AI won't find you a job. But it is great at removing busywork: breaking a posting into requirements, comparing it with your experience, researching a company, drafting a letter, quizzing you before an interview. Set it up properly once and it saves you several hours a week.
This is part 6 of the series. It covers the assistants and prompts themselves. How to connect them to your mail, calendar and spreadsheets is in part 7 on MCP connectors.
Which assistant to pick
The honest answer: any of the big three works for a job search. The differences are in the details.
| Assistant | Price (November 2026) | Why it helps in a job search |
|---|---|---|
| Claude | Free. Pro 20 dollars a month (17 billed annually). Max from 100 dollars | Projects for persistent context, connectors to Gmail, Calendar and Drive, Research mode, Claude in Chrome extension on paid plans |
| ChatGPT | Free. Go, Plus and Pro are paid, check the current price on the vendor's site | Projects, deep research, custom GPTs, agent mode |
| Gemini | Free. Google AI Pro, 909.99 UAH in Ukraine | Deep Research, tight integration with Gmail, Docs and Drive |
| Perplexity | Free. Pro is paid, check the current price on the vendor's site | Search with sources, handy for company research. Comet browser is free |
Availability: Claude, ChatGPT and Gemini work in Ukraine, except the temporarily occupied territories. Check the current country lists on each service's site.
My advice: start with the free plan of the assistant you already use. Pay for a subscription only during the active search, and only if you hit the limits.
Step 1: a base profile
This is the single most important thing you can do. One document that holds everything about you as a developer, in more detail than your CV:
- every project: what the product did, your role, the stack, what exactly you built, the result in numbers if you have them;
- skills with an honest level ("React: daily for 3 years", "PostgreSQL: simple queries, indexes in theory only");
- what you're looking for: level, format, domain, salary, what you don't want;
- languages, location, time zone.
Put the document into a Project (Claude, ChatGPT) or a Gem (Gemini). Now every chat has the context, and you don't have to explain who you are each time. Leave your phone, email and address out; they aren't needed.
Step 2: prompts that work
Job analysis.
Here is a job posting. Pull out: required skills, nice-to-have skills, level, domain, working language, red flags (vague duties, unrealistic requirements). Compare with my profile in this project and tell me honestly how well I fit, what I have, what I'm missing and what to brush up before the interview. Don't flatter me.
Company research (better in Perplexity or a Research mode, with sources):
Collect on company X: product, business model, size, stack (if public), news from the last year, reviews of the hiring process on DOU and Glassdoor. Give a link for every fact. If you couldn't find something, say so.
CV tailoring and letters. Prompts are in part 4. The key rule: "don't invent facts that aren't in my profile".
Replying to a recruiter.
The recruiter wrote this. Draft a short reply, 3 to 5 sentences, friendly and to the point. My range is such and such; I'm free for a call Tuesday or Thursday after 15:00 Kyiv time.
Interview practice.
You are interviewing me for a Middle Frontend Developer role at a product company. Ask me one question at a time on JavaScript, React and the browser, starting easy. After each answer, score it from 0 to 3, say what was missing and ask a follow-up, like a real interviewer.
Post-interview review.
Here are the questions I got and how I answered. Where was I weak? What should I review before the next round?
Agents and browser assistants
In 2026 assistants don't just answer; they act: Claude in Chrome, ChatGPT agent mode, the ChatGPT Atlas and Perplexity Comet browsers. They can open a site, find openings, fill in a form.
Where this genuinely helps: collecting openings from several company career pages into a sheet, finding a team's contacts on a company site, checking whether a role is still open.
Where to stop:
- Don't let an agent apply to jobs on its own. It will answer form questions for you, and those answers will be generic. You won't know exactly what was sent.
- Don't let an agent message people in your name without review. Drafts yes; sending by hand only.
- Remember prompt injection. A page the agent reads may contain hidden text with instructions for the AI. An agent in a logged-in browser with access to your mail is a risk. More in part 7.
- Platform rules apply to agents too. An agent crawling LinkedIn at scale is, from LinkedIn's point of view, a bot.
Privacy
Before you put your CV and correspondence in, check the training settings for your data.
- Claude: Settings → Privacy, a toggle along the lines of "Help improve Claude". Incognito chats are available.
- ChatGPT: Settings → Data Controls → "Improve the model for everyone".
- Gemini: turn off Gemini Apps Activity. This also turns off chat history.
- Perplexity: turn off AI data retention in settings.
Menu labels change, so look for a Privacy or Data section. And regardless of settings, don't paste passport details, NDA-covered contracts or other people's personal data into a chat.
Where AI does harm
- Identical letters. Recruiters already recognise the "I am thrilled to apply" style. Rewrite in your own words.
- Invented experience. AI happily "improves" you. Check every line.
- Hints during the interview. Covered in part 8. In short: don't.
- Replacing learning. AI can explain the event loop, but you are the one answering in the interview. Understanding comes from practice. That's what ShortLesson's short lessons are for, with the Ask AI tutor answering about the exact spot in a lesson you didn't get.
Next
Your assistant knows you, but it can't see your mail, calendar or tracker. MCP connectors change that; see part 7.