s from the brief), specific instructions about what to avoid (phrases, jargon, tone patterns we don’t like), and a short example paragraph in our voice for calibration.
- Give the model the full brief, not a summary of it.
- Specify what to avoid as precisely as what to include.
- Paste in one to two example paragraphs from a post you’re proud of — calibration beats instruction.
- Ask for the draft in one response, not section by section, so the model maintains coherence.
A well-structured prompt to a capable model with a solid brief typically produces a draft that needs editing, not rewriting. In our workflow, the draft stage takes about four minutes of human time — the time to write the prompt and review the output before handing it to the editor.
For more on how we approach AI content creation as a service, the structure is essentially the same at scale.
Stage 3: Edit — the human pass that makes it publishable
Editing AI output is a different skill than editing human writing. The model doesn’t repeat itself out of boredom or lose its thread — but it does have characteristic failure modes. Once you know what to look for, a thorough edit takes fifteen to twenty minutes for a 1,500-word post.
The six things we always check
- Generic openings. The model defaults to scene-setting preamble. Cut to the first substantive sentence.
- Passive constructions. AI writing trends passive. Flip to active wherever it reads as hedging.
- Unsupported claims. Flag any statistic or “studies show” assertion that isn’t sourced in the brief. Delete or verify before publishing.
- Phrase repetition. The model sometimes locks onto a word — “crucial,” “essential,” “key” — and uses it in every section. Find-and-replace runs.
- Missing specificity. Vague examples are the most common editorial weakness. Replace “a B2B company” with a plausible specific scenario.
- Voice drift. Mid-post, the model often relaxes into a more corporate register. The last third of a draft usually needs the most voice work.
We use a shared Notion checklist that the editor ticks off per post. This keeps quality consistent whether it’s one person editing all posts or the workload is split.
Stage 4: SEO pass — keeping the keyword honest
After the voice edit, we do a quick SEO pass. This doesn’t mean stuffing the keyword — it means verifying that the target phrase appears where it should without reading as planted.
- Target keyword in the H1 (post title).
- Target keyword in the first paragraph, naturally.
- Target keyword in at least one H2.
- Meta description written (140–155 characters, keyword included).
- Internal links confirmed — anchor text is descriptive, not “click here.”
- TL;DR / FAQ block present (we use a custom meta field; it generates FAQPage schema automatically).
For posts targeting AI search — which is most of our content now — we also check that the opening paragraph answers the question directly. AI Overviews and ChatGPT citations heavily favour posts that give a crisp answer in the first 100 words. We covered this pattern in detail in our guide on how to get cited by ChatGPT.
Stage 5: Publish — scheduling and the one step most teams skip
Publishing is mostly mechanical, but the step most teams skip is the scheduling discipline. Content published in bursts — three posts this week, nothing for six weeks — sends weak signals to search engines and makes it hard to build audience expectations.
For a two-person team, we recommend committing to one post per week at minimum, scheduled in advance. If you batch-produce content (brief five posts, draft five posts, edit five posts in one session), you can maintain that cadence with roughly four to five hours of focused work per week across both team members.
- Schedule posts at least a week out — don’t publish same-day unless it’s time-sensitive news.
- Keep a “ready to publish” buffer of two to three posts. It’s your insurance against a bad week.
- Set a recurring calendar block for content production. Unscheduled work doesn’t get done consistently.
- Review scheduled posts the day before they go live — catch anything that aged poorly since drafting.
The feedback loop: how the workflow improves itself
A static workflow stops improving. The compounding advantage of a well-maintained AI content workflow comes from the feedback loop — the system that routes performance data back into better briefs and better prompts.
What we review monthly
- Top-performing posts by organic traffic. What angle, format, or keyword pattern do they share? Bias the brief queue toward those patterns.
- Draft quality rating. We score each draft 1–3 before editing (“needed major rework / needed moderate editing / nearly ready”). Track this over time. If scores drop, the prompts need updating.
- Editing time per post. An upward trend means briefs are getting thinner or voice drift is increasing. Investigate before it becomes a problem.
- Keyword ranking movement. Posts that rank page 2–3 are candidates for a refresh rather than a new post on the same topic.
This monthly review takes about an hour. We do it at the end of each month before building the next month’s brief queue. The insight compounds — by month three, brief quality is measurably better than month one, which means editing time drops and post quality rises.
Tools we actually use in this workflow
We’ll name tools specifically because vague recommendations aren’t useful. This stack has worked for us; the specific tools matter less than having one clear tool per stage.
- Brief and queue: Notion database with a custom template. One page per post, status tags to move through pipeline stages.
- Drafting: Claude (claude.ai or API) for long-form. Occasional ChatGPT-4o for variation experiments.
- Editing: Google Docs for collaboration; Hemingway App for a quick passive-voice check before final copy-paste.
- SEO checking: Ahrefs for keyword validation at the brief stage; manual check for on-page elements before scheduling.
- Publishing: WordPress with a custom snippet for rendering (ours). Any headless CMS or standard WP installation works fine.
- Performance review: Google Search Console + GA4. We don’t use a reporting layer for this — the raw data is enough at two-person scale.
The total tooling cost for a two-person team running this pipeline is roughly €80–120/month, depending on which AI tier you’re on. That’s not nothing, but it’s well below the cost of a freelance writer producing the same volume. Our AI automation services page covers how we help teams implement and maintain workflows like this.
Common questions from teams getting started
How long does it take to set this up?
Budget a focused day: two hours to build the Notion brief database and template, one hour to document your voice guidelines extract, one hour to write your first three briefs from scratch. After that, each new brief takes fifteen to twenty minutes. The editing checklist is an afternoon’s work to write and then just becomes habit.
What if one person has to do everything?
It’s doable. The main risk is that brief quality suffers when you’re also the one editing and publishing — you lose the outsider perspective that catches when a post makes assumptions the reader won’t share. The fix is a brief review step: write the brief, wait 24 hours, read it as if you’re the AI. Add anything you’d ask if you were being briefed cold.
How do we handle topics that are too thin for a full post?
Mark them in the queue and come back when you have more to say. A post you can only fill to 900 words of genuine substance isn’t ready. Padding to hit a word count produces exactly the generic filler that neither readers nor search engines reward. Better to push it a month and do it properly.
What to expect in the first 90 days
In our experience building and maintaining content pipelines for clients, the first 90 days follow a predictable arc. Weeks one to two are setup and first runs — expect higher editing time as you calibrate prompts and brief quality. Weeks three to six are the tightening phase — you’ll find two or three recurring editing issues and build those into the checklist. By weeks seven to twelve, the workflow is genuinely running, editing time has dropped, and the brief queue starts building faster than you can publish.
- Month 1: Setup + calibration. Expect 3–4 hours per post (combined).
- Month 2: Tightening. Expect 2–2.5 hours per post.
- Month 3: Rhythm. Expect 1.5–2 hours per post, and a clear sense of what your best content looks like.
Those numbers assume one 1,500–2,000 word post per week. At higher volumes, the economics improve further — the fixed cost of building and maintaining the system amortises across more output.
If you’d rather have us build and run this for you, or if you want a structured setup session, get in touch — we’re happy to walk through what makes sense for your team’s size and publishing goals.