AI Career & Income / Practical guide
Use Claude to verify roles, match genuine evidence, tailor truthful applications and practise interviews. Includes a blank 100-opportunity CSV, fictional examples and human review checkpoints.
A folder containing 100 tailored resumes can look like progress while hiding a weak search. Some roles may be closed, outside your hiring region or asking for experience you do not have. The better unit of work is a decision you can explain: apply with evidence, hold for a missing fact, or skip for a clear reason.
Claude can help keep that decision attached to the job, application and next step. This guide builds a 100-opportunity pipeline around that continuity. It includes a blank CSV, a fictional three-role example and prompts that prepare work for your review. You do not need to connect your inbox or automate submissions to use it.
Our research and testing policy explains the evidence boundary.
Download the blank 100-opportunity job tracker to follow along as you verify roles, tailor applications and record real progress; it has 27 fields and 100 empty opportunity slots, not 100 actual jobs.
Write your target before searching: role family, seniority, permitted hiring locations, work arrangement, relevant experience and non-negotiable constraints. “Remote analyst” is insufficient if the employer hires only in a country where you cannot work. Decide which facts you must verify and which preferences you can discuss rather than silently converting every preference into a requirement.
The workflow is Discover → Verify → Match → Tailor → Human Review → Apply → Track → Interview Practice → Follow Up. Claude can help with extraction, comparison, drafting and reminders. The candidate remains responsible for career facts, work eligibility, employer instructions and every submission.
Review targeting after each ten verified opportunities. Ten is a suggested checkpoint to make reflection manageable, not a statistical threshold or a quota. Ask whether the roles share realistic requirements, whether your evidence is relevant and which unknowns keep blocking a decision. If the first batch is poorly targeted, improve the search before filling the remaining slots.
Anthropic documents web, Windows/macOS desktop and mobile access. Its supported-location list includes Pakistan and the United States. General access does not establish that every beta, connector or payment method is available in every account. Use the official sign-in flow and the features actually shown to you.
Create a job-search project if it is available, or start with a separate chat and local files. Projects can retain instructions and selected knowledge across chats. A minimal packet contains a master factsheet, a role-target note and the current tracker. Keep a local master copy so a generated revision cannot quietly become the only version of your career history.
Give facts stable labels. For the fictional candidate below, F01 is a personal SQL practice project using synthetic order data; F02 is a volunteer task preparing a weekly spreadsheet summary. The candidate has no paid analyst employment. Those distinctions matter more than a polished summary that turns practice into professional experience.
Use small CSV or text files and inspect what Claude can read. The upload guide supports formats including CSV, PDF, DOCX and TXT; spreadsheet handling and file creation depend on the relevant capability and settings. Chat, project and computing contexts have different limits. Non-PDF extraction can lose visual or formatting information, so do not assume an uploaded resume was read exactly as it looks.
Help organize my job search using only my approved factsheet and tracker.
Treat every career claim as requiring an evidence ID.
Do not invent employers, titles, qualifications, dates, metrics, clients or years of experience.
Separate required criteria, preferences and unknowns.
Keep Apply, Hold and Skip decisions distinct from actual submission status.
Do not send messages, submit applications or answer legal/eligibility questions for me.
If information is missing, ask a specific question or mark it unknown.
Return proposed edits and decisions for human review.Do not upload identity documents, bank details, a home address, references’ private contacts or confidential work merely because they appear in a job portal. A minimized factsheet is enough for most drafting. Follow any employer restriction on AI assistance, particularly for assessments and interviews.
Start with employers’ official careers pages and native job alerts. LinkedIn’s alerts offer daily or weekly notifications. Indeed also provides signed-in alerts, with regional limitations and no promise of a daily email when there are no new matches. Treat an alert as a lead to verify, not a current vacancy receipt.
For normal job-board browsing, make your own minimal factual notes and record the official employer link where possible. Do not ask Claude to scrape LinkedIn, copy an entire jobs database, reuse browser cookies, bypass CAPTCHA or run a bulk-apply extension. LinkedIn prohibits third-party scraping and automation. Indeed restricts unapproved automated access and application bots while specifically recognizing permitted AI connectors.
The Claude Marketplace lists an Anthropic-verified connector made by Indeed, requiring sign-in, with search_jobs and get_job_details. Indeed’s linked beta documentation also describes profile-experience and company-data capabilities. The published tool lists differ; inspect the tools and permissions actually offered to your account rather than assuming every listed capability is available.
The general connection path is Customize → Connectors → the official directory → Indeed → review capabilities and permissions → legitimate vendor sign-in. An older interface may use different labels. Connect only if it is available and you want that data exchange; native alerts and manual verification remain a complete alternative. Neither reviewed connector page establishes worldwide or Pakistan-specific connector availability.
Use the connector for supported discovery and details, not as an application bot. The reviewed pages do not document application submission through it. Profile retrieval is optional and can expose personal experience; it is unnecessary for a first search based on your own role criteria.
Find a small batch of possible [role family/seniority] opportunities matching [location/work-mode constraints].
Use current official employer careers pages or the official Indeed connector only where available and authorized.
Do not scrape LinkedIn/Indeed, bypass restrictions or submit anything.
Return employer, exact title, posting ID, source URL, stated location and required criteria.
Mark missing deadline, eligibility or current-status information unknown.
Keep possible matches separate from verified open roles.
If a permitted source is unavailable, report that limit instead of inventing listings.Claude’s web search can provide citations and fetch pages, subject to access and account controls. Read the original posting yourself. An AI citation can point to an outdated vacancy, a staffing intermediary or a page whose country restriction appears outside the extracted paragraph.
Check employer identity, the exact posting and its current application path. Confirm title, location, permitted hiring countries, essential criteria, deadline and timezone when supplied. If the employer does not state a deadline, leave it unknown. Do not ask Claude to infer a closing date from the age of the posting.
Keep the canonical employer URL and posting ID for deduplication. The same vacancy may arrive through several alerts, and a refreshed listing is not necessarily a new opportunity. A URL with a sign-in token or private application session belongs outside a shared tracker. Record the stable public posting instead.
Set a decision before drafting. Apply means you have chosen to proceed, Hold means a missing fact prevents that choice, and Skip means there is a reason to stop. These decisions are different from the process stage. A closed posting can be Skip/Closed without implying that you applied and were rejected.
| Opportunity | Verified or missing fact | Decision / stage | Next useful action |
|---|---|---|---|
| J001 — fictional Cedar Metrics junior analyst | SQL and reporting fit F01/F02; no paid analyst experience | Apply / Drafting | Prepare factual materials and inspect portal requirements |
| J002 — fictional Harbour Systems operations analyst | Remote hiring country, current status and deadline unknown | Hold / Needs verification | Verify the employer source and eligibility before drafting |
| J003 — fictional Pine Ledger reporting assistant | Fictional posting is closed | Skip / Closed | Retain for deduplication; do not count as a rejection |
These are fabricated teaching examples, not real vacancies or observed hiring decisions. Their actual-event dates, URLs and confirmation fields are blank in the sample CSV. You should fill those fields only from events that happen in your own search.
Ask Claude to separate essential requirements from preferences, then map each one to your factsheet. “Comfortable with SQL” can link to F01. “Three years of professional analyst experience” cannot be satisfied by renaming a practice project. Where evidence is missing, retain the gap and decide whether the role still makes sense.
Use this verified posting or my permitted factual notes and approved factsheet.
Return: requirement; required/preferred; source line; genuine evidence ID; gap; question.
Do not invent a match percentage or an ATS score.
Do not convert coursework, volunteering or a personal project into paid employment.
Preserve exact eligibility and experience conditions.
Suggest Apply, Hold or Skip with reasons, but leave the decision to me.
If the source is incomplete or contradictory, identify what must be checked first.Read the posting again after the matrix is prepared. Extraction can miss a requirement embedded in an application question or collapse “preferred” into “required.” Verify any consequential eligibility answer yourself. Claude should help expose the decision, not make a legal or employment-eligibility determination from assumptions.
Tailoring should improve the order and clarity of relevant evidence. It should not produce a new career history. Keep a master factsheet, then save a role-specific resume version under the opportunity ID. Ask Claude to show each proposed edit next to the original fact and its evidence label.
For fictional J001, an honest bullet could read: “Built SQL queries to summarize synthetic order data in a personal practice project; documented checks for missing values.” It identifies the task and scope without claiming production deployment, paid employment or measured business impact. “Improved client reporting efficiency by 40%” would be unsupported and should be removed, even if the sentence sounds stronger.
Use concrete action verbs, relevant tools and your own contribution. Include a metric only when you can substantiate it. Keep standard headings and a readable layout, then follow the employer’s requested file type. Review the exported file as well as the chat text. No layout, keyword choice or resume prompt guarantees an ATS pass.
Tailor my approved resume for [verified role] using only factsheet evidence.
First show proposed edit | original fact | evidence ID | reason.
Prioritize relevant genuine work and preserve dates, titles and scope.
Label projects and volunteering accurately. Do not invent metrics or credentials.
Flag unsupported claims for removal; do not make them sound more credible.
Then produce a readable draft with standard headings in the employer's requested format.
List every unresolved factual question and remind me to inspect the exported file before submitting.A cover letter needs a factual reason for this role, one or two relevant examples and a clear connection to the employer’s stated work. Ask for missing motivation rather than accepting invented enthusiasm for a company you have not investigated. Verify any company fact on its own official site.
Using the verified role, approved factsheet and my genuine reason for applying, draft:
1. A concise cover letter if one is needed.
2. An optional short message for [legitimate recruiter/contact channel].
Use only supported claims and employer facts with source links.
Do not imply a prior relationship or invent achievements.
If a required detail is missing, ask me instead of filling it in.
Show intended recipient/channel separately. Do not send, submit or schedule either draft.
Respect any no-contact or AI-use instruction in the posting.For the fictional candidate, a useful opening says that the junior role’s SQL and reporting tasks align with F01 and F02. It should not claim an experienced analyst’s client portfolio. The purpose is to make the real evidence easy to understand, not disguise the gap.
Download the blank 100-opportunity CSV. It has 27 columns and IDs J001–J100; all other cells are empty. No account, macro or tracking service is required.
The CSV is a data template, not a standalone application or a formatted workbook. It does not retain spreadsheet dropdowns, colours, formulas or frozen panes. Open it in your chosen spreadsheet or text editor and keep a private working copy. Unused ID-only rows are not discovered jobs and must not be counted as such.
| Field group | Exact columns | What to record |
|---|---|---|
| Identity and source | opportunity_id; employer; role_title; employer_job_url; posting_id; source_channel | Stable ID, verified employer source and deduplication key |
| Verification and constraints | date_discovered; last_verified_date; location; work_mode; work_authorization_check; deadline_and_timezone | Actual checks, employer conditions and explicit unknowns |
| Fit and decision | must_have_requirements; evidence_refs; gaps_or_unknowns; decision; stage | Required criteria, genuine factsheet IDs and a human decision |
| Application evidence | resume_version; cover_letter_version; human_review_date; submitted_at; confirmation_reference | Exact reviewed files and actual submission evidence |
| Next step and result | follow_up_due; next_action; next_action_due; outcome; notes | One useful action and observed result; silence remains unconfirmed |
Use stage values consistently: Discovered, Needs verification, Reviewed, Drafting, Ready for human review, Applied, Interview or Closed. Keep decision values Apply, Hold and Skip separate. Leave event dates blank until the event occurs. Use YYYY-MM-DD for dates and an explicit timezone for a deadline or submission timestamp.
The fictional sample above demonstrates three populated records. J001 is still Drafting, so its submitted_at and confirmation_reference remain blank. J002 has a verification task, not an invented application date. J003 records a closed posting, not a rejected application. A tracker becomes misleading when planned actions are formatted as completed events.
Keep URLs free of credentials and session parameters. If you later import untrusted text into a spreadsheet, review cells that begin with formula characters such as =, +, - or @ before opening them. The supplied blank template contains no formulas or candidate data; that does not make every future pasted value safe.
Compare my current tracker with this small batch of verified notes.
Match existing opportunities by employer/posting ID or canonical URL before adding rows.
Show proposed field changes with the supporting event or source.
Do not infer submission, rejection, interview, deadline or eligibility.
Keep unknown dates empty. Count only rows with actual opportunity details.
Flag duplicates and stale verification. Preserve stable opportunity IDs.
Return a change table for review and a new CSV version only after I confirm the changes.
Do not write to connected services or contact anyone.Before submitting, inspect the actual role, destination and latest documents. Check every claim, dates, eligibility answers, spelling, contact details and filename. Confirm the export is readable and the portal accepts its format. Read the employer’s assessment and AI-use instructions; drafting assistance does not grant permission to use AI in a restricted assessment.
Submit through the employer’s or platform’s normal workflow yourself. Record Applied only after a real confirmation or a verified application-history entry. Clicking Submit, generating a resume or hearing “done” from an assistant is insufficient evidence. If the response is uncertain, inspect the portal and legitimate confirmation channel before repeating the application.
Keep confirmation references minimal and private. You usually need a date and enough information to find the receipt, not a public copy of a private portal URL. Do not place an identity number or security answer in notes simply because the application form asked for it.
Here is how the pieces fit together for the fictional Cedar Metrics role. The invented posting asks for SQL, spreadsheet reporting and clear written summaries. The invented candidate has F01 and F02, with no paid analyst employment. They first check the real-world equivalents of the employer source, open status, hiring country and portal instructions; in a real search, any unresolved essential condition would put the role on Hold.
The requirement matrix maps SQL to F01 and spreadsheet summaries to F02, while keeping paid analyst experience absent. The candidate chooses Apply/Drafting. Their resume uses the accurate personal-project bullet above and describes F02 as volunteer reporting work. It does not adopt an experienced analyst title or a made-up efficiency result.
| Posting criterion | Candidate evidence | Review decision |
|---|---|---|
| Required: SQL querying | F01: personal SQL project on synthetic orders | Relevant practice; describe scope honestly |
| Required: spreadsheet summaries | F02: volunteer weekly reporting task | Use the actual contribution; do not invent business impact |
| Required: clear written explanation | F01 check notes and F02 unresolved-entry explanation | Inspect the real equivalents before claiming proficiency |
| Preferred: paid analyst experience | None in the fictional factsheet | Keep the gap; never relabel projects as employment |
When the drafts are complete, the candidate moves the stage to Ready for human review. They check the wording against F01/F02 and the employer’s instructions, then inspect the exported resume and letter. After the final check, they record the actual human-review date and retain that stage while awaiting manual submission. They complete the portal themselves, answering eligibility questions from their own verified facts.
A concrete review would reject “professional analyst” and the unsupported 40% saving, confirm the personal-project and volunteer labels, and check whether the employer accepts these examples for a junior role. The fictional posting treats paid experience as preferred, so that absence does not automatically defeat its three essential criteria. The candidate still owns the suitability decision.
If a genuine confirmation arrives, they record its actual timestamp and a private reference and change the stage to Applied. Without confirmation, the stage stays unresolved while they inspect application history; they do not invent a receipt or immediately submit again. No such submission occurred in this fictional example, so the sample CSV deliberately remains Drafting with blank event dates.
Next they practise a SQL question on missing values and a behavioral question about explaining a data uncertainty. If a real interview invitation later arrives, they record the actual event and prepare against that interview’s instructions. If no reply arrives, the weekly review keeps silence separate from rejection. J002 remains on Hold until its hiring-country question is resolved, and the closed J003 stays outside the active queue.
Use the verified requirements to choose practice topics. For a technical role, ask Claude to pose one question, wait for your attempt and then discuss the reasoning. Check important technical explanations against the relevant official documentation. A convincing model explanation may still contain an error, and a synthetic exercise is not evidence that you can perform the task unaided.
Act as a practice interviewer for [verified role requirements].
Ask one question at a time and wait for my attempt before giving feedback.
For technical answers, discuss assumptions and verify claims against official documentation.
For behavioral answers, use only experiences I provide; ask about my own action and contribution.
Help me clarify situation, task, action and result without inventing a story or metric.
Flag gaps for study rather than supplying an answer to pass off as experience.
This is private practice, not assistance during a restricted live assessment.For behavioral practice, distinguish what you did from what the team achieved. The fictional candidate can describe how they checked a volunteer spreadsheet and explained uncertainty to its owner. They should not memorize a fabricated story about leading an analytics department. If a result was qualitative, say what changed without inventing a percentage.
For learning from permitted source material, the NotebookLM study workflow offers a separate method for checking explanations against the original source.
A useful follow-up belongs to a specific application and a legitimate contact channel. Follow the posting’s instructions and stop when asked. Choose timing based on the employer’s process and any stated response window; there is no universal interval that guarantees a reply. Keep messages concise and factual.
Claude’s Google Workspace connector can read and draft Gmail messages and also send, reply or forward, with approval by default. That is a reason to deliberately retain draft-only instructions and inspect actual permission and organization settings. It is not a read-only product. Connecting an inbox is optional; you can draft from a minimized note and send the final message yourself.
From my confirmed application events and current tracker, prepare:
A. A follow-up draft for [one opportunity] only if the employer's instructions allow it.
B. A weekly report counting verified opportunities, decisions, confirmed applications and actual replies separately.
Distinguish missing values from zero, recent applications from older ones, and silence from rejection.
List stale checks, unresolved holds and next actions with dates only where supported.
Suggest one improvement to targeting or evidence for my review.
Do not invent a response rate, infer an ATS failure or send any message.For a fictional arithmetic example, six confirmed submissions and one documented interview invitation are different counts from ten verified opportunities. They are not this website’s results or a forecast for the reader. Compare cohorts only after allowing for their age and context. A small sample or a silent inbox cannot identify which resume sentence caused an employer decision.
Each week: remove closed roles from the active queue; resolve eligibility holds; check upcoming deadlines; inspect all ready-to-submit files; reconcile actual confirmations; practise one evidence gap; and choose one change to evaluate next. Keep previous versions so you can tell what changed. Do not rewrite the whole strategy in response to one rejection or send repeated messages to manufacture activity.
Connectors can reduce copying between approved services, but they also add permissions and retained data. Review the trusted provider, actual actions and access scope. A connection is not an exemption from a job platform’s terms. Do not install an unrelated scraper because its name mentions Claude or Indeed.
Current Anthropic documentation describes recurring tasks on paid plans, including remote runs using account files and connectors when the computer is asleep. Tasks needing local files or applications run locally; older local Pro/Max tasks created before 6 October have a separate local path. The unified chat/Cowork interface is rolling out gradually, so accounts can have different controls.
A good first scheduled job is a draft weekly tracker review using a selected file, an explicit timezone and no permission to contact employers. Inspect the instructions, source version, approval mode and first result. A task that depends on a file available only on your computer cannot be treated as a guaranteed cloud run. If scheduling is unavailable, a manual reminder preserves the workflow.
Propose a weekly job-search review at [day/time/timezone] using only [approved tracker source].
First show the schedule, source, local/cloud requirements and approval mode for my review.
Return stale verification, unresolved decisions, upcoming supported deadlines and one improvement suggestion.
Do not submit applications, send messages or modify employer/candidate records.
If the source is missing, stale or inaccessible, report the gap and stop.
Tell me how to inspect, pause or change this task in my actual interface.
Do not create recurring work until I approve its exact setup.Usage and context limits remain separate. Break work into manageable batches, save reviewed outputs and start a fresh focused chat when needed. A 100-row template does not promise 100 unlimited generations in one session. Avoid pasting the whole search history into every request when only one verified role and a short factsheet are needed.
Our ChatGPT apps permissions guide covers the broader distinction between connecting a service and approving an external action.
Review your actual Claude data controls and account terms before supplying personal work. Consumer model-improvement choices, safety exceptions and feedback handling differ from business arrangements. Raw connector-content exclusions do not mean text copied into a chat is categorically excluded. Retrieved Google connector data can remain with its associated chat.
Deleting history and erasing every back-end copy are different claims. Anthropic describes ordinary retention timelines and exceptions. Incognito is not zero retention or a guarantee that an organization’s exports or compliance systems cannot include the conversation. Use minimized material and the appropriate account rather than relying on a privacy label alone.
| Problem | What to check | Safer recovery |
|---|---|---|
| Role looks suitable but eligibility is unclear | Official hiring-country and work-authorization requirements | Set Hold; resolve the question before tailoring or applying |
| Claude invents a career claim | Master factsheet and evidence ID | Remove it and rerun the claim audit on the entire draft |
| File output loses content or formatting | Downloaded file, extraction and requested format | Use a smaller text packet; inspect export before submission |
| Connector or schedule is unavailable | Account rollout, permissions and local/cloud requirements | Use the manual workflow; do not bypass access restrictions |
| Submission confirmation is missing | Portal application history and legitimate receipts | Inspect first; keep status uncertain until verified |
| No response arrives | Actual submission age and employer instructions | Keep silence unconfirmed; review targeting without inventing a cause |
If untrusted job text tells an assistant to reveal your files or contact someone, treat it as source content, not an instruction you authorized. Keep sensitive records outside the task, inspect connector actions and stop unexpected requests. No prompt alone prevents every malicious instruction or factual mistake.
This guide does not authorize or recommend a bulk-apply bot. It organizes up to 100 relevant opportunities with supported discovery and reviewed drafts. Submit through the legitimate employer or platform process yourself. The official Indeed connector pages reviewed here do not document application submission.
Only the employer’s actual hiring conditions and your genuine eligibility answer that question. Pakistan appearing on Claude’s supported-country list does not establish worldwide remote-job eligibility or availability of every connector. Keep a role on Hold when the hiring country is unclear.
No. Use truthful relevant language, standard headings, a readable export and the employer’s requested format. Do not claim an ATS score or outcome you have not measured, and do not add unsupported qualifications to match keywords.
Neither is required for the basic manual pipeline. Start with a small factsheet, verified role and CSV. Consider a supported connection or paid scheduling only when it solves a specific bottleneck and you accept its permissions, limits and data handling.
Verify one role, map its requirements to genuine evidence and choose Apply, Hold or Skip. If Apply is justified, draft one truthful resume version and inspect the export. Leave submission fields blank until you actually submit and confirm. That complete cycle teaches more than generating 100 unchecked documents.
Product facts are checked against the sources below. Access dates describe research, not hands-on testing.
Official web, Windows/macOS desktop and iOS/Android entry points; supported-location and minimum-age requirements; prompting and variable usage limits. No account execution was performed for this article.
Current supported-location list includes Pakistan and the United States. General Claude access does not establish availability of every connector, beta or subscription payment method.
Current projects retain files and instructions across chats; Free users can create up to five projects. New cloud-thread projects beta begins with select Pro/Max Claude Code users; do not assume every chat account has the beta. Paid project RAG differs from basic access.
CSV, PDF, DOCX, TXT and other listed uploads; XLSX requires code execution/file creation. Chat and project upload limits differ; non-PDF documents use text extraction. Long-PDF visual handling is restricted; article uses small text/CSV inputs rather than claiming universal size support.
Code execution/file creation on Free/Pro/Max/Team/Enterprise across web, desktop and mobile; creation of downloadable spreadsheets, documents and PDFs; CSV analysis. Organization controls and network egress differ; MCP connections require separate review. Current page lists a 30 MB computing/file limit unlike the general chat-upload page.
Current supported models, cited web search and direct page fetch; Team/Enterprise owner enablement, usage cost and variable links/connectivity. Legacy interface has a Web search toggle; the new unified interface searches when helpful without that toggle. No promise to access restricted job boards.
Web connectors available to all users; Customize>Connectors and directory/authentication workflow; source-system permissions still apply. Team/Enterprise organization enablement and action restrictions; third-party service terms/infrastructure differ. Free custom connector limit is one; a connector is not a platform-access exemption.
Gmail search/read/draft/send/reply/forward with approval by default; Calendar events and Drive data; separate live Docs/Sheets/Slides editing beta. Gmail exposes attachment metadata, not content. Retrieved data retained with chat; raw connector-data training exclusion has copied-into-chat/model-improvement caveat. Source-system access and organization controls apply.
Recurring/on-demand tasks on all paid plans; current remote runs use connectors and account files even when the computer is asleep. Pro/Max tasks created before 6 October that ran locally remain local; tasks requiring local files/apps run locally. Scheduled sidebar, schedule/instruction/approval-mode review and pause/delete controls; gradual unified-interface rollout.
Gradual Pro/Max web/desktop/mobile rollout merges chat and Cowork; users on the same plan can have different interfaces. Cloud work differs from local-file/app tasks that need Desktop open. Do not give one interface path as universal or equate cloud scheduling with local computer access.
Paid five-hour session and weekly usage limits; shared usage across product surfaces, model/effort/task-length effects; context length is separate from usage. 100 tracker rows do not imply 100 unlimited generations, autonomous submissions or guaranteed task completion.
Consumer training depends on user choice plus safety/explicit-opt-in exceptions; raw connector content exclusion does not cover direct copies into chat. Incognito not used for model improvement; feedback may include the related conversation. Do not describe personal Claude accounts as universally no-training.
Consumer chat deletion removes history immediately and ordinarily clears back-end storage within 30 days; model-improvement/feedback/safety/legal retention exceptions differ. Turning off improvement does not undo already trained/in-progress models. No immediate universal erasure or short-lived-only storage claim.
Incognito outside projects excludes ordinary history and memory/model-improvement use, but retention remains 30 days by default or longer under organizational settings. Team/Enterprise exports and Enterprise Compliance API can include incognito. It is not a private-to-employer or zero-retention guarantee.
LinkedIn prohibits third-party scraping software and tools that automate activity on its website; a Claude connector or browser agent does not automatically authorize LinkedIn scraping, messaging or application automation.
Section 8.2 prohibits scraping/copying services through software/scripts and bots/unauthorized automated access or messages; false identities and credential/cookie sharing are prohibited. Use normal human access and platform-supported actions, not external bulk bots or security-control bypass.
Native job-search alerts offer daily/weekly email/app notification choices and can be managed after search. Discovery via native alerts does not constitute application submission or waive third-party automation restrictions.
Indeed’s “Applying to Jobs — Limits” and “Site Rules” prohibit outside-authorized-tool application bots and automated access/data mining/submission absent permission. “External Sites — Third Party Automations” permits specifically sanctioned third-party AI connectors under conditions; this is not blanket automation permission. Applications/interviews/delivery are not guaranteed; employer legitimacy must be checked.
Signed-in native search/alert creation, daily/weekly frequencies and pause/edit controls. International searches may not allow alert creation; no new matching jobs means no guaranteed daily email. Source is US-localized, not proof of equal regional features.
Official Anthropic-verified connector made by Indeed, sign-in required; listing exposes search_jobs/get_job_details for discovery, requirements and direct links. It does not document application submission. Publisher documentation is linked; catalogue presence is not proof of regional availability or executed account access.
Official beta Claude Connector; Job Search, Job Detail, Get Resume (profile experience) and Get Company Data documented; account login required in Claude.ai/Desktop. Tool descriptions differ from two-tool marketplace listing; inspect actual granted tools. Neither reviewed source documents application submission or a complete country list.
Continue with context
AI Productivity Tools / Practical guide
Choose ChatGPT apps for live search, workspace sync and supported actions. Understand plugins, provider access, approvals and data handling before connecting your tools.
Read guideAI Learning & Productivity / Practical guide
Use NotebookLM, now Gemini Notebook, to organize a bounded study set, inspect citations and turn reviewed explanations into practice. Keep source freshness, privacy and limits in view.
Read guide
