# How to turn a job description into a scored shortlist with Claude (https://www.hyreflow.ai/blog/how-to-turn-a-job-description-into-a-shortlist-with-claude)

By Niklas Huetzen, Co-founder, hyreflow. Published 12 Oct 2026.

> To turn a job description into a scored shortlist with Claude, connect hyreflow as a connector, paste the job description and ask for a ranked shortlist. Claude reads the must-haves, searches several people databases, pulls each candidate's dated work history, scores every person against the brief with the evidence quoted, and looks up personal emails only for those who pass. You read the list before a client sees it.

- A score built on a job title is a guess. The shortlist gets good when Claude scores on dated work history.
- In a real run on 10 October 2026, a shortlist of ten with seven personal emails cost 25.85 hyreflow credits, about $2.53.
- The score orders the list. It does not replace the call, and it is only as sharp as the must-haves you give it.

Claude can score a candidate against a job description in seconds. The hard part is giving it real candidates and real work history to score. This guide shows the prompt, one real run and what it cost.

## What does Claude need to turn a job description into a shortlist? [#what-does-claude-need-to-turn-a-job-description-into-a-shortlist]

Claude needs two things to turn a job description into a shortlist: candidate data and the job description as text, a file or a link. Claude has no candidate database of its own, so the people come from a connector such as hyreflow.

Connecting hyreflow to Claude takes a few minutes and nothing is installed. Anthropic makes custom connectors available "for users on Free, Pro, Max, Team, and Enterprise plans", so a paid Claude plan is not required. The steps are in [How to source candidates with Claude](/blog/how-to-source-candidates-with-claude) and the [hyreflow setup guide](/docs/mcp-setup).

Shortlisting with AI is still uncommon. In the CIPD's 2026 resourcing survey of 1,014 UK HR professionals, 16% of organisations used AI to shortlist candidates based on a job description. The survey covers AI tools in general and did not ask about Claude or hyreflow.

<Stat value="16%" source="CIPD, Resourcing and talent planning report 2026" href="https://www.cipd.org/globalassets/media/knowledge/knowledge-hub/reports/2026-pdfs/9218-resource-and-talent-planning-2026-report-web.pdf">
  of UK organisations use AI to shortlist candidates based on a job description. Among small and medium-sized employers, 80% use no AI in any recruitment process.
</Stat>

## What should a job description contain before Claude scores candidates? [#what-should-a-job-description-contain-before-claude-scores-candidates]

A job description needs two or three real must-haves before Claude scores candidates against the job description. The must-haves are what the score is built on. "Strong communicator and a self-starter" gives Claude nothing to check against a work history.

Before you paste the job description, add three things in your own words:

* The must-haves that would make you reject a CV: a platform, a certification, a minimum number of years, a commute.
* What disqualifies someone: contractors for a permanent role, people who moved in the last six months, a sector you cannot place from.
* The rules the client only said out loud. The playbook for [scoring a longlist against the unwritten brief](/skills/score-against-the-unwritten-brief) covers these.

<Callout type="tip">
  If the brief came from a recorded call, start one step earlier. hyreflow can turn the recording into a written search brief with its open questions, as described in the [briefing call playbook](/skills/briefing-call-to-search-brief).
</Callout>

## What prompt turns a job description into a scored shortlist? [#what-prompt-turns-a-job-description-into-a-scored-shortlist]

The prompt that turns a job description into a scored shortlist asks for the outcome and sets the order of work. Paste the job description, then say how many candidates you want, how to score them and which contact details to find.

The prompt below was run on 10 October 2026 through the hyreflow connector. The job description was a public advert for a Senior Controls Engineer at a warehouse automation firm in Northamptonshire, UK. The advert asked for expert Siemens TIA Portal skills, PLC and HMI development and on-site commissioning.

Prompt (Claude):

> Use hyreflow and turn this job spec into a ranked shortlist of 10 candidates. Search the general market and, separately, the direct competitors I list below. Pull each person's real work history before you score them. Score everyone out of 100 against the must-haves, drop anyone who fails one, and quote the evidence. Then find a personal email for the people who made the list, never a work email. Tell me the cost before you spend anything.

What comes back: Ten ranked candidates with scores from 90 down to 68, a line of evidence and a gap for each one, and a personal email for seven of the ten.

The dollar figure uses the Starter rate of $49 a month for 500 credits, read from the [hyreflow pricing page](/pricing) on 10 October 2026. The full prompt with every step is in the [job spec to shortlist playbook](/skills/job-spec-to-shortlist).

## How does hyreflow score candidates against a job description? [#how-does-hyreflow-score-candidates-against-a-job-description]

hyreflow scores candidates against a job description in a fixed order. It searches wide, pulls the dated work history, scores on that history, and only then looks up contact details. Contact details are the most expensive step, so they run last and only for people who passed.

<Flow
  steps="[
  { title: 'Search', detail: '76 people found' },
  { title: 'First read', detail: '18 worth a closer look' },
  { title: 'Work history', detail: 'profiles for all 18' },
  { title: 'Score', detail: '10 passed the brief' },
  { title: 'Personal email', detail: '7 of 10 found', end: true },
]"
  caption="<><b>The run on 10 October 2026.</b> Contact details were looked up last, and only for the ten people who passed the scoring.</>"
/>

<WorksWith tools="['aiark', 'prospeo', 'leadmagic', 'wiza', 'fullenrich']" label="Tools this run called" />

A title search finds the words, not the job. The first search asked for controls and automation engineers near Northampton, and about a third of the 30 people it returned were industrial controls engineers. The rest were software test automation engineers, project controls planners and automotive software engineers, whose titles share the same words.

Dated work history changed the ranking four times in this run:

* One candidate's current title did not say "senior". The history showed more than 20 years of controls work at a direct competitor, including an earlier senior title.
* Another candidate had a senior title, held for under a year, with about five years of experience in total.
* One profile gave two different answers about the current employer. Claude kept the candidate and flagged it as a question for the first call.
* Two people listed the right Siemens skills, but their profiles carried no dated controls roles. Claude left both unscored instead of guessing.

The shortlist came back as a ranked table. The top five are shown here with names, employers and exact dates removed.

| Rank | Score | Evidence from the work history                                                                         | Gap to check on a call                                                |
| ---- | ----- | ------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------- |
| 1    | 90    | Lead-level controls engineer, more than ten years across several direct competitors                    | Already at lead level; Siemens TIA Portal is not named on the profile |
| 2    | 88    | Several years in senior controls roles, a long spell as a controls engineer before that                | Current employer needs confirming                                     |
| 3    | 85    | More than 20 years of controls design and software at one direct competitor, within commuting distance | One employer for a whole career; platform not named                   |
| 4    | 84    | More than ten years as a controls engineer, the latest years at a direct competitor                    | Siemens platform not named                                            |
| 5    | 80    | More than ten years in warehouse automation across several firms                                       | Changed employer recently; may be site support, not design            |

The score is Claude's reading of the evidence, not a measurement. Use it to order the list and to see the reasoning, then make the decision yourself.

## What does a scored shortlist cost with hyreflow? [#what-does-a-scored-shortlist-cost-with-hyreflow]

The scored shortlist in this run cost 25.85 hyreflow credits, about $2.53 at the Starter rate on 10 October 2026. Credits are spent when a tool returns data from a provider, so the reading and the scoring added nothing to the bill.

<CostPerRun
  runs="[
  { task: 'Four people searches', output: '76 people. One search returned nobody and was not charged', credits: '4.0', usd: '$0.39', date: '10 Oct 2026' },
  { task: 'Work history for 18 people', output: '18 profiles with dated roles', credits: '0.85', usd: '$0.08', date: '10 Oct 2026' },
  { task: 'Scoring 18 people against the brief', output: '10 passed, 1 failed a must-have, 2 left unscored, 5 ranked lower', credits: '0', usd: '$0', date: '10 Oct 2026' },
  { task: 'Personal email for the 10 who passed', output: '7 found, 3 not found and not charged', credits: '21', usd: '$2.06', date: '10 Oct 2026' },
]"
  note="One run, not an average. Every figure is what hyreflow charged on hyreflow credits. One search and the work history step first ran on a data-provider account connected to the workspace, which uses no hyreflow credits; both were run again on hyreflow credits to get these figures."
/>

Most of the spend was contact data. hyreflow asks several contact-data providers in turn and stops at the first verified result. Five of the seven personal emails came from the last provider asked, at 3.8 credits each, and two came from an earlier one at 1 credit each.

For a larger pool, the [job spec to shortlist playbook](/skills/job-spec-to-shortlist) shows an estimate by number of candidates. It is an estimate, not a quote. You can also ask for the count and the cost first, with the playbook for [sizing a search before you spend](/skills/size-a-search-before-you-spend).

## How do you refine a shortlist that misses the brief? [#how-do-you-refine-a-shortlist-that-misses-the-brief]

You refine a shortlist that misses the brief by telling Claude what was wrong in plain language. The search and the work history are already paid for, so a second scoring pass on the same people uses no hyreflow credits. Personal emails for anyone newly added to the list are charged as before.

Three corrections do most of the work:

* "Too senior" or "too junior", with the reason. Claude re-scores the same pool against the corrected level.
* "More like number three." Claude uses that profile as the pattern and searches again.
* "Only people at these companies." The playbook for [sourcing from named competitors](/skills/source-from-named-competitors) covers this.

Start with ten candidates, read them, then ask for more. A shortlist of ten shows you whether the brief is right for a few dollars. Two hundred candidates on a wrong brief spends the credits before you find out.

When the shortlist is right, the same conversation can [load it into your ATS without creating duplicates](/skills/push-a-shortlist-into-your-ats). Agencies can take the strongest candidate to market with [an MPC or spec campaign](/blog/how-to-run-a-spec-campaign-with-claude).

In-house talent acquisition teams can run the same scoring on people they already hold. The playbook for [finding candidates already in your own ATS](/skills/match-your-ats-to-a-live-role) scores past applicants against an open role. That matters when, in Greenhouse's 2025 AI in Hiring Report, 34% of recruiters in the US said they spend up to half their week filtering spam and junk applications.

<InlineCTA title="Run it on a live role" action="Connect hyreflow to Claude" href="/mcp">
  New hyreflow workspaces start with 25 free credits. Try the prompt above on a job you are working today.
</InlineCTA>

## Where does a scored shortlist from Claude fall short? [#where-does-a-scored-shortlist-from-claude-fall-short]

<Limits>
  A scored shortlist from Claude and hyreflow is a first pass, not a decision. The score reads seniority, sector and must-haves from a work history; it cannot hear motivation, notice period or salary. Profile data has gaps: in this run two of 18 profiles lacked the dated roles needed for a score, and several strong candidates did not name the required platform at all. Personal email coverage is strongest in the US and the UK and weaker in the DACH region and Asia. Very small employers and location-bound roles return thinner lists.
</Limits>

<Callout type="warning">
  Read every shortlist before a client sees it, and check the flagged gaps on a call. Do not ask for filters on age, gender, ethnicity or any other protected characteristic. The hyreflow playbook scores on skills, experience and evidence only.
</Callout>

## Questions recruiters ask about scoring a shortlist with Claude [#questions-recruiters-ask-about-scoring-a-shortlist-with-claude]

### How does Claude decide the score for each candidate?

Claude reads each candidate's dated work history against the must-haves in the job description and writes down the evidence it found and the gap it could not close. The score is a judgement on that evidence. It is useful for ordering a list, and it is not a test result.

### How many candidates should I ask for?

Start with ten. In the run above, hyreflow found 76 people and pulled work history for 18 to deliver a shortlist of ten. Read those ten, correct the brief if needed, then ask for more.

### What does a credit buy in a shortlist run?

In the run on 10 October 2026, 25.85 credits bought four searches, work history for 18 people and seven personal emails. A personal email cost between 1 and 3.8 credits depending on which provider found it. Lookups that found nothing were not charged in this run; [credits and billing](/docs/credits-and-billing) lists the calls that are charged per request.

### Can Claude score candidates I already have?

Yes. Claude can score any profiles or CVs you paste in against the job description, with or without a connector. What Claude cannot do alone is find new people or look up a personal email. [How to source candidates with Claude](/blog/how-to-source-candidates-with-claude) covers that difference.

### Can I use my own data-provider account?

Yes. You can connect your own data-provider accounts from the Essential plan, and those calls do not use hyreflow credits.

## Sources

- [Resourcing and talent planning report 2026](https://www.cipd.org/globalassets/media/knowledge/knowledge-hub/reports/2026-pdfs/9218-resource-and-talent-planning-2026-report-web.pdf), CIPD. Checked 12 Oct 2026
- [An AI trust crisis: 2025 AI in Hiring Report](https://www.greenhouse.com/newsroom/an-ai-trust-crisis-70-of-hiring-managers-trust-ai-to-make-faster-and-better-hiring-decisions-only-8-of-job-seekers-call-it-fair), Greenhouse. Checked 12 Oct 2026
- [Get started with custom connectors using remote MCP](https://support.claude.com/en/articles/11175166-get-started-with-custom-connectors-using-remote-mcp), Anthropic. Checked 12 Oct 2026
- [Playbook: turn a job spec into a ranked shortlist with contact details](https://www.hyreflow.ai/skills/job-spec-to-shortlist), hyreflow. Checked 10 Oct 2026
- [Credits and billing](https://www.hyreflow.ai/docs/credits-and-billing), hyreflow docs. Checked 10 Oct 2026
- [hyreflow pricing](https://www.hyreflow.ai/pricing), hyreflow. Checked 10 Oct 2026