Can a ChatGPT Dot help with your job search? Job lists, tailored résumés, trackers and 9 real cases

The short version: where a Dot helps your job search, how to set it up, where it gets stuck

Is a Dot worth using for a job search?

Yes, for the repeated legwork around a search, but not as a replacement for your own judgment or your final click. In the cases below it is good at four things. First, turning your résumé into a short list of real postings, linked directly rather than through search pages, and keeping that list in a tracker: 22 matching jobs with salary ranges in a spreadsheet, and direct links to roles such as a software editor job at CNET (cases 2, 3). Second, checking listings before you see them: JobGuard failed three of five, including the top match, and Axton Liu's Dot caught a posting whose dates contradicted each other (cases 5, 4). Third, tailoring a résumé and cover letter for one role and leaving them ready for you to review: three reviewed applications led to one interview (case 1). Fourth, watching for new openings or freelance leads on a schedule while your own computer is off, like an hourly lead check that pushes promising matches to a phone (case 7). On the hiring side it organises applicants and interview logistics and leaves the choices to you (cases 8, 9). It tends to get stuck on an out-of-date résumé, a short list that misses much of the market, listings that have expired or don't add up, contact emails it can't verify, sign-ins, layout, and the application questions only you can answer. It suits people who already know which roles they want and will review every application before it goes out.

Setup steps

1. Put one current résumé where it can find it, and name the file. Ruben Circelli's Dot used an older copy from Gmail while his current one sat in Google Drive (case 3). OpenAI's guide to tracking applications suggests gathering your current résumé, portfolio, role descriptions, deadlines and notes about experience you can verify, and leaving out personal identifiers the drafts don't need (Track job applications).
2. Describe the target and the limits. Give the roles, location or remote, contract type and salary floor, then say what the Dot must not do. Berzaf wrote the list of things his Dot cannot do before the list of things it can (case 5).
3. Ask for a tracker, not a list in chat. OpenAI's guide suggests one row per role with the organisation, deadline and source link, required qualifications, verified evidence of fit, gaps, status and next action, plus one folder per role so drafts and submitted versions don't get mixed up (Track job applications). @alexusadays's tracker had room for submissions, follow-ups and interview dates; Mike Holp's jobs stayed as links in the conversation (cases 2, 6).
4. Make checking part of the job. Ask it to open each posting, confirm it is still open and the apply link works, and keep failures in a separate list with the reason (case 5). Axton Liu still spot-checked the table against the companies' own sites (case 4).
5. Schedule the search with four details: what to check, when to run (with time zone and end date), where results go, and what deserves a message (tasks and memory). Berzaf runs his every weekday morning and Rohan's runs every hour (cases 5, 7).
6. Keep sending for yourself. In Custom rules, set applications and messages to "Ask before taking action" or "Hand off to you" (Getting started with your dot). Approving one message does not give your Dot ongoing permission to contact people for you (Dots privacy, security and safety FAQs). Dr. Kirubel reviewed each of his three applications before it was sent (case 1).
7. Tailor only from what is true. Ask it to separate a posting's requirements from its preferences, mark anything you don't have, and leave the application questions for you; OpenAI's guide says to remove unsupported claims and keep gaps visible (Track job applications). Dr. Kirubel still fixed his CV layout and wrote his own answers (case 1).

Where it gets stuck, and what worked

1. It uses the wrong résumé. Circelli's Dot picked up an old copy from his inbox (case 3). What worked: name the current file and where it lives, and ask the Dot to say which file it used.
2. The list misses most of the market. Circelli notes that it doesn't try every keyword combination or scroll through every page of results, and Mike Holp had three jobs after about 45 minutes (cases 3, 6). What worked: treat it as a daily top-up to your own searching, and give it the keyword combinations and sites you would use yourself.
3. Listings are expired, unverifiable or inconsistent. Three of JobGuard's first five failed its check, and Axton Liu's Dot found contradictory dates (cases 5, 4). What worked: a check before anything reaches the shortlist, a separate list of failures with reasons, and a spot-check of your own on the company's site.
4. There is nobody to write to. Mike Holp's run couldn't verify a contact email for either role, so nothing could be sent (case 6). What worked: ask for a verified contact or none at all, have it draft the message, and send it yourself.
5. Sign-ins, layout and questions need you. Axton Liu took over for a sign-in (case 4), @alexusadays's Dot signed in to LinkedIn on its cloud computer (case 2), and Dr. Kirubel fixed the CV layout and answered the application questions himself (case 1). What worked: expect to step in for sign-ins, open the finished file before it goes anywhere, and keep the questions about you for yourself.
6. It feels slow. Axton Liu found the speed frustrating, and Mike Holp didn't think this job was worth the price of his plan (cases 4, 6). What worked: set it up once as a scheduled job and let it run in the background, then judge it by what the scheduled runs bring back rather than by watching it work (cases 5, 7).
7. Hiring records don't add up. Satya Mallick's test included a skill claimed without evidence and a duplicate application (case 8), and Ted's recruitment sheet had what looked like the same candidate on two different dates (case 9). What worked: ask it to keep claims and evidence apart, mark missing evidence as unknown, merge duplicates and say which, and ask you before anything in a record changes. Who to interview or hire stays with you.

What it can do now, and what it can't yet

What it can do now:

• Compare your résumé with your public LinkedIn profile and make an updated file.
• Find matching postings and link straight to them, with salary ranges where they are listed.
• Check that a listing is still open and point out details that contradict each other.
• Keep a tracker for submissions, follow-ups and interviews.
• Tailor a résumé and cover letter for one role, ready for you to review.
• Run a search or lead check on a schedule and message you only when something new matches.
• Organise applicants, merge duplicates, spot clashing interview records and draft interview questions on the hiring side.

What it can't do yet:

• Cover the whole job market or every keyword combination.
• Reliably find and verify a recruiter's contact email.
• Answer application questions in your words, or guarantee the final layout of your CV.
• Decide which job to take or which applicant to hire.

One message to start with

I'm looking for [role] jobs in [location or remote], [contract type], paying at least [salary]. My current résumé is [file name] in [location]; use only that file, and tell me if you find another version. Build a tracker in [sheet] with one row per job: company, role, link, salary range, deadline, requirements, what in my résumé fits, gaps, status and next action. Before a job goes in, open the posting and check that it is still open and the apply link works; put jobs that fail in a separate tab with the reason. Every weekday at 8:00 [time zone] until [end date], add new matches and message me only if there is something new. Don't apply, message anyone or share my résumé without asking me first.

Quick answers

Can a ChatGPT Dot find jobs for me?

Yes, and the better runs came back with real postings rather than search pages. Ruben Circelli's Dot read his résumé from Gmail and returned direct links to roles such as a software editor job at CNET (case 3). @alexusadays got 22 matching jobs with salary ranges in a tracker spreadsheet (case 2).

Will it apply for me?

Only if you set it up that way, and the people who got results kept the last step themselves. Dr. Kirubel's Dot submitted three tailored applications, each after he had reviewed it, and one led to an interview (case 1). Custom rules let you choose "Take action without asking", "Take action if pre-approved", "Ask before taking action" or "Hand off to you" for each kind of action (Getting started with your dot).

Can I trust the listings it finds?

Ask it to check each one. Berzaf's Dot verified every listing before the shortlist, and three of the first five failed, including the best match (case 5). Axton Liu's Dot pointed out a posting whose dates contradicted each other (case 4).

Which résumé will it use?

Whichever one it finds. Circelli's Dot used an older copy sitting in his inbox, not the current one he keeps in Google Drive (case 3). Tell it which file is current.

Can it help on the hiring side too?

Yes, as an organiser. Satya Mallick's Dot sorted fictional intern applications, kept missing evidence visible, wrote interview questions and left the choice to him (case 8).

Applications sent, and one interview

1. Three tailored applications after review, and one interview (Dr. Kirubel / @AIbalanced, 3 Oct)

In a YouTube video posted on 3 October 2026, Dr. Kirubel (channel @AIbalanced) describes his first 24 hours with a Dot, spent on real research, documents and job applications. The Dot helped him submit three tailored applications, each one after he had reviewed and approved it, and one had already led to an interview. He still had to fix the layout of his CV and supply his own answers to the application questions. He also points out that some workflow screens in the video are reconstructions rather than recordings, and he shows two real submission receipts.

Finding and tracking openings

2. Résumé checked against LinkedIn, 22 matching jobs and a tracker (@alexusadays)

@alexusadays asked his Dot for help with a job search. It compared his résumé with his public LinkedIn profile, added his current position and generated a new file, adding a missing logo when he asked. It then signed in to LinkedIn on its cloud computer and compiled 22 matching jobs with salary ranges, including remote US roles listed at $180–200k and $155–206k. Everything went into a job-tracker spreadsheet with room for submissions, follow-ups and interview dates. Along the way it also reminded him that an electricity bill was coming due.

3. Direct links to real postings from a résumé in Gmail (Ruben Circelli / PCMag, 13 Oct)

Ruben Circelli (PCMag, 13 Oct 2026) connected Gmail, told his Dot it could find his résumé there, and asked for good job opportunities based on his experience. The Dot found relevant roles, such as a software editor position at CNET, and linked straight to each posting. The two other agents in his test linked to lists on job aggregator sites or suggested positions that had already expired. He found two limits. The Dot worked from an older version of his résumé in his inbox, while the current one sits in Google Drive. And it can't replace combing through listings yourself, because it doesn't try every keyword combination or scroll through every page of results.

4. An internship hunt for a test student profile (Axton Liu, 2 Oct)

In a Chinese-language video (2 Oct 2026), Axton Liu gave his Dot a made-up college student's profile and the job of finding an internship. Working in its cloud browser, with him taking over for a sign-in, it built a table of researched openings, rewrote the résumé and cover letter for the role he picked, and pointed out a posting whose dates contradicted each other. He still spot-checked the table against the companies' own sites, and found the speed frustrating. A daily check for new openings was scheduled but had not run yet when he recorded.

Checking that a listing is real

5. JobGuard: every listing verified before it reaches the shortlist (Berzaf)

Berzaf built a Dot called JobGuard to hunt remote contract work, with one rule: every listing is verified before it goes on the shortlist. The first run returned five roles. Two passed verification and three failed, including the one with the highest match score. He wrote the list of things the Dot cannot do before the list of things it can, then scheduled the search for every weekday morning. The demo used a fictional résumé, and nothing was submitted.

Where it fell short

6. 45 minutes, three jobs and no contact emails (Mike Holp, 1 Oct)

In a video published on 1 October 2026, Mike Holp tried to build a job agent in Dots: find remote AI automation jobs, tailor his résumé to each one, find the contact email and draft the outreach. After about 45 minutes he had a basic version. It found three jobs, wrote a cover letter and appeared to tailor the résumé. It couldn't verify a contact email for either role, so there was nothing to send, and the jobs stayed as links in the chat instead of going into a tracker. He found it slow, and for this job he didn't think it was worth the price of the plan.

Freelancers: leads instead of listings

7. Hourly freelance leads sent to his phone (Rohan / @Rohan2Katara)

Rohan turned phone notifications into a freelance lead inbox. Every hour his Dot checks relevant subreddits and searches set keyword combinations on X and LinkedIn for people looking to hire. Because the Dot runs on its own cloud computer, the search keeps going while his PC is off or he is asleep. Through a custom plugin built in Codex, the Dot files each lead into the X, Reddit or LinkedIn section of Rohan's own productivity app, and the app sends a phone notification when a match looks promising. Each lead says why it fits his work and links to the original post; one was someone wanting a website on a $500 budget. Deciding whether to take it and writing to the client stay with Rohan.

On the hiring side

8. Applicants sorted, and the choice left to him (Satya Mallick, 30 Sep)

Satya Mallick (LinkedIn, 30 Sep 2026) gave a Dot four fictional intern applications and a role brief, with clear limits: organise the evidence, keep missing information visible and leave decisions to him. Applicant two listed Python, but the project they described was a browser interface with no Python sample. The Dot kept both facts side by side and marked the evidence as unknown instead of treating the claim as proof. When he changed the priority to project work first, it reordered the columns without changing anyone's eligibility. When he added two more applications plus a duplicate, it returned six unique applicants, not seven, and noticed that one summary mentioned three unanswered questions without saying what they were. Asked for an interview draft for applicant one, it wrote five project-specific questions and left the invitation as a draft. He picked the applicant himself. He is clear that this was one staged session with fictional records and no comparison against a regular chat.

9. A duplicate candidate spotted the day before an interview (Ted / @tedzted, 30 Sep)

Ted (Threads, in Traditional Chinese, 30 September 2026) had connected Asana, Google Drive and Gmail to his Dot. While they were talking about something else, it noticed an interview in Asana for the next day, went through his Google files and found two rows in the recruitment spreadsheet that looked like the same candidate: the same surname, Lin, at 3 PM, but on different dates. It asked him to confirm whether something was wrong. HR confirmed it was the same person and one of the forms had not been updated.

More asks you can copy

Daily shortlist (from cases 3 and 5):

Every weekday at 8:00 [time zone] until [end date], look for [role] jobs in [location or remote], [contract type], paying at least [salary]. Use my résumé in [file location], which is the current version. Before a job goes on the shortlist, open the posting and check that it is still open and the application link works. Put jobs that fail the check in a separate list with the reason. Add new matches to [sheet] with the company, title, salary range, link and date found. Don't apply, message anyone or share my résumé. Message me only if there is something new.

Résumé check before a search (from case 2):

Compare my résumé in [file] with my public LinkedIn profile at [URL]. List anything that is missing or out of date, such as my current role. Make a new version of the file with those fixes and show me the changes before I use it.

One tailored application, ready for me to send (from case 1):

For the job at [link], rewrite my résumé and a cover letter for this role using only what is true in my current résumé. Mark anything the posting asks for that I don't have. Leave every application question I need to answer myself empty and list them. Don't submit anything; show me the finished files first.

Freelance leads (from case 7):

Every hour, check [subreddits] and search [keywords] on [sites] for people looking to hire for [kind of work]. For each match, give me one line on why it fits my work, the budget if stated, and a link to the original post. Send only matches that look promising to [where I'll see them]. Don't reply to anyone.

Applicant tracker for a small hire (from case 8):

Here are [number] applications and the role brief. Build a tracker with one row per applicant and columns for [skills], project evidence and open questions. Keep a claimed skill separate from evidence that shows it, and mark missing evidence as unknown. Merge duplicates and tell me which ones you merged. Don't rank or reject anyone; I decide.

Interview-day check for a hiring manager (from case 9):

Every weekday at 7:30 [time zone], look at tomorrow's interviews in [calendar or task app] and compare them with [recruitment sheet]. If a name, date or time doesn't match, or two rows look like the same person, tell me which rows and why. Don't change the sheet or contact anyone; I'll check with HR.

Related

• Use cases hub · Research watch · Teachers & students · Small business & clients · Email triage
• Dots + Gmail · Cloud computer · Getting started · Features · Pricing