# ChatGPT for policy analysts on Mac in 2026: briefs, research and spoken help with Lazur

> Set up ChatGPT for policy analysts on Mac with Option+Space, projects, file limits and claim checks. Add Lazur for spoken help while you work.

URL: https://www.lazur.app/for/policy-analysts/chatgpt
Updated: 2026-11-09

ChatGPT for policy analysts on Mac is most useful for the work around a policy: summaries, tables, plain-English drafts and rewrites. Use it to speed up those jobs, and keep the judgement, the options and the sign-off with you. Lazur adds voice: hold a key, ask about what is on screen, and hear the answer while you keep reading.

## Quick answer

- ChatGPT on Mac suits summaries, tables, plain-English drafts and rewrites. Keep the policy choices and the sign-off with you.
- Put each brief in its own project. Per-project file caps reported for Free are 5 files, Plus and Go 25, and Pro 40.
- Option+Space opens Quick Chat from another app. The ChatGPT Mac app changed in July 2026, so check which version you run.
- Hold ⌥⌘ to ask Lazur about anything on screen. Hold fn to dictate text at the cursor in any app.

Most policy analysts use ChatGPT for the work around the policy, not the policy itself. That means reading long consultation responses, turning messy notes into a summary, reshaping a table for a briefing, and rewriting a paragraph in plain English. Those tasks are where the tool earns its place. Choosing the options is still your job.

![A policy analyst at a Mac laptop reviews a brief, one everyday setting for ChatGPT for policy analysts on Mac.](https://images.lazur.app/content-images/role-app/policy-analysts-chatgpt/chatgpt-for-policy-analysts-on-mac-1-e8b16f.jpg)

## Where ChatGPT fits in a policy analyst's week

A UK study of think tanks shows the pattern clearly. The researchers invited 554 policy analysts and policy researchers from 94 think tanks to an online survey. They then interviewed eight people in depth. The abstract does not say how many replied, so read the findings as a signal, not a measure.

The authors concluded that generative AI plays a rather insignificant role in policy analysis. It was used only in limited ways, mostly for menial tasks and for dissemination and publication. Ideation and policy formulation stayed human-led, with some AI support.

Analysts also worried about the lack of concrete policies governing AI use. They worried that inappropriate use could lower the quality of their work. That worry is a good starting point for your own team rules.

Those findings match the day-to-day work many analysts describe. These are the jobs worth trying first:

- Summarise a long consultation response to a fixed word limit, then check each quoted claim against the original.
- Turn meeting notes into a list of decisions, owners and open questions.
- Rewrite a paragraph for a minister's office in plainer English, with every figure left unchanged.
- Reshape a messy table into clean columns that a briefing deck can use.
- Draft a stakeholder email that asks for comments by a date you choose.
- List the assumptions in a model description, and mark the ones you have not tested.

## Set up ChatGPT on your Mac before the first brief

Start with the shortcut. Press Option+Space to open Quick Chat from any app. OpenAI's release notes say this shortcut shows the floating controls and focuses Quick Chat. Use it to check a definition or a date while a PDF stays open.

Quick Chat is faster than switching windows, but it is still ChatGPT. The same data rules apply to anything you type into it. Keep confidential extracts out of quick questions unless your team allows them.

The app itself changed this year. OpenAI merged the Codex app into the ChatGPT desktop app for macOS on July 9, 2026. The updated app is available on every plan, including Free. The previous app was renamed ChatGPT Classic. A banner in the app says Classic retires on January 30, 2027.

> **Test before a deadline:** 9to5Mac reports that OpenAI will test a combined Chat and Work view with a small group of users. The article calls the new interface extremely confusing and says the native macOS app was replaced by an Electron-based version. Run your usual brief through it on a practice document first.

Work and Codex modes share plug-ins, according to 9to5Mac. Codex mode shows more technical detail that Work hides. Most policy tasks sit comfortably in Work mode. Check the mode before you start, so you know which view you are reading.

## Set up one project per brief, not one long chat

A project keeps the files and conversations for one piece of work together. Put one brief, one consultation or one review in each project. A long chat that mixes several briefs makes answers drift. It also makes the latest version of a table hard to find.

| Plan | Files per project | Practical meaning for a brief |
| --- | --- | --- |
| Free | 5 | A short note with a few sources |
| Plus and Go | 25 | A consultation response set of moderate size |
| Pro, Business, Enterprise and Edu | 40 | A full evidence base in one place |

_Per-project file caps, as listed by a third-party guide. Check your plan page before you plan around them._

Those counts come from fast.io, a third-party guide. It attributes the per-project caps to each plan, but it does not cite OpenAI for those counts. Treat the numbers as a planning guide, and confirm them in your own account before a large upload.

Set up each project the same way every time. This routine works for most analysts:

1. Name the project with the brief title and the month it was last changed.
2. Upload only the final version of each source, and remove superseded drafts.
3. Write your house rules in the first message: word limits, tone and which sources are allowed.
4. Keep a running list of open questions inside the project, so the next session starts from it.
5. Copy the final text into your document, then check every figure against the original.

## Use local projects for folders on your Mac

Local projects can now include multiple related folders. That helps if your evidence sits in one folder per region or per year. OpenAI's release notes from late July 2026 describe this change. Local Work conversations stay on the computer, which matters for sensitive files.

Before you point ChatGPT at a folder, check what is in it. A folder of consultation responses may hold names, addresses and contact details. Move anything personal out first, or use a copy with the personal fields removed.

- **554** Policy analysts and researchers invited to a UK think-tank survey, drawn from 94 think tanks ([source](https://research-portal.uws.ac.uk/en/publications/generative-artificial-intelligence-in-policy-analysis-a-study-of-/))
- **31%** Workers who used AI in the last week and said it saved them one to two hours ([source](https://www.census.gov/library/stories/2026/08/ai-use-at-work.html))
- **10%** Workers who used AI in the last week and said it saved them no time ([source](https://www.census.gov/library/stories/2026/08/ai-use-at-work.html))
- **40** Files per ChatGPT project on Pro, Business, Enterprise and Edu plans, per a third-party count ([source](https://fast.io/resources/chatgpt-projects-file-limit/))

## Check every claim before it reaches a minister

The biggest mistake is trusting a clean summary. A model can rephrase a finding and drop the caveat that made it true. It can also move a figure from one paragraph to another and change what it describes. Read the output against the source, not against your memory of it.

Ask for a claim list, not a summary. Request one line per claim, with the section or page where it appears. Then compare each line with the source yourself. A line with no reference is a line you cannot use yet.

These mistakes show up most often in analyst drafts:

- Pasting a restricted document without checking your organisation's rules first.
- Accepting a figure that was rounded or moved between paragraphs.
- Letting the draft choose the policy options. Options are your call.
- Leaving old drafts in the project, so the model summarises a version nobody approved.
- Dictating a note with names and addresses straight into a shared document.

## Write prompts that hold up under review

A good prompt names the audience, the length, the source and what you do not want. Compare two requests. The first asks for a summary. The second asks for a summary of the consultation response for a senior official, in plain English, with each claim tied to a section and no recommendations.

The second prompt gives the model fewer ways to go wrong. It also gives you a checklist to review the output against. Save the prompt in the project, so you can reuse it next time without retyping it.

## Draft faster without losing your voice

Use ChatGPT for the first pass, then rewrite the opening and the recommendation yourself. Those two parts carry the judgement, and readers notice when they sound generic. The method notes and tables are often where the saved time sits.

The time effect is real for many workers, but it is uneven. In a U.S. Census Bureau survey from March 2026, 31% of workers who used AI in the last week said it saved them one to two hours. Another 10% said it saved no time, and 4% said it took more time.

Measure your own result. Keep a plain log for a few weeks, with the task, the minutes with ChatGPT and the minutes without. Your own numbers will tell you more than a national survey does.

![Hands typing on a laptop keyboard while drafting an office document with AI help.](https://images.lazur.app/content-images/role-app/policy-analysts-chatgpt/chatgpt-for-policy-analysts-on-mac-3-5655ce.jpg)

## Worked example: summarising a consultation response

Say you have a long consultation response from a housing charity, and the deadline is Friday. Start a project, upload the response, and ask for three things. Ask for the main claims, the evidence offered for each claim, and the parts the response does not address.

Read the output against the response. Check that each claim appears in the text and that the evidence matches it. Then ask ChatGPT to draft a one-page summary in plain English, using only the claims you approved. Edit the result in your own words before it goes in the brief.

Watch for these edge cases, because they break the first draft:

- Scanned PDFs with poor text recognition, where a number can be read wrongly.
- Tables with merged cells, which can lose their row labels when pasted.
- Responses in another language, where a legal term may be translated loosely.
- Quotes that are cut short, so the meaning changes without anyone noticing.

## When ChatGPT gets it wrong

Do not argue with the model until it agrees. Point to the line in the source that contradicts the output, and ask it to fix that one claim. If the same error returns, rewrite the claim yourself. Note the error in the project, so the next draft avoids it.

## Use Lazur next to ChatGPT, not instead of it

Lazur is a voice assistant for the Mac that lives in the notch. Hold ⌥⌘ to ask about anything on screen, and Lazur answers out loud. Ask where something is, and Lazur draws a circle and a label around it. It does not click buttons, and it does not operate ChatGPT for you.

Ask how to do something, and Lazur guides you one click at a time, waiting for each click. That helps with settings you use once a quarter. It also helps when you cannot remember which menu holds the setting you need.

Hold fn to dictate. Clean, formatted text lands at the cursor in any app. Lazur also keeps snippets, a personal dictionary and clipboard slots. Add your department's acronyms to the personal dictionary, so dictation spells them right.

![A researcher reads printed documents and notes before writing a policy summary.](https://images.lazur.app/content-images/role-app/policy-analysts-chatgpt/chatgpt-for-policy-analysts-on-mac-2-7b3e10.jpg)

## What to say to Lazur while you work in ChatGPT

The tasks below are things analysts say in the middle of a job. Keep the ChatGPT window open, and speak the question. Lazur answers from what is on screen, so the question works best when the answer is visible.

Ask one thing at a time. Lazur answers more usefully when the question is narrow, such as which claim has no source. A broad question, such as whether the brief is good, gets a broad answer. Use your own judgement for that.

## Privacy, confidential material and the limits

Treat every upload as a decision. A consultation response may carry personal data. A draft brief may be confidential long before it is published. Check your team's rules before you upload anything. The UWS study found analysts flagged the lack of concrete policies on generative AI at think tanks. Your team may have the same gap.

Lazur listens and looks at the screen only while you hold the shortcut. Audio and the screenshot go to AI providers to answer, and they are not stored. Lazur content is never used to train AI models. Read the privacy terms yourself, rather than relying on a summary.

Check the Computer History setting before you turn it on. OpenAI says it is available in the macOS desktop app for Pro, Business and Enterprise customers. It is off by default, and it expanded to the EEA, Switzerland and the UK in August 2026. Keep it off for confidential briefs.

## Next steps this week

1. Open ChatGPT, press Option+Space, and confirm that Quick Chat appears.
2. Create one project for your current brief, and upload only the final sources.
3. Run one summary task on a document you have already reviewed, then compare the output.
4. Write your house rules in one short note, and keep it in the project.
5. Try Lazur for a week, and note which questions it answers well.

Start small. Pick one task that repeats each week. That shows you where ChatGPT saves time and where it adds checking. For the wider comparison, read [ChatGPT on Mac vs Lazur](https://www.lazur.app/apps/chatgpt) and [Lazur vs ChatGPT](https://www.lazur.app/compare/lazur-vs-chatgpt). If the voice side fits your work, get Lazur from the [download page](https://www.lazur.app/download).

## With Lazur

Lazur is a voice assistant for the Mac that lives in the notch. Hold ⌥ ⌘ and ask; it sees your screen, answers out loud, points at buttons and walks you through apps. Hold fn to dictate. Free for 3 days, then $18/month: https://www.lazur.app/download

- “Lazur, which figure in this draft has no source?”: Lazur circles the unsourced figure on screen and says out loud what is missing. (Points at it)
- “Lazur, where is the Project button in this window?”: Lazur circles the control, labels it, and says what it does.
- “Lazur, walk me through making a new project.”: Lazur gives one step at a time and waits for each click before it moves on.
- “Lazur, read the last paragraph of this brief out loud.”: Lazur reads the visible text aloud, so you can check the tone while your eyes rest.
- “Lazur, turn these notes into a plain-English paragraph I can paste.”: Lazur drafts the paragraph as text you can review, edit and paste yourself.
- “Dictate the note on the consultation reply.”: With fn held, your words land at the cursor in any app, cleaned up.
- “Lazur, find the latest published figures on AI use at work.”: Lazur runs the search in its own Chrome tabs and returns a summary with a report.
- “Lazur, set a timer for the stakeholder read-through.”: Lazur starts the timer and says so when it ends.
- “Lazur, which claim on this page has no source?”: Lazur circles the unsourced claim and says what is missing.

## FAQ

### Does ChatGPT work for policy analysis on a Mac?

Yes, for the work around policy. Use it to summarise, reshape tables, draft plain-English text and check wording. A UK study found think tanks used generative AI in limited ways, mostly for menial tasks and publication. Keep the judgement with you.

### Which ChatGPT app should I use on Mac in 2026?

OpenAI merged the Codex app into the ChatGPT desktop app on July 9, 2026. The old app is ChatGPT Classic, which a banner says retires on January 30, 2027. Check which version is open before you start a deadline task.

### How many files can I put in one ChatGPT project?

A third-party guide lists 5 files on Free, 25 on Plus and Go, and 40 on Pro, Business, Enterprise and Edu. Those counts are not cited to OpenAI in that guide. Check your plan page before you plan a large evidence base.

### What does Option+Space do in ChatGPT on Mac?

It shows the floating controls and focuses Quick Chat, so you can type a request from another app. It suits a quick definition or a date check while you read a PDF. Do not paste confidential text into it without checking your team's rules.

### Can Lazur see my ChatGPT window?

Lazur sees your screen only while you hold ⌥⌘. It answers out loud and points at things on screen. It does not click buttons or operate ChatGPT for you, and it is not a plugin or add-in for any app.

### Is it safe to use ChatGPT for confidential policy work?

Only if your organisation allows it. The UWS study found analysts flagged a lack of concrete rules on generative AI. Read your team's guidance, remove personal data, and keep Computer History off for confidential briefs.

### Does Lazur work on Windows?

No. Lazur runs on macOS, on Apple silicon and Intel Macs. Windows is planned, but it is not available yet.

## Sources

1. [Generative artificial intelligence in policy analysis: a study of UK think tanks](https://research-portal.uws.ac.uk/en/publications/generative-artificial-intelligence-in-policy-analysis-a-study-of-/), UWS Academic Portal, checked 2026-11-09
2. [ChatGPT What's New (release notes)](https://learn.chatgpt.com/docs/whats-new), OpenAI, checked 2026-11-09
3. [OpenAI unveils ChatGPT Work agent, GPT-5.6 models now available](https://9to5mac.com/2026/07/09/openai-announcing-the-next-chapter-for-chatgpt-today-watch-here/), 9to5Mac, checked 2026-11-09
4. [OpenAI is retiring ChatGPT Classic app, will soon combine Chat and Work tabs in new app](https://9to5mac.com/2026/11/02/openai-is-retiring-chatgpt-classic-app-will-soon-combine-chat-and-work-tabs-in-new-app/), 9to5Mac, checked 2026-11-09
5. [About a Third of Workers Who Used AI in the Last Week Said They Completed Tasks One to Two Hours Faster](https://www.census.gov/library/stories/2026/08/ai-use-at-work.html), U.S. Census Bureau, checked 2026-11-09
6. [ChatGPT Projects File Limit: Plan Caps and Search Workarounds](https://fast.io/resources/chatgpt-projects-file-limit/), fast.io (third party), checked 2026-11-09