A consulting firm prepares client status updates faster with AI by gathering the week's project evidence into one place, having the AI fill a fixed template from that evidence only, and keeping a person responsible for the status rating, risks, and sponsor asks. You save time on assembly and first drafts. You don't save it by skipping the review.
The workflow below is meant to repeat every week on every engagement. It uses the tools and meetings you already have. Each section answers one part of the job: inputs, template, prompt, review, client separation, decisions, and measuring the result.
What should the AI draft from?
An AI client status update should be drafted only from evidence you can point to: task activity from your project tool, the RAID log, meeting notes, and decisions recorded since the last update. If a fact isn't in those inputs, the draft shouldn't contain it. Gather the inputs first and the drafting step gets short.
Pull these for the reporting period:
- Task activity. Completed, started, and overdue tasks from your project tool, exported or copied as a list with owners and due dates.
- RAID items. New or changed risks, assumptions, issues, and dependencies, with owners.
- Meeting notes. Notes or transcripts from client and internal meetings, clearly labelled with which is which.
- Decisions. Decisions made, and decisions still waiting on the sponsor.
- Last week's update. So the AI can say what changed rather than repeating everything.
When sources disagree, decide in advance which one wins. Enterprise DNA's guide recommends ranking sources by how far to trust them and flagging conflicts between them. A simple rule: the project plan beats chat, and a recorded decision beats a remembered one. Write the rule down so the AI and your reviewers apply it the same way.
If you have no integrations, paste the inputs in by hand. It's slower to gather, but the rest of the workflow is the same. Deltek makes the wider point that AI only helps when it runs on the firm's own data, and that cleaning data and standardising workflows comes first.
How do I set up a template the AI can fill?
A status update template gives the AI fixed sections to fill and gives the sponsor the same layout every week. Keep the template short, put what you need from the sponsor near the top, and agree the layout with the sponsor once. Then the AI is filling slots from evidence, not inventing a structure.
AdAI's guide suggests a standard structure of executive summary, progress against plan, key decisions made, risks and issues, asks of the client, and the next two weeks. Here is a version you can copy:
Subject: [Client] [Project] status, [date]
Overall status: [Green / Amber / Red], set by [lead name]
Summary: [A few sentences on the most important development this period]
Needed from you
- [Decision or review], by [date]. If we don't hear by then: [consequence]
Progress against plan
- [Deliverable]: [done / on track / at risk], evidence: [link or reference]
Decisions made
- [Decision], made by [name] on [date]
Risks and issues
- [Risk], owner [name], next step [action]
Next period
- [Planned work], owner [name]
Put "Needed from you" above progress. The ask is the part that changes what happens next, so place it where the sponsor sees it first.
What prompt should I give the AI?
A good status update prompt tells the AI its role, the task, the evidence it may use, and the exact output format, and forbids anything outside the evidence. The four-part prompt in this Claude Cowork article uses role, task, context, and format. Add rules for sources and gaps.
Copy and adapt:
You are a project lead writing a weekly status update for [client sponsor name].
Task: fill the template below using only the evidence provided.
Evidence: [paste task activity, RAID items, client meeting notes, decisions, last update]
Rules:
- Use only facts in the evidence. After each progress line, name the source item.
- If a section has no evidence, write "[No update: check]" instead of guessing.
- Leave Overall status blank for me to set.
- Do not include anything from notes labelled INTERNAL.
- Plain, factual tone for [client sector]. No reassurance the evidence doesn't support.
Format: [paste template]
Enterprise DNA's blog post describes three things a status agent needs: data access, a library of templates, and voice guidelines. Keep a short voice note per client in your prompt, such as how formal they like it and which terms they use.
What must a person check before it goes to the client?
A person must check the status rating, every risk and blocker, every ask to the sponsor, and anything that names a person or a commitment. AI drafts tend to sound smoother than the project really is, so the review targets accuracy and candour. AdAI notes that AI cannot guess what counts as a risk.
Use this checklist on every draft:
- Status rating. Set it yourself. Ask: would I say this colour out loud to the sponsor in a meeting?
- Traceability. Spot-check progress lines against their named source. Remove any line you can't trace.
- Risks. Compare the draft against your RAID log. Add anything the AI left out, especially items that are uncomfortable.
- Asks. Confirm each ask has a clear decision, a named sponsor who makes it, and a date.
- Internal material. Make sure staffing issues, margin, and internal debate stayed out.
- Gaps. Resolve every "[No update: check]" marker, by finding the fact or deleting the line.
Enterprise DNA's guide routes every draft to the engagement manager for approval and lists triggers that should force a human review. Write your own triggers, for example a status change, a new red risk, or a slipped milestone, and have a senior person read those updates too.
How do I keep each client's data separate?
Keeping client data separate means every client gets its own workspace, project, or conversation in your AI tool, and nothing from one engagement is pasted into another. Check your contracts and NDAs before any client data goes into a third-party AI tool, and use only tools your firm has approved.
Practical rules:
- Start a fresh conversation or project for each client, every time. Never reuse a thread that contains another client's notes.
- Name files and notes with the client in the title so mix-ups are visible.
- Label internal notes INTERNAL and tell the AI to exclude them, as in the prompt above. Enterprise DNA also recommends keeping internal issues separate from client issues.
- Ask your IT or security lead which retention and training settings apply to your AI tool, and record the answer.
- Remove personal details the update doesn't need before pasting.
How do I turn the update into a decision request?
A status update becomes a decision request when the ask is specific, the options are written out, and the cost of waiting is stated. Write each ask so the sponsor can answer in a line: what you need, the options, your recommendation, the date, and what happens if the date passes.
Wording you can use:
We need your decision on [topic] by [date]. Options: [A] or [B]. We recommend [A] because [reason]. If we don't have a decision by [date], [deliverable] moves to [new date].
Follow up before the date, not after it. Unblockd sends asks to the sponsor by email with one-tap answers (no account), or answered by replying to the email. Whatever you use, record the response in your decision log so next week's draft has it as evidence, and tell the team what the decision means for their next step.
How do I know the workflow is faster?
You know an AI status workflow is faster by timing it on your own engagements, not by trusting vendor figures. Before you change anything, note how long gathering, drafting, and review take. Then run the new workflow on an engagement you know well and compare.
Track three things for each update: time to gather inputs, time to review and edit the draft, and corrections the reviewer had to make. If corrections stay high, fix the inputs or the prompt before rolling out further. If gathering is the slow part, tidy the project tool and RAID log first. Expand to other engagements once reviewers trust the drafts, and keep the template, prompt, and checklist in a shared place so every lead uses the same version.