How to Write a Client Report With AI - and Get the Client's Comments on It (2026)
By Marcin, founder of Markloop · September 2026
Short answer: give ChatGPT or Claude your real data and let it draft the structure, the summary and the wording, as one designed HTML page. Check every number against the source yourself. Then send the client one link they can comment on, and let your AI turn their comments into the next version at the same link - which also works as one link for a whole monthly engagement.
What AI does well in a client report, and what it can't know
A client report is mostly organizing: turning exports, notes and numbers into something a busy person reads in five minutes. AI is good at that. It doesn't know which numbers are right, which finding matters most to this client, or what you can deliver next month.
| Let the AI do it | Keep it yours |
|---|---|
| Turn raw exports into tables and charts | Where the numbers come from, and whether they're right |
| Write a three-line summary a manager will read | Which finding matters most to this client |
| Group issues and label impact and effort | What you recommend, and what it costs |
| Keep the same structure month after month | Anything you promise: dates, owners, results |
Step 1 · Give it the real data
Export what the report is based on - a CSV from Search Console or your analytics tool, a crawl summary, your call notes - and paste or attach it. Then set the rules:
Write a Q3 SEO audit for Brightline Dental from the attached data. Use only numbers from the data. If something is missing, write [TODO] instead of estimating. For every finding give the evidence, why it matters, the fix, and an impact label (High, Medium, Low).
The [TODO] rule matters more than any other line in the prompt. Without it, a model fills gaps with numbers that look plausible.
Step 2 · Structure it for someone who reads the first screen
Most clients read the summary and the first finding, and skim the rest. Put what they need to decide at the top, and push raw data to the end:
Ask for it as one page:
Turn this into one standalone HTML page with inline CSS: a three-line summary at the top, one card per finding with an impact label, a next-steps table with owner and date, and the raw data in a collapsed appendix. No external images or scripts.
One HTML file keeps the layout and charts exactly as the AI made them and opens in any browser. If the client needs a PDF for their files, you can still produce one at the end.
Step 3 · Check the numbers before the client does
A wrong number in a report costs more trust than a typo in a proposal. Two prompts help, and then you read it yourself:
List every number in this report with the row of the source data it came from. Flag any number you can't trace.List every claim that isn't backed by the attached data.
Also check the recommendations against your capacity. The model will happily suggest a site-wide rewrite for next week.
Step 4 · Send one link the client can comment on
A PDF attachment gets a reply like "looks good, but can we do the second thing first?", and you're left guessing which second thing. On a comment link, the client puts the question on the finding it's about. In the example above, Ben asked to move the schema fixes up because new services launch in October. You replied in the same thread and changed the plan.
With Markloop there are two ways in. No setup: download the HTML file your AI made and upload it. Or connect your AI once (Claude, ChatGPT, Cursor or Grok; setup is a one-time step, shown in the app after signup) and ask it:
Publish this audit to my Markloop project as a new document.
Turn on comments for the link and send it. The client types a name and comments, no account needed. You can give the link an expiry date or revoke it later, and you get one email per document summing up new comments. Who can see the link and where the data lives is covered in the FAQ, and the 2-minute walkthrough shows the whole round.
Step 5 · Publish the next version at the same link
When the comments are in, go back to the same chat:
Pull the comments on the Brightline audit. Move the schema fixes to week 1 in the next-steps table, update the summary, and publish the next version.
Your AI gets each comment with the text it was left on, so "the schema fixes" is never ambiguous. It edits the page on your machine, publishes v2 at the same link and marks the comment as addressed. Anything it didn't address stays open for the next round. Markloop doesn't email the client about new versions, so send a one-line "v2 is up" note.
Monthly reports: one link for the whole engagement
For a retainer, don't start a new document every month. Publish each month's report as the next version of the same document. The client keeps one bookmark, and you keep a full history of what was raised and what was addressed.
A few things to know. The link always opens the latest version. Older versions stay in your history, and the client sees them only if you turn version history on for that link (it's off by default). Comments stay with the version they were made on, so July's discussion doesn't clutter September's page. When the client needs a copy for their files, "Print / save as PDF" gives a clean PDF of any version, without the comments.
More prompts for client reports
Rewrite the summary in three sentences for a CEO who hasn't read last month's report.Compare this month's numbers with last month's and add a "what changed" line under each chart.Add a "what we need from you" section: decisions, access and content the client owes us.Read this as the client's CFO. Which recommendation would they question, and why?
FAQ
Can ChatGPT or Claude write a client report from my data?
Yes, if you give it the actual exports and tell it to use only numbers from the data. It's good at structure, summaries and turning tables into charts. You still check every number against the source and decide what to recommend.
Does the client need an account to comment?
No. They open the share link, type a name and comment. The name isn't verified, so if you need to know exactly who said what, invite them to the project by email instead; they sign in with an account.
Can the client still get a PDF?
Yes. Print / save as PDF gives a clean PDF of any version, without the comments. The comments can be exported separately as plain files.
Should I send a new link every month?
No. Publish each month as a new version of the same document. The link always opens the latest report, and the client can see earlier months if you turn on version history for the link.
Send this month's report as one link
Publish it from ChatGPT or Claude, let the client comment on the findings, and have your AI turn the comments into the next version.
14-day free trial · No credit card required · Reviewers always free
Related: Claude artifacts explained · How to write a client proposal with AI · HTML to Google Docs: what breaks, and when to send a link · HTML is the new Markdown
