ChatGPT Work tab merged into chat? What it means for your CV

What changed when OpenAI merged ChatGPT's Work tab into chat?

On 2 November 2026, OpenAI told subscribers it's merging the Chat and Work tabs in ChatGPT, and that ChatGPT and Codex will draw on the same usage allowance. Until now the desktop app had two sections, ChatGPT and Codex, and the ChatGPT section was split again into Chat and Work. A small group of testers gets the combined version in the coming weeks, and a wider rollout follows. Codex stays a separate experience inside the desktop app. If you're not in the test group, you'll keep the two-tab layout for a while. Two things haven't been announced yet: how big the shared pool will be, and whether the web and mobile apps get the same merge.

Work launched on 9 July 2026. It was the mode for when you wanted a finished file rather than a conversation: a brief, a deck, an analysis, a document you could review and use. In the desktop app it can open and edit files on your computer and run programs. On the web it runs in the cloud. Each Work session gets a scratch folder that persists across chats. That's why so many job seekers started dropping a CV, a job ad and a cover-letter draft into Work and asking for a polished Word file back. Work has been a paid feature on plans from $20 a month upward. Free users and the cheaper Go tier don't get it.

So why should a job seeker care about a tab layout? Partly because it's a trend: Anthropic folded Claude's chat and Cowork into a single screen in September. But mostly because the merge changes two things you rely on when you tailor applications. First, the line between talking a problem through and producing the file gets blurrier. You'll pick a mode less deliberately, and you may spend heavier compute on tasks that didn't need it. Second, from December your CV rewrites share a budget with anything you run in Codex, like the portfolio site you're coding to back up a developer application. Neither change is a disaster. Both reward a tighter workflow, and the rest of this piece gives you one.

Where do your CV projects and files live after the merge?

Your work isn't going anywhere. The merge changes how you get into ChatGPT, not where things are stored. Nothing in the announcement suggests chats, projects or uploaded files will be removed. Work-style tasks will simply start from the same screen as ordinary chat. What OpenAI hasn't explained publicly is how existing Work sessions and their scratch folders will show up in the combined view. So don't assume a three-week-old Work thread will be easy to find once your layout changes. If a tailored CV matters, it shouldn't live only in a chat sidebar. Download it, name it after the employer and the role, and keep it somewhere you control the version history.

Set up one project for your whole job search, not one per application. Put your master CV in the project files. That's the long version with every role, metric and certification. Add a short note on your target titles, your minimum salary and the locations or remote setups you'll accept. Each application then becomes one chat inside that project: paste the job ad, ask for the tailored version, export it. This setup survives interface changes because the context lives in the project, not in a particular tab. Paste in current salary ranges and market details yourself, because some of ChatGPT's knowledge can go stale without fresh data from you. You also avoid the classic mess of five slightly different CVs scattered across chats.

If you use the desktop app, remember that Work can read and write files directly on your computer. That's handy when you want ChatGPT to open your Word CV, edit it and save a copy next to the original. It's also a reason to be careful. Point it at a dedicated job-search folder, not a documents folder full of payslips, reference letters and passport scans. Treat every file it writes as a draft. Open it yourself, check the formatting survived, and confirm that dates, employer names and job titles still match reality. A file can look perfect in Word and still parse badly in an applicant tracking system, and the chat won't warn you.

How does the shared Codex usage pool cut into CV rewriting?

From December, Plus, Pro and Business subscribers get one token-based allowance shared across ChatGPT and Codex. It replaces the separate counters each part of the app had before. OpenAI says most people won't notice. It also says that, depending on how much you use and which model you pick, some people will wait for resets more often. Heavy users running long, reasoning-heavy sessions are the ones most likely to notice. That describes a job seeker on a Sunday afternoon pasting in eight job ads and asking for eight rewrites, eight cover letters and a round of revisions on each. Tokens count what you send as well as what you get back, so re-sending a two-page CV and a long job ad on every turn adds up fast.

Plus subscribers have felt this squeeze before. On 25 August OpenAI brought back a five-hour cap on Work and Codex for Plus, while Pro kept unlimited access. At the end of October it doubled Pro limits and dropped a planned cut to the $200 tier. The pattern is clear: the $20 plan gets a real but limited budget, and now that budget is shared. If you also use Codex, say to build a portfolio project for an engineering role or to script a job-application tracker, those sessions come out of the same pot as your CV edits. It's also not confirmed whether light chat on the fast default model will stay outside the meter once the pools combine.

Here's my unpopular view: CV work rarely needs the heaviest model or the highest reasoning setting. Rewriting bullets to match a job ad is a writing task, not a research problem. A mid-tier model on a light or medium reasoning level handles it well. Save the expensive settings for the one or two jobs you really care about, where you want the model to weigh trade-offs. Which projects should go? How should you frame a career gap? Does a title change need explaining? Using maximum reasoning for every small tweak is like hiring a senior lawyer to proofread a two-line email. You'll hit the reset wall sooner, and the output will barely differ from what a lighter setting gives you.

What's the leanest ChatGPT CV tailoring workflow now?

The leanest workflow does the tailoring in one structured pass instead of a dozen back-and-forth messages. Most wasted usage comes from feeding things in bit by bit. You paste the CV and ask for feedback, then paste the job ad and ask for changes, then ask again because the first draft drifted off course. Every turn re-sends the whole context. Instead, put everything in one opening message: your master CV (or a pointer to it in the project files), the full job ad pasted as text, three to five constraints, and a clear description of the output you want. A model that has the whole brief from the start writes a better first draft. You'll usually need one or two correction rounds, not eight.

Here's a prompt that works: "Using my master CV in this project, tailor a two-page CV for the job ad below. Keep every employer, title and date exactly as written. Reorder and rewrite bullets so the most relevant achievements come first. Use the ad's wording where it truthfully applies, and don't add any skill I haven't listed. Then list the ad's requirements I don't clearly meet, plus three questions you'd need answered to close those gaps." That last line matters most. Models tend to cover over gaps, and this turns that habit into a list you can act on. You get one focused follow-up instead of an endless loop of "make it better."

Group the rest of your requests. If you're applying for three similar roles, ask for one shared base version plus a short list of changes for each ad, not three full rewrites. Write cover letters only after the CV is final, and do it in the same chat so the model reuses the context instead of rebuilding it. Ask for plain text first and formatting last. Reformatting a document four times is the fastest way to burn tokens for nothing. If you hit a reset partway through your search, that's a good moment to try another assistant. Our comparison of which Claude model works best for resume writing covers when the cheaper model is enough and when the top one is worth it.

Why shouldn't ChatGPT grade the CV it just wrote?

Because a model reviewing its own draft is grading its own work, and it'll nearly always give itself a pass. Ask ChatGPT whether the CV it just wrote is ATS-friendly and you'll usually get a yes plus a few cosmetic tips. It can't see how a real applicant tracking system will read your file. Language models also tend to agree with whatever is already in the conversation. And different models don't even agree with each other: if you've ever run the same CV past two different assistants, you've seen how far their verdicts can split. The chat is a capable writer. As an examiner of its own output, it's lenient and biased.

The checks that matter happen on the file, not the text. A tailored CV can read beautifully in the chat window and still break in a parser. A table can scramble your job history. Dates can be in a format the system can't read. A header can swallow your contact details, and a two-column template exported from a design tool can come out jumbled. The model never sees any of this, because it worked on words while the ATS reads the final PDF or Word file. Polished bullets don't help if the parser files your latest job under Education. So once you've downloaded your tailored version, test that exact file, the one you'll actually upload, not the draft in your chat history.

This is where an independent score is worth having. Upload the final file to HRLens's free CV analysis and you'll get an overall score out of 100, five category scores including ATS compatibility and clarity and structure, a list of your strengths and a visual layout analysis. Paste the job description too. Paste it as text, because LinkedIn and Indeed block link fetching. The analysis then lists the skills the ad asks for that your CV doesn't show. If you want to check your CV against a specific job ad before every application, it's the same quick flow. Ten minutes of checking is cheap insurance after an hour of tailoring, and it doesn't touch your ChatGPT allowance.

Frequently asked questions