Client Project Context Stack
Give every client and deliverable its own reusable AI context
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 7
- Confidence
- 99%
The Client Project Context Stack creates a dedicated AI workspace for each client and, when necessary, each type of deliverable. Reference files hold company facts, product information, preferred writers, style samples, and previously approved work. Persistent instructions define how the model should behave, including tone, structure, exclusions, and accuracy expectations. New tasks inherit this context, reducing the need to restate foundational information and enabling a solo operator to manage more clients. Because project features may fail to apply every instruction consistently, the workflow also reinforces critical directions in the current chat. The stack improves over time as approved outputs are added, turning client feedback into reusable context rather than a correction that disappears after one task.
Origin
Barbara Jovanovic adopted separate ChatGPT projects for clients and deliverables to scale her solo agency while retaining detailed product knowledge and voice consistency. Extracted from Marketing Against The Grain.
Core principles
- 01Persistent context is infrastructure for scalable AI work
- 02Different deliverables may require different instructions
- 03Product facts and style references serve different purposes
- 04Approved outputs are high-value voice examples
- 05Critical instructions should be reinforced when tools apply them inconsistently
How to run it
- 1
Partition the work
Create a project for each client and separate deliverable types when their structures or voices differ.
Pro tip Start with separate projects for clearly different outputs such as blogs and social posts.
Watch out Excessive fragmentation can make shared context difficult to maintain.
- 2
Load factual context
Add a comprehensive reference covering the company, product, terminology, audience, and relevant history.
Pro tip Maintain one canonical document that is easy to update.
Watch out Stale product details can make every future output inaccurate.
- 3
Load voice references
Add writing samples, admired authors, and examples that represent the desired style.
Pro tip Ask the client which writers and publications they genuinely read.
Watch out Style imitation should guide qualities rather than copy another writer's expressions.
- 4
Capture approved work
Add outputs the client explicitly praises as examples of successful voice and structure.
Pro tip Record why the client approved an example when that feedback is available.
Watch out Do not add mediocre work merely because it was published.
- 5
Set persistent behavior
Write project instructions for tone, language, structure, evidence, and prohibited habits.
Pro tip Separate positive directions from explicit exclusions.
Watch out Project instructions may not be followed reliably in every generation.
- 6
Reinforce and review
Restate critical instructions in the active chat and verify the output against the project context.
Pro tip Prioritize factual and brand-critical constraints when repeating directions.
Watch out Persistent context does not eliminate the need for human review.
- 7
Maintain the stack
Update references whenever products, positioning, voice preferences, or approved examples change.
Pro tip Schedule periodic reviews for long-running clients.
Watch out An unmanaged context stack gradually becomes contradictory.
In the wild
An agency creates a dedicated blog project containing product documentation, a downloaded document of style samples, favorite writers named by the founder, persistent drafting instructions, and articles the founder previously praised. Each weekly draft begins with this context already available.
→ Recurring drafts require less setup and remain more consistent with the client's voice.
Common mistakes
Mixing every deliverable together
A single project may apply blog conventions to social posts or otherwise blur format-specific behavior.
Saving facts but not voice
Product documentation improves accuracy but does not teach the model how the client should sound.
Treating instructions as automatic
Current project systems may ignore persistent instructions, so critical requirements still need verification and reinforcement.
Is it for you?
Best for
Agencies and internal teams producing recurring deliverables for multiple brands or executives.
Not ideal for
One-off tasks where building and maintaining persistent context would cost more than it saves.
From the transcript
“We have basically a project for each client and for each of their deliverables.”
“And when we create a piece that he's like, that was great. Love that. Love that style that he established. I also just put it…”
“And then in the instructions, this is arguably more important than the documents. You just instruct it how you want it to behave every single…”
From the episode
I Run a 6-Figure Agency with ZERO Employees (AI Only)
Barbara Jovanovic