Build an AI Job Search Assistant with Google Sheets

Piotr Grudzień
Published 9 min read
A Monzo backend role saved in Google Sheets with Shortlist Desk's confirmation

An AI job search assistant can check a company’s current openings, explain how a posting fits your preferences, and save the roles you choose to Google Sheets. You will build Shortlist Desk, a personal assistant that keeps the reason for your interest beside the original job link.

Imagine your search is fairly specific: backend engineering, London or UK remote, and a chance to use your Python experience. You already know one company you would like to join. The useful next step is finding a role worth reading closely, then remembering both its appeal and its compromises.

A Monzo backend role saved in Google Sheets with Shortlist Desk's confirmation

The goal: one role with its original link and the tradeoff you approved, saved in Google Sheets. The real saved row is paired with Quickchat AI's confirmation.

The example follows Monzo’s Backend Engineer III posting. It allows London or UK-remote work, while its application code is primarily Go. That makes it a useful near match for our fictional reader: a location fit with a language tradeoff, rather than an invented “perfect match.” Openings change; the source was checked on 17 November 2026.

The Google connection, saved row and save conversation were captured from the connected test Agent. Configuration screenshots use a local copy of the same setup.

Download the Shortlist Desk starter kit. It includes the spreadsheet header, four complete Action recipes, the Main Prompt and a test checklist. Use the files as copyable settings, not as an automatic import.

Follow the steps to choose your criteria, create the assistant, read current jobs, check the fit, connect your shortlist, and save and verify a role.

Step 1: Which companies and jobs do you want to track?

Start with one employer and a short set of preferences. A small, explicit search is easier to inspect than a long list of vaguely relevant vacancies.

  1. Open the employer’s careers page and follow its job links. For this example, use Monzo’s public Greenhouse board. The board token is monzo.
  2. Write down your role, location and work preferences. Our reader wants individual-contributor backend work, London or explicitly UK-remote, with Python experience relevant to the discussion.
  3. Separate requirements from preferences. If Python must be the main production language, say so. If you would learn Go for the right role, say that instead.

Search criteria separating backend work, London or UK remote, and a preference for Python

Decide which criteria rule a job out and which start a conversation. These are fictional search preferences, not an assessment of a real applicant.

You do not need a CV for this build. The assistant compares your stated preferences with the posting; it does not decide whether you are qualified or eligible to work in a country.

Step 2: How do you create your AI job search assistant?

Create a Quickchat AI Agent and give it one job: inspect configured company boards and help you keep a personal shortlist.

  1. Create a Quickchat AI account and open your Agent.
  2. Under Identity, name it Shortlist Desk. Paste main-prompt.txt from the kit into Main Prompt and save.
  3. Keep this personal workflow in the authenticated dashboard’s AI Preview. Under Your Website → Install → Availability and access, turn Widget & Embed access off and set Chat Page Access to Nobody. Anyone who can reach an Agent with a Sheet-reading Action could ask it about that Sheet.

Shortlist Desk's Identity settings and Main Prompt in Quickchat AI

The Main Prompt sets the decision rules. The Actions you add next give the assistant access to current evidence.

The prompt makes three distinctions: posting evidence versus inference, proposing a save versus confirming it, and saved versus applied. Those distinctions will matter more than giving the assistant a job-hunting personality.

Step 3: How do you read a company’s current job openings?

Use two read-only Actions: one lists the employer’s jobs, and the other reads a selected posting. Greenhouse’s Job Board API allows these public GET requests without an API key.

Open Actions & MCPs → Add Action → API Action. Create the two Actions below, using the kit’s exact descriptions. The API Action builder guide explains the editor and its plain-English builder.

Editor fieldAction 1Action 2
API Action NameList Monzo jobsRead selected job
MethodGETGET
What to ask the user firstNo parametersRequired Text parameter job_id
HeadersAccept: application/jsonAccept: application/json
BodyEmptyEmpty

In API Endpoint, use these URLs respectively:

https://boards-api.greenhouse.io/v1/boards/monzo/jobs
https://boards-api.greenhouse.io/v1/boards/monzo/jobs/{{job_id}}

For the detail Action, set the job_id description to: “The exact job ID returned by List Monzo jobs and Read selected job. Never invent an ID.” Insert the parameter through Add AI Data so it becomes a variable chip in the URL.

Keep each response available to the Agent. The list’s meta.total lets you check coverage; job_id comes from the posting’s id, not internal_job_id. Fetch compact listings first, then descriptions only for the roles you discuss. If the returned list is incomplete or too large to process, report that limit instead of calling it a complete search.

Activate both read Actions. In AI Preview, send:

I work mainly in Python and want an individual-contributor backend role in London or explicitly UK remote. Check current Monzo jobs and explain the fit and any gaps for Backend Engineer III. Keep it concise.

Step 4: How can the assistant explain whether a role fits?

A useful match explanation pairs each claim with something the employer actually wrote. Open the original posting alongside the assistant’s answer and check the location, responsibilities and requirements yourself.

The employer's backend posting beside the assistant's explanation of the UK location fit and Go tradeoff

Compare both sides of the answer. A role can fit your location and interests while still asking you to learn another language.

For the example role, London or Remote (UK) supports our location preference. The posting’s description of Go as its main application language means the assistant should not call this a Python role. If Python is non-negotiable, ask it to exclude the role. If you are open to Go, keep that tradeoff in the note you save.

Try a second question:

Does UK remote mean I can do this job while living in Spain? Use only what this posting says.

The expected answer distinguishes remote work within the UK from permission to work abroad. Silence about a country is not permission. The same rule applies to salary and sponsorship: use an explicit statement or leave the answer unknown.

This is also your first tuning loop. If an answer overstates a fit, inspect the called detail Action in Inbox, identify the unsupported sentence, tighten the Action’s description, and rerun the same question. Change one instruction at a time so you can tell what helped.

Step 5: How do you connect a job shortlist to Google Sheets?

Connect Google Sheets after the assistant can find and explain a role. The integration creates a spreadsheet and starter logging Action; you will add a Shortlist tab and two purpose-built Actions.

  1. Open Actions & MCPs → Add Action → Google Sheets → Connect. Sign in to your Google account and approve the requested access.
  2. Open the spreadsheet from the connection dialog. The integration uses per-file access, not unrestricted Drive access. The Google Sheets connection guide explains this setup.
  3. Disable the generated log_lead_to_google_sheet Action while configuring the shortlist. Leave its name intact; the connection uses it to find the starter.
  4. Add a tab named exactly Shortlist. Paste Shortlist.tsv from the kit into cell A1. Freeze row 1, enable text wrapping, and give the title and match-note columns enough width to read. Keep all ten columns visible and in the supplied A:J order, with source_key in A1 and checked_at in J1.

The Google Sheets connection surface in Quickchat AI

Use the spreadsheet created by this connection. Copying an unrelated Sheet's ID into an Action does not grant access to it.

The ten columns keep the visible decision beside the identifiers needed to check it:

source_key | company | job_id | title | location | job_url | status | match_note | source_updated_at | checked_at

Shortlist column diagram grouping the employer facts, decision note and source identifiers

Template diagram. The status starts as saved. Keep both the provider's update time and the time you checked the posting.

For the optional six-column reading view shown at the start, add a separate tab named Saved roles and paste the kit’s saved-roles-view.txt formula into A1. It displays the company, role, location, original link, status and match note. Edit statuses and notes in Shortlist; Saved roles updates from that tab.

The Saved roles tab in Google Sheets with a Monzo backend role, original posting link and approved match note

The optional Saved roles view shows the decision in six columns. Both Actions still read and write the ten-column Shortlist tab.

Create Read my shortlist and Save selected role using action-recipes.json in the kit. Find your spreadsheet ID inside the generated starter Action’s existing URL and copy it into both recipes in place of APP_CREATED_SHEET_ID. Keep that ID fixed in the editor.

SettingRead my shortlistSave selected role
MethodGETPOST
RangeShortlist!A1:J501Shortlist!A:J:append
Authorization headerBearer {{google_sheets_access_token}}Bearer {{google_sheets_access_token}}
BodyEmptyThe kit’s one-row JSON body
Query parametersNonevalueInputOption=RAW, insertDataOption=INSERT_ROWS

The read URL ends in /values/Shortlist!A1:J501. The save URL ends in /values/Shortlist!A:J:append. Both begin with https://sheets.googleapis.com/v4/spreadsheets/ followed by the fixed spreadsheet ID. The save Action also needs Content-Type: application/json.

The row key is greenhouse:monzo:<job_id>. The request fixes the company to Monzo and the status to saved; {{current_timestamp}} supplies the check time in UTC. The model fills the returned job fields and your approved match note. Google documents the read contract and append contract.

After reviewing the settings, activate Read my shortlist and Save selected role. The POST Action stays inactive until you switch it on.

Step 6: How do you save a role and check the result?

Saving is a short conversation with a visible checkpoint. The assistant reads the shortlist, proposes a role and note, and waits for a separate confirmation.

Send “Save Backend Engineer III to my shortlist.” Read the proposed note before approving it. It should retain the Go tradeoff, not flatten the role into a generic recommendation.

Shortlist Desk proposes a role and match note before asking for confirmation

Review the decision you will want to remember later, not just the title.

Reply “Confirm save with that match note.” The prompt tells the assistant to recheck the posting and existing keys, append one row, then read the shortlist again.

Diagram of reading the shortlist, confirming a role, rechecking it, appending and reading back

The repeated read catches ordinary accidental repeats. It is not an atomic uniqueness guarantee if multiple conversations save simultaneously.

Inspect the result in two places:

  1. In Inbox, expand the called Action. Check that updates.updatedRange spans columns A:J, such as Shortlist!A2:J2 for the first saved row.
  2. In your Google Sheet, inspect that exact row. Each value must sit under its matching header, from source_key in column A to checked_at in J. Verify the job ID, original URL and approved note. A successful reply alone is not proof that Google received the row.

Our first live check placed the row in B:K while edge columns were hidden. We made all ten columns visible and reran the same recipe; the row then aligned with A:J.

Quickchat AI confirms Shortlist!A2:J2 and shows successful save and readback Actions

One confirmed save, followed by a readback. The append returned Shortlist!A2:J2, and the follow-up read found the same source key and approved note.

Ask to save the same role again. The assistant should find its exact key and keep one row. If a request times out, read before retrying. The recipe reads 500 data rows; expand its configured range before the shortlist reaches that limit.

Shortlist Desk finds the saved Monzo role and confirms that no duplicate was added

A second confirmation found the existing source key. The Sheet still contained one saved row, and no second append ran.

You can later change saved to applied or add interview notes in the Sheet yourself. Applications still happen through the employer’s original link.

Step 7: How do you test the assistant and add another company?

Test the decisions as well as the happy path. Use Testing → Simulation with the kit’s test messages, then inspect the corresponding Inbox calls. Simulation can execute active writes, so start with the save Action disabled and enable it only for your test Sheet.

The Quickchat AI testing surface for the Shortlist Desk demo

Keep the same questions when you change a prompt. Check the actual Action result as well as the wording of the reply.
TestWhat to verify
Ask for Python-only backend workA Go role is not presented as a Python match
Ask about remote work from SpainUK remote is not expanded to every country
Ask to save, without confirmingNo append occurs
Confirm, then repeat the saveOne matching source key remains
Posting disappears or Google access failsNo invented current posting or successful save

To use a different Greenhouse employer, copy the kit and replace monzo in both Greenhouse URLs, Monzo in the save body’s company cell, and greenhouse:monzo: in the body, descriptions and Main Prompt. Update the employer references in the Action names and descriptions too. Verify that employer’s board token, then repeat the read, fit, confirmation and readback tests. Keep each copy tied to one fixed employer and its own Agent.

If you prefer assistant-led setup, the verified Quickchat AI ChatGPT app and Claude integration explain the account connection. Give your assistant the kit and ask it to configure the four Actions, preserve the fixed board and spreadsheet, and leave saving inactive until you finish Google consent and review the settings.

Your finish line is one role with a reason, a tradeoff and a link you can trust enough to investigate. Repeat the search when you want to check the employer again; the shortlist keeps the context you would otherwise have to reconstruct.

Frequently asked questions

Can an AI job search assistant search every company?

This build searches the public Greenhouse boards you configure. It starts with Monzo and checks when you ask. It does not search every employer, LinkedIn or private applicant-tracking systems.

Do I need a Greenhouse API key?

No. Greenhouse’s public Job Board GET endpoints allow unauthenticated reads of published jobs. This tutorial uses the list and job-detail endpoints. Submitting applications is a separate capability and is not configured here.

Is this a Google Sheets job application tracker?

It starts as a shortlist of roles you choose to investigate. Each row contains the company, job ID, original URL, location, match note and saved status. Change the status yourself after you apply; saving a row does not submit an application.

Can I use an existing Google Sheet?

Follow this recipe with the spreadsheet created by Quickchat AI’s Google Sheets connection. The connection uses per-file access and does not grant unrestricted access to Drive. Pasting an arbitrary existing spreadsheet ID into an Action does not grant permission to that file.

Does remote mean I can work from any country?

No. Read the posting’s explicit location and remote-work conditions. Remote (UK) supports working remotely within the UK, not automatic eligibility from another country. When the posting does not answer a location question, the assistant should say so.

Will the assistant apply for jobs or send daily alerts?

No. This recipe checks openings on request and saves roles after confirmation. Apply through the employer’s original link. Scheduled checks and notifications require a separate trigger and delivery workflow.

Can ChatGPT or Claude help me set this up?

Yes. They can adapt the starter kit’s prompts and HTTP recipes. An authorized Quickchat AI account-management connection can also configure supported Agent settings. Google sign-in and consent remain with the account owner.

Do I need to write code?

No. You configure four HTTP Actions in Quickchat AI, paste the supplied Main Prompt and add a Shortlist tab to the app-created Google Sheet. The downloadable kit contains the complete field values and a test checklist; it is not an automatic Agent import.

Quickchat AI Quickchat AI Ask Quickchat AI a question