ChatGPT and Claude can help you apply to jobs, but in 2026 neither one can run your applications for you the way an auto-apply service does. Both write good answers to screening questions and solid cover letter drafts, and their browser agents (ChatGPT agent mode and Claude in Chrome) can fill in a simple application while you watch, but they stall on logins, captchas, Workday account creation and long multi-page forms. They also do not watch company career pages for new postings, so even when they finish an application, they cannot get you in early.

If you want something that actually applies for you, you have two real options: build your own agent from a GitHub project and keep it running, or pay for a hosted service such as Scoutify's Sniper, which applies on the employer's own career site minutes after a matching job posts. I built Scoutify, so weigh my recommendation with that in mind. This guide explains where each route works, including the cases where the free routes are the better choice.

The short answer

ChatGPT can draft answers, tailor resume bullets and write cover letters, and its agent mode can fill in a simple form if you sign in for it and approve the submit.
Claude can do the same writing work, Claude in Chrome can fill in forms in your own browser, and Claude Code can drive a browser through Playwright if you are comfortable writing and fixing code.
Neither assistant watches career pages, so neither one applies to a job the morning it posts unless you find the posting yourself.
Open-source GitHub projects can submit applications, but the best-known one dropped LinkedIn after LinkedIn objected, and all of them need regular repairs.
A hosted agent such as Sniper costs $25 a week for up to 500 applications after 25 free ones, and it is worth paying for when you want applications sent early without supervising each one.

Disclosure: I, Jules Lemée, founded Scoutify and wrote this guide. I checked the current state of ChatGPT agent mode, Claude in Chrome, Claude Code and the GitHub projects below on September 30, 2026. These products change every few months, so check each vendor's own help pages before you rely on a detail.

What ChatGPT and Claude can do for a job application today

It helps to split a job application into four tasks: finding a posting that fits you, writing the material, filling in and submitting the employer's form, and confirming that the application arrived. General assistants are very good at the second task, and they have recently become partly useful at the third. They do almost nothing for the first and the fourth, and I think those two are the ones that decide whether a recruiter ever reads your application.

ChatGPT agent mode

ChatGPT's agent mode runs a browser on a virtual computer that OpenAI hosts, and you can watch it click through pages in the chat window. It can search a job board, read a listing, open the application and type your details into simple fields. When a site needs you to sign in, it hands the browser back to you, and it asks for your permission before actions with real consequences, which includes pressing submit on an application. Agent mode is part of the paid ChatGPT plans rather than the free one, and OpenAI limits how many agent tasks each plan can run; on the $20 Plus plan, that limit has been reported at about 40 tasks a month, which a serious search can use up in a week.

Posts went viral in 2026 claiming that agent mode applied to hundreds of jobs overnight, but I have not seen one that showed the employers' confirmation emails. Even LoopCV, which sells its own auto-apply product and therefore has every reason to be generous about tools it can sit next to, wrote in July 2026 that agent mode cannot reliably handle multi-step applicant tracking flows, captchas or logins with two-factor authentication.

Claude in Chrome and computer use

Claude in Chrome is Anthropic's browser extension, and it became generally available on every paid Claude plan (Pro, Max, Team and Enterprise) in late August 2026. Unlike ChatGPT's agent, it works inside your own Chrome with your own logins, which removes some of the sign-in friction. It can read the page, fill in fields and click through steps, and by default it pauses to ask you before actions that need your approval. Anthropic's safety guidance says it is prohibited from bypassing captchas, so any application with a captcha stops until you solve it yourself.

Anthropic's computer use goes one step further, because it lets Claude operate a whole desktop through screenshots, clicks and typing. Underneath, it works the same way as the extension: it handles one task at a time, you watch it, and it is usually slower than you would be on a form you have already filled in a dozen times.

Claude Code with Playwright

This is the route that most searches for "auto apply to jobs using Claude" and "AI job search Claude Code" are really about. Claude Code is Anthropic's coding agent, it is included in the Pro plan, and it can control a real browser through Playwright, either by writing a Playwright script or by driving the browser directly through the Playwright MCP server. A developer can get it to fill in a Greenhouse form in an afternoon.

The hard part starts after that afternoon. Each applicant tracking system lays out its forms differently, every employer adds its own screening questions, and when one of those pages changes, a script that worked last month can fail quietly. You end up maintaining software in the middle of a job search, which is a strange way to spend the weeks when you most need to be preparing for interviews.

"Apply with Claude" connectors

You will also find pages that offer to let you "apply with Claude" or "apply with ChatGPT". LoopCV, for example, publishes a Claude connector: you add its server to Claude, describe the jobs you want in plain English, and LoopCV's paid plans submit the applications. That can be a reasonable product, but it is worth being clear about who does the work. Claude is the chat window, and LoopCV's own service does the searching and the applying, so you are really judging LoopCV rather than Claude. Our LoopCV review covers how that service applies and what it costs.

Where general agents get stuck

The failure points are predictable once you have filled in a few hundred applications yourself, and they are mostly the same for ChatGPT and Claude.

Logins and Workday accounts

Most Workday career sites ask you to create an account with that one employer before you can apply, then verify your email address, and then sign in again. Each company on Workday is a separate account, so applying to 30 Workday companies can mean 30 accounts and 30 passwords. A general agent can type into the sign-up form, but it cannot open your inbox to click the verification link unless you give it access, and I would not hand an experimental agent both my email and my password manager just to save a few minutes.

Captchas and automated-traffic checks

Many career sites put a captcha, or a quieter check for automated traffic, in front of the submit button. Claude in Chrome will not bypass one, ChatGPT's agent sometimes clears a simple checkbox and sometimes does not, and a script you wrote yourself has to deal with it on its own. Applicant tracking systems have also started screening the applications themselves. In February 2026, Greenhouse launched Real Talent, which flags applications with fraud signals such as a location that does not match the IP address, disposable or internet phone numbers, and patterns repeated across many applicants. Greenhouse says no application is rejected automatically, but a flagged application reaches the recruiter with a warning attached, and that is not where you want yours to land.

The captcha is only the visible part. Detection systems try to cover every little step of the human fingerprint, including the size of the browser, whether the computer has a GPU, the time zone and whether the traffic comes through a proxy, and any competent anti-bot solution checks hundreds of signals like these. A person filling in the form on their own laptop matches those checks because it is their real browser on their real machine, but a home-made agent running in a cloud browser or behind a shared proxy can look wrong on several of them at once. We see this with Sniper too: Ashby flags shared IP addresses, so we gate Sniper's Ashby applications instead of pushing them through.

Multi-page forms and screening questions

A Workday application usually runs across five or six steps: your information, your experience, the employer's application questions, voluntary disclosures, self-identification and a review page. Each step has dropdowns that only accept the employer's own wording, date pickers that behave differently from site to site, and questions such as "Are you legally authorized to work in the United States?" that must be answered exactly right. A general agent can do one page well and then lose track on the next one, and when it picks the wrong option in a dropdown, nothing on the page tells you until a recruiter reads it.

Nothing watches the career page

This is the limit that matters most, and it is the one people notice last. ChatGPT and Claude act when you ask them to. Neither one watches thousands of career pages and wakes up when a matching job appears, so the job you hand them is a job you already found, usually on a board where hundreds of other people found it too. Recruiters often review applications in the order they arrive and stop once they have enough people to interview, so an application that is well written but sent two days late is competing for whatever attention is left. Our auto-apply speed report looks at how soon after a posting different tools actually apply.

No proof that it arrived

When an agent says it submitted an application, the only evidence is the agent's own message. The employer's confirmation email goes to your inbox, mixed in with everything else, and if the submit failed silently, there is nothing to find. Our guide to the job application confirmation email explains what a real confirmation looks like and what it means when one never arrives.

Job application bots on GitHub: the do-it-yourself route

"Auto apply jobs GitHub" and "job application bot GitHub" are popular searches, and I understand the appeal: the code is free, you control it, and you can point it at whichever model you like. The state of these projects in 2026 is worth knowing before you spend a weekend on one.

AIHawk (Jobs_Applier_AI_Agent)

AIHawk was the best-known open-source applier. It went viral in 2024 for filling in LinkedIn Easy Apply forms with answers written by a language model, and it collected close to 30,000 stars on GitHub. LinkedIn objected, saying that it does not permit third-party software that automates activity on its site, and in early 2025 the project removed its LinkedIn features. When I checked on September 30, 2026, the original repository address redirected to a different project from the same author (a stealth Playwright browser tool for AI agents), and the forks that still carry the job applier code had mostly not been updated since 2025.

career-ops

career-ops is the most popular Claude Code project for job searching, with more than 73,000 stars on September 30, 2026. It is a well-built tool, and it is honest about what it is. It scans job portals including Ashby, Greenhouse and Lever, scores each listing against your profile, tailors a PDF resume, drafts answers and tracks your pipeline, and its documentation says plainly that it drafts and fills but never submits: you review the form and press submit yourself. Its author has written that he built it during his own search and evaluated more than 740 listings with it before landing a role.

If you want a careful search where you read every application before it goes, career-ops is a good fit, and it pairs well with an alert or auto-apply tool that handles timing. It is not an auto-apply tool, though, so people who find it while searching for one usually discover that they still have to submit every form by hand.

Other scripts

Searching GitHub turns up many smaller projects with names like "ai-job-agent" and "linkedin-auto-apply". Most of them target LinkedIn Easy Apply, because it is one form layout, and most have a handful of stars and no commits in the past year. I would read the date of the last commit before I read the README.

The maintenance burden

Running your own applier is a small software project. You need a machine that stays on (or a server), an API key for a model, a profile file with your answers, and the patience to fix the script when a site renames a field. By my rough estimate, the model usage costs a few cents to a few tens of cents per application, depending on the model and the number of questions, but your time costs much more than that. If the script breaks on a Tuesday night, it stays broken until you notice, and a script that fails quietly can send half-filled applications under your name.

LinkedIn's terms and your account

Most GitHub appliers target LinkedIn, and LinkedIn's User Agreement prohibits using "bots or other unauthorized automated methods to access the Services", as well as software, scripts and browser extensions that scrape it. LinkedIn enforces those terms by restricting accounts, and your LinkedIn profile is worth more to your career than any batch of Easy Apply submissions. Our guide to LinkedIn auto apply covers the risks and the safer alternatives in detail.

Time and cost compared

The table below compares four ways to get applications out. The time for applying by hand comes from my own experience of about 20 minutes per application on Greenhouse and Workday, the prices are each vendor's as of September 2026, and the rest are my estimates, so treat the table as a starting point rather than a measurement.

Route	What you pay	Your time	How early it applies	Proof it arrived
Applying by hand	You pay nothing beyond your time.	Each application takes about 20 minutes, and a new Workday account adds more.	You apply when you find the job, which is usually after it reaches a job board.	The confirmation email arrives in your own inbox.
ChatGPT agent mode or Claude in Chrome	A paid plan starts at about $20 a month, with limits on agent use.	You still sign in, solve captchas, check answers and approve each submit, so the saving on long forms is small.	It applies only to jobs you have already found.	The agent reports what it did, and the employer's email goes to your inbox.
A GitHub project or a Claude Code script	The code is free, and model usage costs a few cents or more per application.	Setup takes a day or more, and repairs continue for as long as you use it.	It applies as early as your own job feed allows, and most projects read job boards.	You get whatever logging you build, plus the employer's email in your inbox.
Scoutify Sniper	Sniper costs $25 a week for up to 500 applications after 25 free ones, and Auto Apply costs $15 a week for up to 200.	You set your filters, upload resumes and complete your apply profile once, and after that you review results.	It applies minutes after a matching job posts on a supported career site.	The employer's confirmation email lands in Scoutify Mail next to the application.

One more number helps. If Sniper sends the full 500 applications in a week, $25 works out to 5 cents per application. A narrow search sends far fewer, so your real cost per application will be higher, but it still costs less than the 20 minutes each one would take you by hand. There is also a fifth route, which is paying a person to apply for you, and our guide to paying someone to apply to jobs compares those services with software.

When a hosted agent is worth paying for

I built Sniper because of my second internship cycle. I already had alerts telling me about new jobs within minutes, but I was filling in the same fields on Greenhouse and Workday over and over, at about 20 minutes per application, when I should have been preparing for interviews. A general assistant would not have fixed that for me, because the slow part was never the writing. The slow part was the forms, plus the need to find each job early enough for the forms to matter.

We built Sniper's agent ourselves, starting in February 2026, and the career sites showed us quickly why a home-made agent is hard to keep working. In March, our Workday path reported applications as submitted when Workday had actually returned an error inside a response that looked successful, so our system said "submitted" for applications that were never sent. Lever's captcha wall beat us outright, so we switched Lever off in April. Problems like these are what a job seeker running their own script has to find and fix alone, in the middle of a search.

Sniper does the parts that general agents skip. Scoutify watches 50,000+ company career pages, and when a new posting passes your filters (role, years of experience, location, job type and visa sponsorship), Sniper opens the employer's own application on Workday, Greenhouse or SmartRecruiters, fills it in with your resume and profile, and submits it, often minutes after the job went up. Ashby and Lever are rolling out now. Sniper submits around the clock, so it keeps applying while you are at work, in class or asleep, which a supervised agent cannot do, because a supervised agent needs you sitting in front of it.

It also gives you proof. Most employers send a confirmation email when their system receives an application, and for most Sniper applications that email lands in Scoutify Mail, usually within two minutes, next to the application it belongs to. You can open any application in the web app and see the resume and answers that were sent. Sniper never uses your LinkedIn account and never applies on linkedin.com, so your profile is never part of the risk.

Sniper is the better choice in these situations:

You want to apply to new jobs at companies that hire through their own career sites, and you want to be among the first applicants.
You would rather spend your evenings preparing for interviews than supervising an agent or fixing a script.
You want to check the employer's own confirmation for each application instead of trusting a dashboard.

A general assistant or a do-it-yourself setup is the better choice in these situations:

You are applying to a small number of roles and want to write every answer yourself, in which case ChatGPT or Claude is a good writing partner and career-ops is a good research tool.
Your target jobs exist only on LinkedIn or Indeed, because Sniper does not apply there.
You enjoy building tools, and you accept that you will maintain one during your search.

Sniper has limits of its own. It does not write cover letters or rewrite your resume; it sends your own resume, choosing the best fit for each job from up to 100 that you upload. It also cannot make an employer reply. If you are unsure whether auto-applying helps at all, our guide to whether auto-apply works covers what recruiters see, and how many jobs to apply to per day covers how much volume makes sense.

If you have tried ChatGPT or Claude and found yourself babysitting forms, let Sniper apply to new jobs as soon as they post. Your first 25 applications are free with no card, and you can compare the employers' confirmations with what your agent used to report.

How to use ChatGPT and Claude alongside an auto-apply tool

The best setup I know uses each tool for what it is good at. Use ChatGPT or Claude to sharpen your resume, prepare answers to the questions you will see again and again, and write cover letters for the handful of roles you care about most. Our ChatGPT prompts for job search and the ChatGPT cover letter guide cover the writing side, and AI resume builders vs ChatGPT explains when a dedicated resume builder helps. Then let a tool that watches career pages handle timing and submission, and spend the time you get back on interviews.

If you would rather keep the submit button for yourself, an autofill extension is the middle ground. Our comparison of autofill and auto-apply explains the difference, and Scoutify vs Simplify compares the most popular autofill extension with Sniper.

Frequently asked questions

Can ChatGPT apply to jobs for me?

ChatGPT can help you apply, but it cannot run your applications on its own. Its agent mode, which comes with the paid plans, can fill in a simple form while you watch, but it hands the browser back to you for logins, asks your permission before submitting, and often stalls on multi-page forms such as Workday's. It also does not watch career pages, so it only applies to jobs you have already found.

Can Claude apply to jobs for me?

Claude can fill in some applications under your supervision, but it is not an auto-apply service. Claude in Chrome, which is available on paid Claude plans, fills in forms in your own browser and stops at captchas, and Claude Code can drive a browser through Playwright if you write and maintain the code. Neither one watches career pages for new postings, so neither one applies early on its own.

Is there an AI that applies to jobs for you?

Yes, hosted auto-apply services apply for you without supervision. Scoutify's Sniper watches 50,000+ company career pages and applies on the employer's own Workday, Greenhouse or SmartRecruiters form minutes after a matching job posts, and it costs $25 a week for up to 500 applications after 25 free ones. Our guide to the best auto-apply tools compares it with AIApply, JobCopilot, Jobright and others.

Are GitHub auto-apply bots safe?

GitHub auto-apply bots carry two main risks. Most of them automate LinkedIn, whose User Agreement prohibits bots and other unauthorized automation and which restricts accounts that break that rule, and a script that breaks quietly can send incomplete applications under your name. If you use one, read the code, keep it away from your LinkedIn account and check what it sends.

Is using AI to apply for jobs cheating?

Using AI to apply for jobs is not cheating, as long as everything in the application is true. Employers judge the resume and answers you submit, and a tool that sends your real resume with accurate answers is doing the same paperwork you would do by hand. The line is misrepresentation, which means invented experience, false answers to screening questions, or generated material that you have not read.

Can ChatGPT or Claude Apply to Jobs for You? What Works in 2026