Retainer Friday: you are behind on one workstream and waiting on brand assets. Asking a model to “make this update sound positive and professional” often returns “strong progress across workstreams” with the blocker as a footnote. Your client plans launches around that email. Spin is not courtesy; it is a scheduling hazard.
This is the freelance twin of async update writing without spin for internal teams. Same mechanics, different reader: a paying client, not a manager on Slack.
Why models drift toward upbeat language
Generative systems are often tuned toward agreeable, helpful-sounding prose. OpenAI publicly discussed and rolled back a sycophancy-related drift in an April 2025 GPT-4o update (OpenAI, “Sycophancy in GPT-4o”). Risk-management framing treats that drift as a predictable failure mode when the stakes are client planning, not as a quirk to ignore (NIST AI Risk Management Framework). If your prompt is vague (“write a client update for Project Orion”), the model has no access to the real status and will invent a plausible narrative. Plausible is not accountable.
Do not let an AI-drafted client update upgrade “blocked” or “behind” into language that implies on-track delivery. If the honest status is delayed, say delayed, with what you need to unblock.
Workflow for solo operators
Step 1: Facts card first (no AI)
Fill in plain language:
- Shipped / done since last update
- Blocked on (or “nothing”)
- Next concrete actions and dates you have actually checked
- Confidence: on track / at risk / blocked - need X by date
- Scope changes since last update (link your scope change log)
Step 2: Format-only prompt
Turn these facts into a short client status email.
Keep tone plain and professional.
Do not add progress, confidence, or timelines beyond what I wrote.
If something is blocked, put that near the top, not as a footnote.
Facts:
[paste facts card]
Step 3: Spin-word scan
| Phrase to hunt | Replace with |
|---|---|
| Great progress / crushing it | Specific shipped items only |
| Nearly there / almost done | A date you believe, or “no solid ETA yet” |
| On track (when blocked) | Blocker first + ask |
| Should be fine | Your real confidence line |
Step 4: Privacy pass before send
If you used AI to format the email, confirm you did not paste credentials, personal data about the client’s customers, or confidential annexes into the chat. Status facts rarely need those details.
Step 5: Escalation pattern
If the same item is “at risk” for three updates, propose a call or a scope decision. Do not let the weekly email absorb a stalled decision forever. When AI assistance on deliverables is material, follow disclosure norms that keep your name accountable (keep your name on AI-assisted work).
Save a “facts card” snippet in your email tool. Refuse to open the AI tab until all five lines are filled.
Illustrative scenario (labeled)
Illustrative scenario, not a measured case: A freelance PM sends an AI-polished update saying the integration is “on track for Friday.” Internally they know the API key from the client never arrived. The client’s launch comms go out Wednesday. The repair costs more than an early blunt email would have.
Retainer vs project updates
On retainers, a quiet week still needs a factual note: what moved, what did not, what you need. Silence trains clients to assume progress. On fixed projects, updates should track milestone language from the SOW so “done” means the same thing on both sides. If the SOW is vague, fix the SOW language at the next change window rather than inventing certainty in email.
Sample facts-to-email shape
Illustrative only:
Facts: Shipped homepage wireframes v2. Blocked on final brand icons from client design lead. Next: UI kit once icons arrive; cannot commit to Friday visual QA without them. Confidence: blocked - need icons by Wednesday to keep Friday.
Email shape after format-only pass should still lead with the blocker and the Wednesday ask. If the draft buries that under “solid progress on homepage,” reject the draft.
Validation and fallback
Validation: a stranger reading only the email would not plan differently than your facts card justifies.
Fallback: send the facts card as a bulleted email with no AI pass. Ugly and clear beats pretty and wrong.
Misleading commercial communications can raise consumer-protection issues when you sell to consumers (FTC advertising overview; FTC crackdown on deceptive AI claims; EU UCPD) - even a status email can set expectations about delivery. For B2B retainers the duty is still practical: do not create false certainty.
Desk version: client update facts card.
