A client context doc is 1 page per client that tells AI what a human strategist already knows. Put in the client's goals, audience, offer, approved claims, voice rules, banned words, approvers, current priorities and past mistakes. Load it into ChatGPT, Claude or Gemini once. Then every draft starts from the client, not from a blank prompt.
AI is not guessing because it is dumb. It is guessing because nobody told it.
Agency owners blame the model. "The copy is generic." "It missed the tone." "It made up a claim."
Here is the shift. The model did exactly what you asked. You gave it a task with no client. So it filled the gap with the average of the internet. Generic in, generic out.
Your best account manager would make the same mistakes on day 1. They get good because they learn the client. AI never gets that chance, because the client lives in people's heads and Slack threads.
The fix is not a better prompt. It is a better memory.
Why prompts alone keep failing
It is not your team's fault. Prompts were sold as the answer. But a prompt is per task. Context is per client.
So every person writes their own prompt. Each one pastes a different chunk of the brief. One version says the client hates the word "cheap." Another forgot. The outputs drift, and your senior people spend their day fixing drafts.
I lived this. I scaled a 25-person marketing agency the hard way, and the bigger it got, the less profitable it became. So I documented every single job. Then I rebuilt each one as AI. The jobs were easy to document. The client context was the part that made it work.
Google says it plainly in its own guide to Gemini Gems: "Provide as much background as possible." A client context doc is how you give that background once, instead of every time.
The 12 sections that stop the guessing
Every section answers a question AI would otherwise guess at.
- Client snapshot. Who they are, what they sell, who they sell to. 3 lines.
- Goals and the 1 number. What winning looks like, in the client's words.
- Audience. Who buys, what they want, what they fear, what they already believe.
- Offer and proof. What you are allowed to claim, with the source for each claim.
- Voice rules. Tone, reading level, sentence length, words they love.
- Banned list. Words, claims, topics and competitors never to use.
- Approvers. Who signs off on what, and how fast they reply.
- Current priorities. What matters this month. This section changes the most.
- Channels and specs. Where the work runs and the format rules for each.
- Past mistakes. What got rejected and why. This is the most useful section.
- Examples of approved work. 2 or 3 pieces the client loved, pasted in full.
- Decision log. Dated notes on choices made, so nobody reopens them.
Sections 6, 10 and 11 do the most work. Rules tell AI what to avoid. Examples show it what good looks like.
The copy-ready template
Copy this into a doc for each client. Fill it in from the kickoff call, the sales notes and the last 3 months of feedback.
CLIENT CONTEXT DOC
Client:
Last updated:
Owner (account manager):
Senior reviewer:
1. CLIENT SNAPSHOT
- What they sell:
- Who they sell to:
- How they make money:
- Why they hired us (their words):
2. GOALS AND THE 1 NUMBER
- 12 month goal (their words):
- The 1 number they check first:
- What would make them renew:
3. AUDIENCE
- Primary buyer:
- What the buyer wants:
- What the buyer fears:
- What the buyer already believes:
- Objections we hear most:
4. OFFER AND PROOF
- Core offer:
- Price or pricing rules we can mention:
- Approved claims (claim + source):
- Claims we cannot make:
5. VOICE RULES
- Tone in 3 words:
- Reading level:
- Sentence length:
- Words and phrases they use:
- Formatting rules:
6. BANNED LIST
- Banned words:
- Banned claims:
- Topics to avoid:
- Competitors never to name:
7. APPROVERS
- Copy approved by:
- Creative approved by:
- Spend approved by:
- Who can veto anything:
- Expected reply time and channel:
8. CURRENT PRIORITIES (update monthly)
- Priority 1:
- Priority 2:
- Priority 3:
- Not a priority right now:
9. CHANNELS AND SPECS
- Channel:
- Format rules:
- Posting or send cadence:
10. PAST MISTAKES
- What was rejected:
- Why:
- Rule we made from it:
11. EXAMPLES OF APPROVED WORK
- Example 1 (paste full text):
- Example 2 (paste full text):
- Why the client liked them:
12. DECISION LOG
- Date / decision / who decided / why:
2 rules for filling it in. First, use the client's exact words wherever you can. Paraphrase loses the voice. Second, every claim in section 4 needs a source. If you cannot point to one, it goes under "claims we cannot make."
Where to put it so AI actually reads it
A doc in a shared drive does nothing if the AI never sees it. Each major tool has a place built for this.
| Tool | Where the doc goes | Source |
|---|---|---|
| ChatGPT | A Project per client. Add instructions in project settings and upload the doc as reference material | OpenAI Help |
| Claude | A Project per client. Set project instructions and add the doc to project knowledge | Claude Help |
| Gemini | A Gem per client. Write instructions and add the doc under Knowledge | Gemini Help |
1 project or Gem per client. Never mix 2 clients in one. That is how a claim from one account ends up in another account's ad.
This is the manual version. Portable Delivery Intelligence is the installed version. Each client's strategy, rules, priorities and approvals sit in one place, connected to the AI tools your team already uses. The context travels with the work, so nobody re-briefs the AI and nobody re-explains the client.
Keep it alive or it goes stale
A context doc goes stale fast. Section 8 changes every month. Section 10 grows every time a client says no.
So set 2 rules.
- The twice rule. If anyone has to correct an AI draft for the same reason twice, the fix goes in the doc.
- The monthly pass. The account manager updates priorities and the decision log once a month. The senior reviewer reads the changes, not the whole doc.
That is how senior people step back. They approve the context once. They stop fixing every draft.
What this does to capacity
Hype is not a plan. Data is. 4 to 8 accounts per account manager is the common ceiling. Much of that ceiling is memory, not hours. Once the client lives in a doc your AI reads, 18 to 25 is our per-operator target with Portable Delivery Intelligence in place. You can see how the math works on our agency capacity page.
Our "25 to 1" model comes from internal mapping: 1 operator plus the system carries the work a 25-person marketing team used to carry. The same delivery system has supported $7M in client revenue. We hold 42 verified Upwork reviews, 5.0, Top 1% Expert-Vetted. Clients include Bowen eBikes and Kajeet. See the proof.
FAQ
How long should a client context doc be?
As long as it takes to fill the 12 sections, and no longer. The examples of approved work and the past mistakes section matter more than length.
Should I put the context doc in the prompt or in a project?
Put it in a project or Gem for that client. Then every chat starts with it, and nobody has to paste it in by hand.
Who should own the doc?
The account manager owns it and updates it. A senior person reviews it once at kickoff and reads the monthly changes.
Is it safe to put client info into AI tools?
Check your client contract and your AI tool's data settings first. Leave out anything the client has not cleared for third party tools.
Bring one client account. We map where your team's time leaks and how many more accounts each person could carry. You keep the map either way. Book your mapping session here.
