If you’ve ever typed up minutes after a four-hour meeting with your eyes half-closed, Anthropic’s Claude AI feels like magic.
It needs the agenda.
It needs the attendance.
It needs the transcript or notes in the right order.
It needs the right permissions if you’re working with sensitive material.
And it needs very clear instructions about what your minutes should look like.
Claude is thoughtful, but it’s not a trained clerk. It doesn’t know your statutes, procedures, templates, or local quirks until you hand them over. You still have to guide it and double-check everything including motions, speakers, votes, formatting, outcomes, and accuracy.
If ChatGPT is the high-school job-shadow kid who shows up in a blazer and insists they totally understand municipal procedure, Claude is your new coworker who reads the entire packet before saying a word. It’s slower, more careful, and far less likely to improvise. Sometimes that’s exactly what you want. Sometimes it means waiting a few extra seconds. But if you need a tool that handles long meetings without drifting off halfway through, Claude is usually the steadier option.
In this guide, we’ll walk through the full Claude workflow for generating minutes step-by-step. We’ll cover how it works, how to access it, and how clerks are using it right now to automate pieces of their workload and produce reliable first-draft minutes. You’ll see every click, every prompt, and the full “DIY” process — from transcript to structured minutes.
If you can use email, you can follow this guide. Nothing here assumes anything more than that.
What Claude can do for minutes today
Let’s cover the basics, especially if Claude feels new or too technical. You upload documents, type instructions, and it returns cleaner, more organized text. However, it doesn’t know your municipality or procedures. It only knows the material you prompt it.
Here’s what makes Claude different from other AI tools.
1. Claude can handle very long meetings without losing the thread.
The newest models, Opus 4.5 and Sonnet 4.5, can read extremely long transcripts in one go. They support context windows of up to 200,000 tokens, which is just a technical way of saying, “You can upload a huge transcript and Claude won’t forget what happened earlier in the meeting.”
This makes a big difference for clerks dealing with:
Four-hour council meetings
Complex planning discussions
Public comment marathons
Budget hearings with multiple speakers
Claude is simply harder to overwhelm.
2. Claude is more careful and less likely to guess.
If ChatGPT is fast and chatty, Claude is slower and more deliberate. It reads your documents more thoroughly. It tries harder to avoid making things up. If something seems unclear, Claude sometimes even says so out loud instead of pretending.
3. Claude supports multiple models depending on your workload.
We recommend using Opus 4.5 for the vast majority of meeting minutes. It is the fastest and most powerful model. Sonnet 4.5 is also a good backup, as it is balanced, accurate, and fast enough.
If you’re generating minutes, stick with Sonnet or Opus.
4. Claude works well with uploaded files.
Just like ChatGPt, you can upload:
PDFs
Word documents
Text files
Transcripts
Scanned pages (sometimes)
Then simply tell Claude what to do with them.
5. Claude does not require installation or special software.
You can use it on any computer through the Anthropic website, the Claude iOS app, or the Claude Android app. Just sign in, upload your files, and start typing.
6. Claude is better than ChatGPT at long, complex discussions.
Its reasoning is stronger, especially for:
Detailed planning and zoning topics
Technical reports
Budget-heavy discussions
Multi-speaker debates
Back-to-back agenda items
It holds structure better and keeps related ideas grouped appropriately, as long as your instructions are clear.
What Claude can’t do well
Even the best Claude models have blind spots. None of these are your fault, they’re simply what happens when a general-purpose AI tries to write official government records.
1. Claude still makes things up when unsure.
Claude will sometimes fill in the blank with something that sounds appropriate but isn’t accurate, especially if someone mumbles, the audio is unclear, a vote isn’t recorded clearly or someone mumbles.
2. Claude rewrites formal or legal language unless you tell it not to.
Its first instinct is to simplify wording. That means ordinances, zoning descriptions, statutory references, motions and boilerplate procedural language may be paraphrased or softened unless you instruct otherwise.
3. Claude can blend agenda items together.
Because it tries to be helpful based on meaning, not sequencing, Claude may merge similar discussions, group public comment with staff reports, or combine back-to-back items into one summary. If you want strict separation, you must tell it.
4. Claude can over-explain or become too cautious.
Clerks have noticed weird issues like:
Extra disclaimers
Long prefaces
Qualifiers like “based on the transcript provided…”
Overly detailed summaries
Responses that sound like they’re written by a policy analyst
This isn’t wrong per se, but it can make minutes longer than needed.
5. Claude does not support voice dictation or image generation.
One of the biggest differences between Claude and ChatGPT and Gemini, is Claude has no voice mode. It also cannot generate images, clean up recordings directly, or create diagrams or charts. It’s strictly a reading, writing and coding tool.
6. Claude is slower.
On average, Claude takes 3–8 seconds to respond to medium requests. Opus is even slower. This isn’t a bug. It’s reading everything carefully. But it does change the feel of the workflow.
7. Claude keeps no audit trail.
It cannot show where each line came from, how it interpreted a motion, and which part of the transcript influenced a section. You are responsible for reviewing and version control.
8. Claude doesn’t follow your local procedures automatically.
You must teach it every time. It doesn’t know your formatting rules, how you record abstentions, how your board handles amendments, your preferred level of detail or your standard legal phrasing
How to generate minutes in Claude
This is the part most clerks tell us they wish someone explained clearly. Claude looks simple. It’s just a chat box and an upload button, but the order you give it information makes all the difference. Claude needs structure upfront so it doesn’t guess or reorganize things on its own.
Step 1: Pick the right Claude model
Just like ChatGPT, the free version of Claude limits message count, file uploads and model power. For minutes, especially long ones, you’ll want the paid plan so you can access Sonnet or Opus without hitting limits. That’s $17 per month.
And similar to ChatGPT, Claude also has multiple models. We recommend using Opus 4.5 for the vast majority of meeting minutes. It is the fastest and most powerful model. Sonnet 4.5 is also a good backup, as it is balanced, accurate, and fast enough.
Step 2: Paste your agenda before anything else
Claude needs to understand the structure of the meeting before it reads the discussion. Otherwise, it will try to infer agenda items based on topic similarities, which can cause blending.
Paste your agenda as plain text in a prompt like this:
“This is the official agenda. Please keep each item separate and do not combine or reorder them.
Agenda Example:
1. Call to Order
2. Approval of Minutes
3. Public Comment
4. Old Business
a. Ordinance 24-11
b. Sidewalk Repair Project
5. New Business
a. Rezoning Request – Maple Ave.
6. Adjournment”
Claude listens closely to instructions like this. This prevents most agenda-blending issues later.
Step 3: Add your attendance list with firm guardrails
Claude will guess names if you don’t give them upfront.
Paste your speaker list right after the agenda. Never later.
Here is an example prompt.
These are the only people who should be listed in attendance. Do not add names that do not appear here.
ATTENDANCE
Present:
- Council President Mark Harris
- Councilor Linda Perez
- Councilor David Kim
- City Administrator John Fields
Absent:
- Councilor Brenda Watts
This prevents it from inventing staff members, adding speakers from previous meetings, and confusing public commenters with officials
Step 4: Upload your transcript or meeting notes the right way
You can upload files using the paperclip icon. Claude accepts PDFs, Word docs, TXT files, CSV, HTML, ODT, RTF, EPUB, JSON and XLSX files.
For minutes, the best approach is upload your written transcript or combined meeting notes as a single file.
Note: If your transcript is extremely long (more than 150 pages), Claude Sonnet or Opus can still read it because of the 200K token window. But, you might find more accurate and helpful results if you break your transcript into chunks.
Then, add this message with your transcript in the prompt.
“I’m about to upload the full meeting transcript. Please read it carefully and wait for all files to finish uploading before responding.
“This transcript corresponds to the agenda you received earlier. Please keep your summary aligned to each agenda item without merging unrelated topics.”
Step 5: Tell Claude exactly what the minutes should look like
Claude is not a clerk. It has no idea what your minutes must include unless you spell it out.
However, Claude responds extremely well to clear, structured instructions. So, you’ll want to use a prompt like this to tell it the right structure and format.
Please draft the meeting minutes using this structure:
- Meeting date, time, and location
- Call to order time
- Attendance list exactly as provided
- Approval of previous minutes (if applicable)
- For each agenda item:
- A short summary of discussion (2–4 sentences)
- All motions, who made them, who seconded, and the vote results
- Amendments, withdrawn motions, or clarifications
- Public comment summarized by topic, not by every individual speaker
- Adjournment time
Please use neutral, factual language. Do not add commentary or interpretation.
Do not invent motions, votes, names, or details not provided in the transcript.
Step 6: Generate the first draft
Now you simply type a prompt like,
“Please generate the first draft of the minutes.”
Then, Claude will read all files, map discussion to agenda items, structure each item cleanly and include motions and votes
Because Claude reads long documents more reliably than ChatGPT, you’ll often notice fewer dropped items or misattributed comments.
However, this process will take a few minutes longer than ChatGPT, but that’s normal. Claude is reading everything more carefully.
Step 7: Identify errors and revise the draft
Claude does best when you give it specific revisions. While it is strongly recommended you review it manually, you can cut down on errors by adding this prompt.
Please reduce the level of detail in public comment.
Please rewrite motions exactly as they appear in the transcript.
Please remove any wording that sounds interpretive.
Please isolate amendments and list them separately beneath the main motion
Please compare each motion in the draft against the transcript and fix any discrepancies.”
This works better in Claude than ChatGPT because Claude handles longer contexts without forgetting earlier sections. But, you’ll still want to check it manually.
Step 8: Final formatting and compliance checks
Claude can help you tidy up formatting:
“Please format the minutes in a clean bulleted structure with agenda numbers bolded.” “Please add the meeting header with the date and time.”
But always do a final review yourself.
Check:
Who made each motion
Who seconded each motion
Votes
Amendments
Order of discussion
Attendance
Adjournment time
Claude is helpful, but it will never replace your judgment or your understanding of local procedure.
When you're satisfied, copy the text into Word or a PDF and finalize it as you normally would.
Real-world Claude use cases for municipal meetings
Clerks often ask: “Okay, but what does Claude actually work well for?”
In real day-to-day clerk work, Claude has some situations where it’s genuinely helpful, and others where it still needs heavy supervision.
Below are the most common meeting types where clerks have tested Claude and what to expect from each one.
City Council meetings
Council meetings are where Claude tends to shine the most, especially if you have:
Long discussions
Policy-heavy agenda items
Back-and-forth between multiple council members
Long public comment sections
Because Claude can read up to 200,000 tokens at once, it’s far less likely to lose the thread of a long conversation. This is where clerks often notice the biggest difference from ChatGPT.
Claude also does well at keeping discussion grouped correctly, maintaining a neutral tone, sticking to agenda order, identifying each motion cleanly, and summarizing big-picture policy debates
But you still need to double-check motions, check who seconded what, confirm the actual vote counts, and remove unnecessary “explanatory” language Claude sometimes inserts
Claude sometimes over-explains or adds extra context around why something might matter. Those lines are easy to delete, but they show up more in council minutes than anywhere else.
Planning & Zoning meetings
These meetings are where Claude is usually the best LLM option. Why? Planning & Zoning discussions are technical, repetitive, filled with jargon, often 2–3 hours long and centered around complex reports
Compared to ChatGPT, Claude is also better at recognizing when an amendment changes the intent of a motion, retaining parcel details or project names, and keeping technical language stable (unless you tell it to simplify)
Planning & Zoning is where Claude feels the most “worth it,” as long as you give it a clean transcript.
Committee meetings
Committee meetings vary widely. Some are casual, some are highly structured. Claude is generally helpful when the committee follows a regular agenda, discussions stay somewhat on-topic, and the transcript is clean
Claude struggles when the committee jumps around between items, members speak informally or interrupt each other, and discussions sprawl with no clear transitions
It also does well at summarizing routine updates, project briefings and status reports. But you’ll need to tighten the write-up afterward because Claude sometimes produces more detail than you actually want in committee minutes.
Meetings with heavy public comments
Claude is better than ChatGPT at grouping similar comments together.
However, you need to give it explicit instructions in your prompts. If you don’t, Claude may list each comment individually, get names wrong, or reorganize comments by theme rather than by order (not always desired)
Claude is solid here, but based on clerk feedback, public comment is still one of the places where manual oversight is essential.
Compliance and risk considerations of using Claude for minutes
Claude is more careful and more accurate than ChatGPT, but it’s still an all-purpose AI. It doesn’t know municipal law, open meeting rules, or your retention policies.
That means there are some key risks to consider.
1. No audit trail. If a resident disputes something, there’s no built-in “proof” of how the AI produced that line.
2. No version history. Claude can’t track edits inside the document the way ClerkMinutes can.
3. Data privacy. Claude’s policies are strong, but it is still a cloud service. Some municipalities require special review before using third-party AI tools.
4. It will still guess. Even careful guessing is still guessing and that’s not good enough for official records.
5. Accessibility rules still apply. Claude doesn’t ensure your final minutes meet required formatting or accessibility standards.
None of these points make Claude bad. They just mean Claude is great at helping clerks move faster, but it is not a replacement for clerk judgment or compliance.
How ClerkMinutes simplifies all of this
Claude is an excellent assistant. It is not a municipal workflow. And that’s exactly where ClerkMinutes is different.
ClerkMinutes isn’t a creative tool you have to babysit. It’s built for one job. That’s turning a meeting into accurate, publish-ready minutes with as little friction as possible. All the rules clerks follow every day are built into the AI workflows. You don’t have to teach it how your meetings work or remind it not to invent things.
Where Claude needs careful supervision, ClerkMinutes starts by removing the parts that cause supervision in the first place.
Here’s what that looks like.
1. You don’t have to prep a transcript, chunk files, or explain the agenda in advance
Claude only works if you feed it everything manually, in the perfect sequence. ClerkMinutes doesn’t need any prompting.
You simply:
Upload your agenda and meeting recording.
Assign speakers.
Wait a few minutes
Review the automatic minutes.
That’s it. ClerkMinutes automatically identifies agenda items, detects transitions, separates topics correctly, and organizes discussion under the right headings
2. Motions, seconds, votes, and amendments are recognized automatically
Claude can summarize discussion, but it doesn’t inherently understand parliamentary procedure. You still have to tell it what a motion looks like and how to format it.
ClerkMinutes, on the other hand detects every motion, captures who made it, captures who seconded it, captures amendments, captures substitute motions, captures withdrawn motions, records the vote correctly, and attaches outcomes to the right agenda item
You don’t have to hunt through a transcript or prompt it repeatedly to “fix the motions.” ClerkMinutes treats motions as first-class citizens in the workflow — because they are.
3. Speakers are identified for you
Claude cannot reliably identify speakers unless you give it a labeled transcript. Even then, it sometimes confuses similar names or uses the wrong label.
Whereas in ClerkMinutes, all you need to do is assign speakers once at the start. Then, it will identify speakers automatically and keeps those names consistent through the entire meeting.
4. Formatting follows your municipality’s conventions automatically
Claude formats minutes however you tell it to, which is helpful but you have to give it the instructions every single time.
ClerkMinutes applies your heading structure, your indentation rules, your level of detail, your formatting conventions and your standard minutes layout
Once you finalize your style once, you don’t need to tell the tool again.
5. ClerkMinutes never invents information
Claude is careful, but it still fills in blanks when transcripts are unclear. ClerkMinutes doesn’t do that. If something is missing, ClerkMinutes will flag it and prompt you to verify. It will not guess names, votes, outcomes, or motion language.
That one difference alone is why many clerks use LLMs for drafts but use ClerkMinutes for anything that needs to be correct.
6. Review takes minutes, not hours
Claude produces a draft, but drafts still require line-by-line review.
In ClerkMinutes you review it all inside one unified screen. You approve sections, adjust text, and publish the final minutes.
7. You get an automatic audit trail
Claude gives you a draft, but nothing else. There’s no record of where each line came from.
ClerkMinutes provides version history, timestamps, source references, etc. If a resident challenges the minutes, you can show exactly when a statement happened and who said it.
8. ClerkMinutes reduces the chance of human error
Claude reduces typing. ClerkMinutes reduces risk.Because ClerkMinutes is designed specifically for clerk workflows, you avoid missing motions, mis-numbering agenda items, blending unrelated topics, formatting inconsistencies and procedural errors.
How to generate minutes in ClerkMinutes
After seeing everything that goes into using Claude, including selecting the right model, pasting the agenda, uploading transcripts correctly, giving step-by-step instructions, revising drafts, and cleaning up formatting, it helps to see how ClerkMinutes works and why it’s often a much faster and more accurate way to generate minutes.
That’s because ClerkMinutes is built specifically for municipal workflows. You don’t have to train it or write prompts and secretely hope the AI “understands what you meant.”
Here’s the entire process from start to finish.
Step 1
Upload Your Agenda
Just drag and drop your agenda into ClerkMinutes. That’s it. You don’t have to paste the text into a chat window or explain the structure item by item. ClerkMinutes reads the agenda automatically and uses it to create the outline of your minutes.
Step 2
Upload Your Meeting Recording
Upload the audio or video recording of your meeting via Zoom, YouTube or a local file. If you already have a transcript, you can upload that too.
Unlike Claude, you don’t have to break the file into chunks, warn the model not to merge items, label anything manually, or paste supporting documents in a specific order
Step 3
Assign Speakers
ClerkMinutes gives you a clean transcription of the meeting and identifies speakers for you.
You simply just need to tell it the speakers when you upload the recording. Then, confirm who’s who, and correct anything mislabeled with a single click once the minutes are generated.
There’s no guessing, no model hallucinating new names, and no “please rewrite this with corrected speakers” loop you sometimes get with Claude.
Once confirmed, ClerkMinutes keeps those names consistent through the entire meeting.
Step 4
Generate your minutes and customize as needed
With your agenda, recording, and speaker assignments in place, ClerkMinutes drafts your minutes for you. They come out already aligned to your agenda, separated into the right sections, and formatted the way municipal minutes should look
From there, you can use ClerkMinutes’ Custom Instructions feature to adjust a sentence, add context, clarify a motion, revise a vote or refine public comment.
Once everything looks correct, download your minutes as a Word document or PDF and you’re done.
So when should you use Claude vs. ClerkMinutes?
Now that you’ve seen both workflows side by side, the pattern becomes pretty clear. Claude is powerful, but it still behaves like an all-in-one tool. It can summarize, rewrite, reorganize, and help you think through meetings quickly, but only after you feed it everything in the right order, give it the right instructions, and double-check the output carefully.
ClerkMinutes, on the other hand, is built for the exact work clerks do every day. It already understands agendas, motions, seconds, amendments, votes, and municipal formatting. That’s why the workflow is shorter and the review process is more predictable.
Use Claude when:
You already have a clean transcript.
You’re comfortable supervising and correcting the output.
You want help summarizing a discussion before drafting minutes.
You need a quick rewrite of an explanation, background section, or staff report.
You’re working on something non-official, like a recap email or FAQ.
You’re experimenting and don’t mind guiding the model carefully.
Claude is great for thinking, outlining, and reshaping language. It’s also good for long policy discussions because of its very large context window. But it still needs your hand on the wheel the entire time.
If the risk is low and you’re mainly using it for convenience, Claude can save you time.
Use ClerkMinutes when:
You need publish-ready, accurate minutes.
You want the agenda structure to be followed automatically.
You want motions and votes captured correctly the first time.
You want to avoid hallucinations or invented details.
You don’t want to label and double check speakers manually.
You don’t want to write long prompts.
You want predictable formatting that matches your municipality.
You want an audit trail linked to your minutes.
You have multiple meetings per month and want a repeatable, reliable workflow.
ClerkMinutes is for the work that needs to be right
If you want to see what your minutes look like in ClerkMinutes, you can try it for free for 14 days by uploading your first meeting here.
HeyGov, Inc.
Sturgeon Bay, Wisconsin
Toll Free: +1 888 HEYGOV1