Using AI to turn your team's WhatsApp or Slack messages…

McKinsey estimates employees spend 28% of the workweek reading and responding to messages — a significant chunk of which contains decisions and action items that disappear into the scroll. This post walks you through the exact steps to export your WhatsApp or Slack chat history, feed it into an AI model, and use AI to summarize WhatsApp messages and extract action items into a clean, structured record of what was decided and who owns what. Done consistently, this eliminates the "wait, did we agree to that?" conversations that quietly eat hours every month.

What you need before you start

ChatGPT{:target="_blank"} — OpenAI's AI assistant, capable of processing pasted or uploaded text and returning structured output. The free tier works for short chats, but GPT-4o requires a ChatGPT Plus{:target="_blank"} subscription at $20/month (pricing checked January 2026). For long exports, this is the plan you need.

Claude{:target="_blank"} — Anthropic's AI assistant, a solid alternative to ChatGPT for this task. Claude Pro{:target="_blank"} is $20/month as of January 2026 and gives access to Claude 3.5 Sonnet and Claude 3.7, both of which handle structured extraction well.

Google Gemini{:target="_blank"} — Worth mentioning separately because Gemini 2.0 Flash supports a context window up to 1 million tokens, making it the most practical option if you're dealing with months of dense group chat history. Available via Google One AI Premium{:target="_blank"} at $19.99/month (pricing checked January 2026).

Time required: 10–15 minutes for the manual method the first time. Under 5 minutes per session once you have your prompt saved. Automated pipeline setup via Zapier or Make: 1–3 hours.

Skill level: The manual method requires no technical skills — just the ability to export a file and paste text. The automation options require a Zapier or Make account and comfort with basic workflow builders.

How to Use AI to Summarize WhatsApp and Slack Group Chat Decisions

For WhatsApp

Open the WhatsApp group chat you want to export.
Tap the three-dot menu (Android) or the group name at the top (iOS) to open chat settings.
Select More > Export Chat. Choose Without Media — the file will be smaller and the AI doesn't need images to extract decisions.
Save or share the resulting .txt file to your device. WhatsApp's official export instructions{:target="_blank"} confirm this works on both iOS and Android.
Open your AI tool of choice. In ChatGPT or Claude, use the file upload option to attach the .txt directly, or open the file and copy-paste the relevant date range of messages.

Here's the catch: WhatsApp's standard (non-Business API) version has no automated extraction. Every export is manual. If your team uses WhatsApp heavily, plan for this as a weekly 5-minute task, not something that happens automatically.

For Slack

Open the Slack channel or DM thread you want to summarise.
For the manual method, scroll to the relevant date range and copy the messages directly. Slack doesn't offer a native text export on free plans without third-party tooling.
Paste the copied messages into your AI tool.

Note: Slack's free plan{:target="_blank"} retains only 90 days of message history as of 2024. If your team is on the free plan, run this summarisation process before the 90-day cutoff or you will permanently lose context on decisions made in older threads.

Write a Prompt That Actually Extracts Decisions and Action Items

The quality of your output depends almost entirely on the specificity of your prompt. A vague "summarise this chat" will return a vague paragraph. The four-part structure below consistently produces usable, structured output across GPT-4o, Claude 3.5 Sonnet, and Gemini 2.0 Flash.

Prompt template — copy and save this:

"Below is an exported team chat from [date range]. Please do the following:

Decisions made — List every clear decision that was reached, even if informal. Quote or paraphrase the key message.

Action items — List every task or commitment someone agreed to do. Include the name of the person responsible and any deadline mentioned.

Unresolved questions — Flag anything that was raised but not concluded. These may need a follow-up.

Key context — Note any important background information referenced in the chat that future readers would need to understand the decisions.

Format the output as a structured list under each of these four headings. Do not include small talk or filler messages.

[Paste chat text here]"

What to expect: a clean, four-section output that typically runs 200–500 words for a week of moderate team chat. Verify it by scanning for at least 2–3 specific decisions you remember making — if they're present and accurate, the extraction is working. If they're missing, the relevant messages may have been too informal in phrasing; adjust future conversations or add a "Decision:" tag in the chat itself to make the AI's job easier.

Skipping the four-part structure and writing a vague prompt is the most common reason people get garbage output and conclude "AI can't do this." The structure is what makes the extraction reliable.

When Something Goes Wrong

The output lists tasks but no names are attached. Root cause: the chat messages didn't include names before each message — this can happen with some WhatsApp export formats or when messages are copied manually. Fix: before pasting, do a quick find-and-replace to ensure each message block starts with a name. Most WhatsApp exports already format as "[date, time] Name: message", but if yours doesn't, a few minutes of manual cleanup resolves it.

The AI hallucinates a decision that wasn't made. Root cause: long, ambiguous discussions where the AI infers consensus from back-and-forth rather than an explicit agreement. Fix: add a line to your prompt — "Only list decisions where there is a clear indication of agreement, not just discussion." This tightens the output significantly.

Gemini or ChatGPT truncates a long export. Root cause: the message history exceeds the model's effective context window. Fix: split the export into weekly chunks before pasting, or use Gemini 2.0 Flash specifically — its 1 million token context window handles even months of dense chat without truncation.

Where to Store the Output So Your Team Can Actually Find It

A summary sitting in your personal ChatGPT history helps no one but you. The output needs to land somewhere the team can search and reference it.

Three practical options, in order of effort:

Notion{:target="_blank"} — Paste the AI output into a Notion page. Notion AI{:target="_blank"} (add-on at $10/member/month as of January 2026, or $8/member/month on annual billing) can then reformat it, create tables, or surface related decisions from older pages. Notion's full-text search also gives you semantic proximity that Slack's keyword search lacks.
Google Docs — A shared "Team Decisions Log" doc, updated weekly, is the lowest-friction option. Not fancy, but searchable and version-controlled. Free.
Trello or Asana — If action items are your primary focus, paste them directly into your project management tool as cards or tasks. This gets action items into your existing workflow rather than creating a separate document to maintain.

Semi-Automated Options for Teams Who Won't Do This Manually Every Week

The honest answer about the manual method: most teams abandon it within a few weeks without a workflow or automation. If that sounds like your team, consider these options.

Slack AI{:target="_blank"} — Slack's native AI add-on can summarise channels and threads and surface action items. Cost: approximately $10/user/month (pricing checked January 2026), available on Pro plan and above only. The trade-off is that this solves the Slack problem but does nothing for WhatsApp. If your team runs entirely in Slack and is already on a paid plan, this is the lowest-friction path.

Zapier{:target="_blank"} + OpenAI or Anthropic API — Build a Zap that triggers on new Slack messages (or a scheduled channel export), sends the text to the OpenAI or Anthropic API with your saved prompt, and pushes the structured output to Notion, Google Docs, or Trello. No coding required, but budget 1–3 hours for initial setup. Zapier's guide to auto-summarising notes{:target="_blank"} covers the core workflow pattern. Zapier's Professional plan starts at $49/month (pricing checked January 2026); the Starter plan at $19.99/month may cover lighter usage.

Make{:target="_blank"} (formerly Integromat) — Similar capability to Zapier, generally lower cost for higher-volume automation. The free tier includes 1,000 operations/month, which is sufficient for a small team running weekly summaries. The Core plan is $9/month (pricing checked January 2026).

Notion AI{:target="_blank"} — If your team already uses Notion, this is underrated. Paste a chat export into any Notion page, highlight it, and use Notion AI to generate a structured summary in-place. It stays inside your existing tool, the output is immediately searchable, and there's no pipeline to maintain.

For heavy WhatsApp users: Respond.io{:target="_blank"} and Wati{:target="_blank"} are WhatsApp Business API platforms that include built-in AI summarisation and tagging as of 2025. These are worth evaluating if WhatsApp is your primary team communication channel and you're processing high volumes — but both require setup against the WhatsApp Business API and involve meaningful per-message or monthly costs.

Privacy and Data Risks Before You Paste Team Conversations Into AI

This section is non-negotiable to read before you start. Pasting team chat logs into consumer-tier ChatGPT or Claude.ai means that data may be used for model training by default, unless you explicitly opt out in your account settings. For most small business conversations about schedules and task assignments, this is low risk. For conversations involving financial data, client details, legal matters, or personnel issues, it is not.

Your options in order of increasing protection:

Opt out of training data use — both OpenAI and Anthropic offer this in account settings. Do this before you paste anything.
Use ChatGPT Team or Enterprise — ChatGPT Team{:target="_blank"} at $30/user/month (annual billing, pricing checked January 2026) explicitly excludes your conversations from training data.
Use the API directly — API usage is not used for model training by default on either OpenAI or Anthropic. If you're building a Zapier or Make workflow, API-based processing is inherently safer than consumer interfaces.
Strip identifying information — before pasting sensitive conversations, do a find-and-replace to remove full names, client company names, and financial figures. This reduces exposure even on consumer plans.

The trade-off is real: the most accessible method (free consumer ChatGPT) has the weakest privacy protections. If your chats contain anything sensitive, take 10 minutes to configure privacy settings or step up to a Team plan.

Building a Lightweight Weekly Habit So This Actually Sticks

Tools don't fail. Habits do. The most common outcome with this workflow is that a team tries it once, gets a great summary, doesn't build a consistent process, and abandons it by week three.

The fix is making this someone's explicit 10-minute responsibility, on a specific day, linked to something that already happens:

Assign one person to own the weekly chat summary. Not "everyone" — that means no one.
Tie it to a fixed day — Monday morning before standup, or Friday before close. The decision review becomes part of the ritual, not an extra task.
Create a shared doc template — a Google Doc or Notion page with a table: Date | Decisions | Action Items | Owner | Deadline. The AI fills the content; the person pastes and reviews.
Review it in your next team meeting — even for 5 minutes. This closes the loop and makes the habit feel useful rather than administrative.

If you automate via Zapier or Make, set the automation to trigger on a fixed schedule and post the output to a dedicated Slack channel or Notion page. That way the summary appears without anyone having to remember — the habit is baked into the system.

What to Do Next

If you've never tried the manual method, do it once this week with last week's most active chat. Use the prompt template above, paste the output into a shared Google Doc, and bring it to your next team meeting. That single test will tell you whether the output quality justifies building the habit or an automation around it.

Once you've validated the output, consider building the Zapier + OpenAI pipeline to make it automatic — the setup investment pays off within the first month for any team with more than three active channels.

For related reading: how to use AI to run more effective team meetings, and how to automate task management with AI and Zapier.

FAQ

Can I use AI to summarize WhatsApp messages for free? Yes, with limitations. The free tier of ChatGPT handles short exports but struggles with long or complex chats. GPT-4o — which produces substantially more accurate structured output — requires ChatGPT Plus at $20/month (pricing checked January 2026). For a team running this weekly, the cost per summary is roughly $0.50–$1.00 of a monthly subscription, which is negligible against the time saved.

How do I use ChatGPT to extract action items from WhatsApp messages? Export your WhatsApp chat as a .txt file, then upload or paste it into ChatGPT with a structured prompt that asks for decisions, action items, unresolved questions, and key context under separate headings. The four-part prompt template in this post is the fastest way to get reliable output. GPT-4o on a Plus plan handles most weekly exports without issue.

How long a chat export can I feed into an AI model? It depends on the model's context window. Claude 3.5 Sonnet and GPT-4o handle most weekly exports comfortably. For exports covering months of a busy group chat, Gemini 2.0 Flash is the practical choice — its 1 million token context window is one of the largest available in a consumer-accessible model as of early 2026, and it handles long exports without truncation.

Is it safe to paste employee conversations into ChatGPT? Consumer ChatGPT may use your conversations for model training unless you opt out. Go to Settings > Data Controls and disable model training before pasting anything work-related. For higher-stakes conversations — anything involving financials, personnel, or client data — use ChatGPT Team{:target="_blank"} ($30/user/month, annual billing) or Claude for Enterprise, both of which explicitly exclude your data from training.

Does Slack have a built-in AI tool for summarising messages and decisions? Yes. Slack AI{:target="_blank"} is a paid add-on at approximately $10/user/month (pricing checked January 2026), available on Slack Pro and above. It summarises channels and threads natively without any export or copy-paste required. The trade-off: for a team of five, that's $50/month on top of your existing Slack subscription — versus $20/month for a single ChatGPT Plus account that covers the whole team if one person runs the summaries manually.

What's the ROI of doing this consistently for a small team? I don't have a study specific to this workflow, so I won't cite a number I can't verify. What I can say from the McKinsey data: if your team of five spends 28% of the workweek on messaging, and even 10% of that involves re-litigating decisions that were never formally recorded, that's roughly 1.1 hours per person per week — 5.5 team hours — spent on conversations that a weekly summary would have prevented. At a modest $30/hour blended rate, that's $165/week or $8,580/year. A ChatGPT Plus subscription costs $240/year. The numbers say this is worth doing.

Using AI to turn your team's WhatsApp or Slack messages into a searchable written record of decisions and action items