How to set up ChatGPT automations for your job search

What changed in ChatGPT on 29 September 2026, and why does it matter for your job search?

At its Dev Day on 29 September 2026, OpenAI gave ChatGPT plugins a dedicated home in the sidebar and moved a step closer to automations that fire on events rather than on your typing. Developers can now build sidebar apps that open fullscreen, conversation panels that sit beside a chat, and file viewers that handle their own formats. There's a new Plugin Creator tool, a redesigned submission flow for the directory, and per-permission approval so you can grant a plugin one type of access without handing it everything. For anyone job hunting, the practical result is that searching, tracking and drafting stop living in six separate browser tabs.

The automation half of the announcement is the part to watch, and it's worth being precise about its status. OpenAI announced support for MCP Events, a proposed specification that would let a plugin trigger work when something happens inside a connected app, and said it's making automations easier to add and manage. That event-triggered piece isn't fully shipped yet. Scheduled tasks, though, already landed earlier in 2026: a Scheduled page and time windows arrived in June, and in August tasks opened to free accounts, gained sharing, and picked up event triggers tied to Gmail, Slack and GitHub inside ChatGPT Work. That's enough to build a real job-search system today.

Here's the bridge, and it's the whole point of this article. Automation multiplies whatever you feed it. A scheduled task that surfaces twelve relevant roles every morning is only useful if the document you send in response is strong enough to survive the first screen. Run a weak CV through an automated pipeline and you don't get more interviews — you get more rejections, faster, and with less idea why. The build below assumes you'll spend an hour fixing the base document before you spend ten minutes wiring up the tasks. That order matters more than any prompt in this piece.

Which ChatGPT automations are actually worth building for a job search?

Four automations carry almost all the value: a daily job-ad digest, a keyword extractor that runs on every ad you shortlist, a follow-up reminder timed to each application, and a weekly CV-gap review. Everything else is decoration. The digest replaces the thirty minutes you currently lose scrolling boards; the extractor turns each ad into the five or six phrases your CV needs to contain; the reminder keeps you from ghosting yourself after a promising first conversation; the weekly review tells you which skill keeps appearing in the ads you want and never appears on your CV. Together they take a scattered search and give it a spine.

Notice what's missing from that list: nothing here applies on your behalf. Scheduled tasks run at most once an hour, unattended ones can be paused, and on a free account a task produces a chat message rather than actions inside other apps. It can search, summarise and draft. Sending, submitting and paying stay in your hands, which is exactly where they belong. I'd argue the fully automated apply-bot is a bad idea even when the technology allows it — employers already see applicant volumes that make a generic submission worthless, and the way to stand out is a tailored application, not a faster one.

Think of the four automations as a funnel rather than a checklist. The digest widens the top. The extractor narrows each ad into a concrete tailoring brief. The reminder protects the conversations you've already started, which convert far better than cold applications. The weekly review closes the loop by feeding what you learned back into the document. Skip any one of them and the system leaks: no digest and you apply to whatever LinkedIn happens to show you; no reminders and half your warm leads go cold; no weekly review and you repeat the same gap in your CV for three months without noticing.

How do you set up each automation, step by step?

Open the Scheduled page in the ChatGPT sidebar, or simply describe the recurring job inside any chat, and ChatGPT will create the task and confirm it back to you. Start with the digest: ask for a weekday morning run that searches for roles posted in the last 24 hours matching your title, location and seniority, returns no more than five, and lists company, role, salary if stated, and one line on why it fits. On a free account, ask for a window such as morning rather than an exact time. Then check Settings and turn on notifications, push or email — without them the task runs and you never see the output.

Automation two is the keyword extractor. Build it as a saved prompt you fire on demand rather than a scheduled task: paste an ad and ask for the exact skills, tools and certifications named, ranked by how many times each appears and whether it sits under required or preferred. That list is your tailoring brief. Automation three is the follow-up: after each application, create a one-off task for seven working days later that reminds you of the company, the role and the name of whoever you spoke to, and drafts a three-sentence nudge. Automation four runs Sunday evening and compares the week's saved ads against your current CV.

Plan limits shape how many of these you can keep live at once, so check the table before you start building. If you're on a free account with three slots, run the daily digest, the weekly CV-gap review, and leave the third slot free for rotating follow-up reminders. One more detail that catches people out: if you pin a task to a specific model, confirm that model is still available. Model retirements happen on published dates, and a task tied to a sunsetting version will need a replacement chosen before it stops working. Review your scheduled list monthly and delete anything you've stopped reading.

Why does your base CV decide the quality of every automated application?

Every automated application is a variation on one source document, so a flaw in the base CV repeats itself across every role you apply to. If your dates parse badly, they parse badly forty times. If your achievements are duties rather than outcomes, no amount of keyword swapping fixes that — the rewrite inherits the weakness and dresses it in the employer's vocabulary. This is the step people skip because it's the least fun part of the process, and it's the step that decides whether the other three hours pay off. Score the document before you scale it, not after a month of silence.

A free CV analysis gives you the baseline in a couple of minutes: an overall score out of 100 rated from Weak to Excellent, five category scores covering experience, tech stack, impact and ownership, clarity and structure, and ATS compatibility, plus your strengths and a visual layout analysis that shows how the file is actually laid out. Upload a PDF, a Word file or even a photo of a printed CV — Hebrew OCR is supported, which matters if you're applying in Israel with a bilingual document. The point is to find out whether the thing feeding your automations is solid before it feeds them.

Then make the analysis job-aware. Paste the job description itself rather than the link, since LinkedIn and Indeed block fetching, and the analysis will list the skills you're missing against that specific ad and tailor its suggestions accordingly. Every analysis also produces ATS-friendly rewritten CVs in six templates that mirror the ad's keywords, which is exactly the artefact your digest automation was surfacing roles for. The paid Full Analysis at $3.99 adds the fix list, the reasons you might be rejected, your blind spots, a hiring probability figure and a salary estimate — useful when you want the diagnosis, not just the score.

How do you stop automated applications from reading like a bot wrote them?

Keep the automation on the research side and keep your own voice on the writing side. Let ChatGPT tell you that an ad names Terraform four times and lists SOC 2 under required; don't let it generate the bullet describing what you actually did with Terraform. The tell isn't AI involvement — recruiters assume that now — it's the flatness that comes from a model inventing specifics it doesn't have. Phrases that could describe anyone, achievements with no number attached, and three candidates opening with the same sentence structure are what get a CV skimmed and set aside within seconds.

There are concrete habits that fix this. Write each bullet yourself in rough form first, with the real number in it: how many users, how much money, how many weeks saved, how big the team. Let the model tighten the sentence, not invent the content. Strip the adjectives it adds back in. If you want a second opinion on whether your draft still sounds machine-made, there's a check built for exactly that question — see whether your CV reads as AI-written before you send it anywhere. Ten minutes of that beats a week of polite rejection emails.

If you're building a CV from scratch rather than fixing one, the chat-based CV builder works the same way your automations do — paste an old CV or just describe what you've done, refine it by chat, then export to PDF or Word in any of six templates. Hebrew CVs are written in Hebrew and exported right-to-left, which saves the formatting mess that usually follows an RTL document through a parser. Build the document once, properly. Then let your scheduled tasks do what they're genuinely good at: finding the right roles every morning so you can spend your energy on the fifteen applications that deserve it.

Frequently asked questions