How to Triple Your Productivity with AI: A Complete Real-User Case Study

“Triple your productivity” sounds like marketing. This article does not invent an all-powerful AI user, and it does not claim that every hour of an entire job becomes three times faster. It uses the documented experience of STADLER co-CEO Julia Stadler and her organization, combined with controlled research and a reproducible time-audit model, to explain where 3x gains are realistic, how to build them, and what should never be delegated to AI.

After using AI for a while, many professionals experience a contradiction:

- Emails are faster;

- Research feels easier;

- The calendar is still full;

- Total workload has not fallen;

- Reviewing AI output sometimes creates new work.

The problem is usually not model intelligence. It is that AI remains an occasional chat box rather than part of the workflow.

Meaningful productivity gains come from redesigning the chain:

```text

Raw information

→ AI capture and retrieval

→ human objectives and standards

→ AI structure and first draft

→ automated transfer and reuse

→ human fact, judgment, and accountability review

→ reusable template

```

If AI writes one paragraph, a task may become 30% faster.

If it also helps capture sources, structure information, draft, reformat, repurpose, and route the output, a repeatable package of knowledge work can approach three times the original productivity.

---

1. The verdict: 3x is possible, but not across an entire job

In this article, 3x productivity means:

Completing the same repeatable package of knowledge tasks, at the same quality standard, with roughly one-third of the original human time.

It does not mean:

- Turning a 40-hour job into 13 hours;

- Delegating accountability to a model;

- Automating every meeting and decision;

- Receiving equal benefit in every occupation;

- Becoming faster merely by buying more tools.

The strongest 3x candidates include:

- First-pass research;

- Document and report drafts;

- Meeting preparation, notes, and action items;

- Repetitive email and customer responses;

- Spreadsheet cleanup and exploratory analysis;

- Rewriting, translation, and multi-channel repurposing;

- Weekly and monthly reporting;

- Searching historical organizational knowledge.

Work that is less likely to reach 3x includes:

- High-stakes final decisions;

- Negotiation;

- Conflict management;

- Cross-functional coordination;

- Physical and field work;

- Original strategic judgment;

- Professional legal, medical, or financial conclusions;

- Rare problems with weak model training data.

---

2. The real case: from half a day for a draft to a usable version in about 20 minutes

The primary case is STADLER, a German industrial company with more than 230 years of history.

STADLER builds automated waste-sorting systems, operates globally, and employs more than 650 people. Co-CEO Julia Stadler helped make AI an everyday productivity layer rather than an isolated experiment.

By March 2026, the company reported:

A representative change described by Julia Stadler was:

- A solid first version previously required roughly half a day;

- AI could produce a structured, usable draft in about 20 minutes;

- People then spent their time on facts, judgment, tone, and improvement rather than a blank page.

AI did not reduce the final deliverable to 20 minutes.

It reduced:

The time from zero to a workable first version.

Final quality still depended on:

- Company context;

- Professional knowledge;

- Data verification;

- Approval;

- Human judgment.

STADLER’s disclosed average was 2.5x faster drafting, not 2.5x across all work. Some repetitive content workflows reached 6x, while common knowledge tasks saved 30%–40%.

That is the correct way to interpret a 3x claim:

Do not expect every task to average 3x. Identify the large, repeatable work packages that can be improved by 3x–6x.

---

3. What the research supports—and what it does not

Professional writing: 40% less time, 18% higher quality

MIT researchers ran an experiment with 444 college-educated professionals.

For suitable writing tasks, access to ChatGPT:

- Reduced average completion time by 40%;

- Increased independently evaluated quality by 18%.

For defined, text-heavy, moderately complex tasks, AI can improve speed and quality at the same time.

Consulting tasks: 25.1% faster, more tasks completed, higher quality

Harvard Business School and BCG conducted a preregistered experiment involving 758 knowledge workers.

Across 18 tasks inside the AI capability frontier, AI users:

- Completed 12.2% more tasks;

- Worked 25.1% faster;

- Produced significantly higher-quality results.

On one complex managerial task outside the frontier, AI users were 19% less likely to reach the correct answer.

This is the “jagged technological frontier”:

- Two tasks can appear similar;

- AI can perform one extremely well;

- It can produce a confident error on the other.

Real customer support: 15% average improvement, larger gains for novices

A 2025 Quarterly Journal of Economics study examined 5,172 customer-support agents.

It found:

- A 15% average increase in successfully resolved issues per hour;

- Roughly 30% improvement among novice and lower-skill workers;

- Limited speed gains and small declines in some quality measures among the most experienced workers.

One major benefit of AI is transferring the practices of strong performers to the rest of the organization.

For experts, however, blindly following average AI recommendations can weaken distinctive judgment.

Enterprise use: typically 40–60 minutes saved per active day

An OpenAI survey across almost 100 enterprises reported that active ChatGPT Enterprise users attributed 40–60 minutes of time saved to AI per working day.

Vodafone’s legal analysis found an average of four hours saved per user per week among Microsoft 365 Copilot users.

These figures are closer to the realistic impact on a complete working day: often 10%–20%, not three times the entire job.

Therefore:

3x is a workflow-level multiplier. Saving 40–60 minutes per day is closer to a normal whole-job gain.

---

4. Define productivity correctly

A misleading calculation is:

```text

Manual report: 2 hours

AI generation: 2 minutes

Productivity gain: 60x

```

This excludes:

- Preparing the source material;

- Writing instructions;

- Waiting;

- Verifying facts;

- Repairing formatting;

- Regeneration;

- Approval;

- Future rework caused by mistakes.

A more reliable model is:

```text

Effective productivity

= usable output that passes acceptance

÷ (human setup + AI interaction + verification + rework)

```

The multiplier is:

```text

Productivity multiplier

= original human time

÷ AI-workflow human time

```

Example:

- Original workflow: 180 minutes;

- AI generation: 10 minutes;

- Input preparation: 15 minutes;

- Verification: 25 minutes;

- Revision: 10 minutes;

- Total AI workflow: 60 minutes.

Therefore:

```text

180 ÷ 60 = 3x

```

Not 18x.

Quality must remain constant

A valid multiplier requires:

- No increase in factual error;

- The final output meeting the original standard;

- No hidden transfer of work to colleagues;

- No new maintenance burden;

- No privacy or compliance violation.

---

5. Start with a one-week time audit

Do not begin by buying five AI subscriptions.

Record one week of work:

Prioritize tasks that are:

Frequent

They recur daily or weekly.

Rule-based

Inputs, outputs, and acceptance criteria can be explained.

Text or structured-data heavy

AI can read and produce the core material.

Detectable and recoverable

Errors can be found before delivery.

Do not start with the hardest work

A poor first AI experiment is:

- Final contract review;

- Company strategy;

- An enterprise forecast;

- A complex financial model.

A better sequence is:

1. Summarization;

2. Rewriting;

3. Extraction;

4. Classification;

5. Template-based drafting;

6. Exploratory data analysis;

7. Multi-step automation;

8. Support for high-stakes decisions.

---

6. The four-layer productivity system

Reaching 3x generally requires more than one chatbot.

Layer 1: capture

Goal: prevent information from disappearing in meetings, email, and chat.

Tool categories:

- AI meeting notes;

- Email summaries;

- Speech-to-text;

- Web clipping;

- Unified inboxes;

- Knowledge bases.

Useful extracted fields:

- Decisions;

- Action items;

- Owners;

- Deadlines;

- Risks;

- Open questions.

If people still replay recordings, copy notes, and rewrite summaries manually, the later AI stages cannot form an end-to-end system.

Layer 2: understand

Goal: turn raw information into verifiable context.

Tool categories:

- ChatGPT Deep Research;

- NotebookLM;

- Enterprise search;

- File Q&A;

- Spreadsheet analysis.

Good uses:

- Multi-source research;

- Source-constrained questions;

- Contract comparison;

- Historical project synthesis;

- Spreadsheet exploration;

- Contradiction detection.

ChatGPT Deep Research can use the public web, selected sites, uploaded files, and enabled applications to produce a cited report.

NotebookLM is useful when the answer should remain grounded in user-supplied sources.

Layer 3: produce

Goal: create an 80%-complete first version.

Tool categories:

- General AI assistants;

- Document AI;

- Presentation generation;

- Spreadsheet assistants;

- Coding assistants;

- Image and video tools.

AI handles:

- Structure;

- Drafts;

- Format conversion;

- Multiple versions;

- Tone adaptation;

- Preliminary charts;

- Boilerplate.

Humans handle:

- Facts;

- Business logic;

- Sensitive wording;

- Tradeoffs;

- Final approval.

Layer 4: automate

Goal: move information automatically from one step to the next.

Tool categories:

- Zapier;

- Make;

- n8n;

- Microsoft Power Automate;

- Internal workflow systems;

- AI agents.

Example:

```text

Meeting ends

→ transcript retrieved

→ action items extracted

→ project tasks created

→ follow-up email drafted

→ CRM updated

→ notes stored in the knowledge base

→ owners asked to confirm

```

Without automation, every AI use begins with copying, pasting, and re-explaining context.

With automation, AI becomes part of the operating system.

---

7. A complete real-user-style workflow

The following workflow is reconstructed from the STADLER method and current tool capabilities for an ordinary knowledge worker.

It is not a minute-by-minute diary released by Julia Stadler. Undisclosed details are not presented as her personal data.

---

Workflow 1: morning information triage

Before AI

1. Open email;

2. Read Slack or Teams;

3. Review the calendar;

4. Recall incomplete work;

5. Build a priority list manually.

This can take 30–45 minutes and often allows the first urgent message to control the day.

AI workflow

With authorized access to:

- Today’s meetings;

- High-priority unread messages;

- Project tasks;

- Yesterday’s action items;

- Weekly objectives;

the assistant produces:

```text

Three most important outcomes today:

1.

2.

3.

Messages requiring a response:

- person

- reason

- suggested response

Risks:

- potential delay

- dependency

- missing information

Suggested time blocks:

- deep work

- communication

- administration

```

Where time is saved

The assistant reduces searching and summarization.

It should not decide priorities solely from the most urgent email. Final prioritization must reflect goals and accountability.

---

Workflow 2: turn meetings into decisions

Before AI

- Find context before the meeting;

- Take notes while speaking;

- Spend 20–30 minutes cleaning them up;

- Create tasks separately;

- Draft follow-up email;

- Lose the decision history later.

AI workflow

Before

Generate a one-page brief from:

- Previous notes;

- Project status;

- Unfinished actions;

- Relevant email.

During

Use an approved transcription assistant so participants can focus on discussion.

After

Extract:

- Decisions;

- Actions;

- Owners;

- Dates;

- Open issues;

- Risks;

- People to notify.

Then route confirmed tasks into project software.

Human confirmation

Check:

- Who actually committed;

- Whether dates are real;

- Whether an opinion was incorrectly recorded as a decision;

- Whether recording was permitted.

---

Workflow 3: reduce two hours of search to a focused research session

Before AI

- Try many keywords;

- Open many pages;

- Copy notes manually;

- Lose track of sources;

- Begin writing late.

AI workflow

1. Define the decision question;

2. Limit geography, dates, and source types;

3. Use Deep Research to gather cited evidence;

4. Place key original documents in NotebookLM;

5. Ask for contradictions and unknowns;

6. Open the most important original sources;

7. Generate the brief only after verification.

Prompt

```text

Research question:

[specific question]

Decision supported:

[why this matters]

Scope:

- geography:

- dates:

- exclusions:

First produce a research plan.

Prioritize primary sources, official documents, and independent research.

Cite every important claim.

Separate facts, inferences, and recommendations.

Use “cannot verify” when necessary.

```

Why human review remains

AI can:

- Cite stale pages;

- Mix incompatible definitions;

- Miss paywalled evidence;

- Present correlation as causation;

- Select evidence that supports the initial thesis.

The gain comes from reducing search cost, not eliminating verification.

---

Workflow 4: turn a half-day draft into 20–60 minutes of structured work

This is the central STADLER use case.

Before AI

1. Collect information;

2. Invent a structure;

3. Start the first paragraph;

4. Rework headings;

5. Add examples;

6. Normalize tone;

7. Fix formatting.

The time sink is often not expertise. It is translating existing knowledge into readable material.

AI workflow

Establish persistent context

In ChatGPT Projects or another managed workspace, save:

- Objective;

- Audience;

- Reference files;

- Brand voice;

- Prohibited claims;

- Examples;

- Acceptance criteria.

Request the outline first

```text

Create a three-level outline from the supplied sources.

For every section include:

- core claim

- supporting evidence

- missing information

- likely objection

Do not draft the full text yet.

```

Draft section by section

Require:

- No invented numbers;

- Explicit use of the sources;

- Unknowns to be marked;

- A defined voice;

- A defined length.

Run an adversarial review

```text

Act as the most skeptical reviewer.

Identify:

1. unsupported facts

2. reasoning gaps

3. repetition

4. overpromising

5. unanswered questions

List issues only. Do not rewrite.

```

Finalize manually

Humans own:

- The central thesis;

- Number verification;

- Professional judgment;

- Sensitive wording;

- Final approval.

Why this can reach 2.5x–6x

AI removes:

- Blank-page time;

- Initial structure;

- Routine wording;

- Format conversion;

- Repeated versions.

It does not remove:

- Expertise;

- Decision-making;

- Review;

- Accountability.

---

Workflow 5: ask questions of the spreadsheet before writing formulas

Before AI

- Clean columns;

- Search for formulas;

- Build pivots;

- Find anomalies;

- Copy charts;

- Write conclusions.

AI workflow

Upload well-structured data and instruct:

```text

Inspect data quality before analysis.

Report:

1. missing values

2. duplicates

3. outliers

4. definition conflicts

5. fields requiring confirmation

After confirmation:

- summarize by month and region

- calculate year-over-year and period-over-period change

- identify the five largest changes

- create tables and charts

- separate measured facts from possible explanations

```

ChatGPT data analysis can analyze uploaded files and create calculations, tables, and charts.

Human checks

Verify:

- Denominators;

- Time windows;

- Currency;

- Gross versus net;

- Duplicate records;

- Missing values;

- Business definitions;

- Misleading visualization.

---

Workflow 6: repurpose one source into six formats

STADLER’s up-to-6x social-content result reflects knowledge reuse, not only faster typing.

One product update can become:

1. Customer email;

2. LinkedIn post;

3. Internal announcement;

4. Sales talking points;

5. FAQ;

6. Presentation outline.

Prompt

```text

The following content is the only factual source:

[source]

Create:

1. a 150-word customer email

2. a short social post

3. an internal announcement

4. a spoken sales explanation

5. five FAQs

6. a six-slide outline

Rules:

- do not add numbers absent from the source

- adapt tone to each audience

- preserve factual consistency

- mark commitments requiring approval

```

Risk

If the source contains an error, automation distributes the error to every channel.

Confirm the source of truth first.

---

8. A reproducible 3x time model

The following standardized week demonstrates how 3x can emerge.

It is not a personal time diary published by Julia Stadler or STADLER. It is a reproducible measurement template informed by the real case and common knowledge-work patterns.

Results:

The lesson:

Occasional chatbot use often produces 1.5x–2.3x improvement. Approaching 3x usually requires persistent context, templates, and automated handoffs.

Why this will not replicate everywhere

The model assumes:

- Repeated tasks;

- Available source material;

- Clear acceptance standards;

- Context access;

- Human review;

- Stable automation;

- Knowledge-intensive work.

If the bottleneck is:

- Customer waiting;

- Approval;

- Conflict;

- Field execution;

- Budget;

- Regulation;

- Supply chain;

AI only improves part of the system.

---

9. A minimum viable tool stack

You do not need ten subscriptions.

Individual knowledge worker

- General assistant: ChatGPT, Claude, or Gemini;

- Source-grounded research: NotebookLM;

- Meetings: an organization-approved meeting assistant;

- Automation: Zapier, Make, or n8n;

- Tasks and knowledge: Notion, Feishu, Google Workspace, or Microsoft 365.

Microsoft organization

- Microsoft 365 Copilot;

- Teams transcription;

- SharePoint;

- Power Automate;

- Excel;

- Enterprise permissions.

Google Workspace team

- Gemini for Workspace;

- NotebookLM;

- Google Meet;

- Docs, Sheets, and Drive;

- Apps Script or third-party automation.

Higher-privacy organization

- Enterprise AI workspace;

- Approved connectors;

- Private knowledge system;

- Permissions and audit logs;

- Human approval;

- Self-hosted automation where necessary.

Selection principle

Choose tools that eliminate copying and pasting, not merely the tool that produces the most impressive isolated answer.

An AI with governed access to:

- Files;

- Meetings;

- Email;

- Projects;

- Tasks;

will usually create more durable productivity than a disconnected chatbot.

---

10. Seven reusable prompts

Task decomposition

```text

I need to complete:

[task]

Acceptance criteria:

[criteria]

Available sources:

[sources]

First:

1. identify missing information

2. break the task into steps

3. mark AI-owned and human-owned work

4. identify risks and review gates

Do not produce the final output yet.

```

Meeting brief

```text

Create a one-page pre-meeting brief from the history:

- prior decisions

- open actions

- key data

- questions that must be resolved

- likely conflicts

- recommended questions

Do not present discussion points as confirmed facts.

```

Source-constrained drafting

```text

The supplied files are the only factual source.

Create:

[output]

Rules:

- do not add external numbers

- cite important claims

- mark uncertainty

- separate facts, inferences, and recommendations

```

Data analysis

```text

Inspect data quality before analysis.

Check:

- missing values

- duplicates

- outliers

- units

- dates

- category definitions

Ask for clarification before calculating.

```

Adversarial review

```text

You are responsible for rejecting this work.

Find:

- unsupported claims

- reasoning gaps

- ignored alternatives

- overpromising

- inconsistent numbers

- errors that could cause loss

Return risks and questions only.

```

Repurposing

```text

Transform the same factual source into:

- executive summary

- customer email

- internal announcement

- social post

- FAQ

- presentation outline

Keep facts consistent. Do not invent information for impact.

```

Retrospective

```text

Compare the manual and AI workflows.

Report:

- time saved

- additional review time

- causes of rework

- reusable templates

- steps that can be automated next

- steps that should remain human

```

---

11. A 30-day implementation plan

Week 1: measure only

Goals:

- Track time;

- Find five repeated tasks;

- Select two low-risk workflows;

- Establish a quality baseline.

Do not redesign everything at once.

Deliverables:

- Time log;

- Task inventory;

- AI suitability score;

- Manual examples.

Week 2: create templates

For two workflows, define:

- Input format;

- Prompt;

- Reference example;

- Acceptance checklist;

- Adversarial review prompt.

Goal:

Move from repeated explanation to a fill-in workflow.

Week 3: connect context

Place relevant information in:

- Project workspaces;

- NotebookLM;

- Enterprise search;

- Shared folders;

- Searchable databases.

Goal:

Reduce repeated uploading and background explanation.

Week 4: automate handoffs

Only automate a stable workflow.

Example:

```text

Form submitted

→ AI classifies

→ summary generated

→ task created

→ response drafted

→ human approval

→ send

```

Final comparison:

- Total time;

- Output;

- Rework;

- Error;

- Delay;

- Mental fatigue;

- Time for high-value work.

---

12. The productivity dashboard

Track:

A valid 3x threshold

Do not claim 3x unless:

- Measurement lasts at least four weeks;

- There are at least ten task samples;

- Final quality is not lower than baseline;

- Review and rework are included;

- Work was not shifted to someone else;

- The result does not depend on one unusually good generation.

---

13. Seven common failure modes

Automating too early

An unstable workflow becomes an automated unstable workflow.

Long prompt, unclear acceptance

“Professional” is less useful than:

- 800 words;

- CFO audience;

- preserve four figures;

- no external facts;

- end with three risks.

Starting every conversation from zero

Without project context, references, and examples, the model must guess repeatedly.

Too many tools

Copying the same material across five AI products can be slower than doing the task manually.

Ignoring review time

Thirty seconds of generation plus 40 minutes of checking is not a 30-second task.

Confusing fluent language with truth

The most dangerous errors are often expressed clearly.

Filling the saved time with low-value output

The objective should not be:

Produce more documents that nobody reads.

Use saved time for:

- Customers;

- Judgment;

- Design;

- Learning;

- Rest;

- The real constraint.

---

14. Data that should not be sent to unapproved AI

Unless the organization has approved the system, avoid uploading:

- Customer personal information;

- Employee health and HR data;

- Non-public financial information;

- Full contracts;

- Passwords and keys;

- Production databases;

- Confidential code;

- Regulated government data.

Safer practices:

- Use enterprise or education plans;

- Confirm training policy;

- Review subprocessors;

- Apply least privilege;

- Remove identifiers;

- Add human approval;

- Keep logs;

- Restrict automatic sending and deletion.

Productivity is not worth losing control of data.

---

15. Who is most likely to gain 3x?

Strong candidates

- Content and marketing;

- Consultants;

- Researchers;

- Product managers;

- Sales;

- Administration and operations;

- Teachers;

- Independent founders;

- Report-heavy roles;

- People with repetitive communication.

Harder environments

- Primarily physical work;

- Non-digitized inputs;

- Every task is unique;

- No process standards;

- Long multi-party coordination;

- External AI is prohibited;

- Errors are hard to detect.

The expert risk

Research frequently shows larger gains for novices.

Experts can be harmed when fluent, average recommendations reduce their vigilance or originality.

Experts should delegate:

- Search;

- Formatting;

- First drafts;

- Repetition;

- Adversarial review.

They should retain:

- Core models;

- Professional conclusions;

- Critical tradeoffs;

- Final accountability.

---

16. A realistic, non-hyped result

The evidence supports three levels of impact.

Whole working day

Mature enterprise users often report:

- 40–60 minutes saved per day;

- Three to five hours saved per week;

- Roughly 10%–25% overall productivity improvement.

Suitable individual tasks

Controlled studies and workflow tests show:

- 40% less time for professional writing;

- 55.8% faster completion of a specific coding task;

- Enterprise knowledge queries reduced from 7.5 minutes to about one minute;

- 2.5x faster first drafts;

- Up to 6x for high-volume content.

Standardized end-to-end workflows

When high-frequency work is:

- Templated;

- Connected to knowledge;

- Automatically routed;

- Protected by human quality gates;

a selected work package can reasonably target 3x.

That does not mean organizational output immediately triples.

The new bottleneck may become:

- Approval;

- Customer demand;

- Sales cycles;

- Decision speed;

- Execution resources;

- Market size.

---

17. Final assessment

The lesson from Julia Stadler and the STADLER team is not simply “tell employees to use ChatGPT.”

They did more:

- Defined appropriate uses;

- Provided official access;

- Created training and guardrails;

- Turned good workflows into custom assistants;

- Expanded across functions;

- Watched actual usage;

- Shifted people from blank-page creation to review and judgment.

Their published outcomes include:

- 30%–40% savings on common knowledge tasks;

- 2.5x faster first drafts;

- Up to 6x acceleration in high-volume content;

- More than 85% daily active use.

The realistic meaning of “triple productivity” is not:

AI does my entire job.

It is:

I redesign repeated, rule-based, reviewable work so AI handles search, organization, drafting, and routing, while I retain objectives, facts, judgment, and accountability.

The sequence worth copying is:

1. Measure;

2. Select;

3. Template;

4. Connect context;

5. Automate handoffs;

6. Add quality gates;

7. Review continuously.

At that point, 3x stops being a marketing claim. It becomes a workflow result that can be verified through time and quality metrics.

Information was updated on June 26, 2026. The real-world case figures come primarily from vendor customer stories and academic studies; customer stories may be subject to selection bias. Investment, medical, legal, financial, and HR decisions require qualified professional review.

---

Sources

1. [STADLER reshapes knowledge work at a 230-year-old company](https://openai.com/index/stadler/)

2. [The state of enterprise AI 2025](https://openai.com/business/guides-and-resources/the-state-of-enterprise-ai-2025-report/)

3. [Navigating the Jagged Technological Frontier](https://www.hbs.edu/faculty/Pages/item.aspx?num=64700)

4. [Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence](https://www.science.org/doi/10.1126/science.adh2586)

5. [Generative AI at Work, The Quarterly Journal of Economics](https://academic.oup.com/qje/article/140/2/889/7990658)

6. [The Impact of AI on Developer Productivity: Evidence from GitHub Copilot](https://www.microsoft.com/en-us/research/publication/the-impact-of-ai-on-developer-productivity-evidence-from-github-copilot/)

7. [Vodafone Microsoft 365 Copilot customer story](https://www.microsoft.com/en/customers/story/19346-vodafone-microsoft-365-copilot)

8. [BBVA scales AI across the organization](https://openai.com/index/bbva-2025/)

9. [Deep research in ChatGPT](https://help.openai.com/en/articles/10500283-deep-research-in-chatgpt)

10. [Projects in ChatGPT](https://help.openai.com/en/articles/10169521-projects-in-chatgpt)

11. [Data analysis with ChatGPT](https://help.openai.com/en/articles/8437071-data-analysis-with-chatgpt)

12. [NotebookLM Help](https://support.google.com/notebooklm/)

13. [Notion AI Meeting Notes](https://www.notion.com/help/ai-meeting-notes)

14. [Notion Enterprise Search](https://www.notion.com/help/enterprise-search)

15. [Zapier AI workflows and agents](https://zapier.com/)