Prompting Toolkit

A practical approach to writing clear, effective prompts using a repeatable three\-step framework that works across Ashby's AI features and any AI tool, along with ready\-to\-use prompt examples for common recruiting tasks\.

How to use this Playbook

Use this playbook as a step\-by\-step guide to prompting more effectively, in Ashby and in any AI tool you already use\.

Follow the lessons in order, or click directly into a section using the navigation panel on the left\. Each step in the framework builds on the last, and the final section brings them together with examples you can use today\.

Who this playbook is for

Recruiters and Sourcers who want faster, more consistent candidate summaries, outreach, and interview prep

Talent acquisition leaders who want repeatable pipeline reviews and reporting they can trust enough to share

Recruiting operations and Ashby Admins who set up the criteria, scorecards, and workflows that other people depend on

Anyone building custom agents in Ashby who wants to sharpen a use case, explore what agents can do, and get them acting autonomously on their behalf

What you'll learn

Define Your Outcome What to decide before you write anything

Design Your Output *Components that shape what comes back
Test and Iterate How to tell whether it worked, and what to change
Ready\-to\-Use Prompts Ready\-to\-use prompts you can adapt today

Card Front

What are prompts?

Card Back

Image

A prompt is the instruction you give an AI tool that tells it what to do and what matters\.

Well\-written prompts lead to clearer, more relevant, and more consistent results\.

Card Front

Where are prompts used in Ashby?

Card Back

In Ashby, prompting is used across several product areas to help generate content, analyze recruiting data, support decision\-making, and connect Ashby workflows with external tools\.

These areas include:

Ashby Agents
AI\-Assisted Application Review
AI auto\-fill in AI Notetaker
Ashby MCP server
AI\-Assisted Report Builder

Before you dive in

The prompting toolkit covers general prompting principles and best practices to help you achieve more useful, consistent results from AI tools\.

While some examples reference specific Ashby AI features, the guidance is largely feature\-agnostic and is intended to apply across a range of prompting contexts\.

Flip through each card to understand key prompting concepts\.

Click each option to learn more about how prompting can be used with Ashby AI features\.

Ashby Agents \(Ashby Assistant\)

Can see & act: Ashby recruiting data available to you, and perform actions on your behalf within your permissions scope\.

Best for: One\-off questions and drafts, such as pipeline checks, summaries, and updates you’ll review and send\.
Ashby Agents \(Custom Agents\)
Can see: Ashby recruiting data that the current user is permitted to access, based on their existing permissions\. Best for: A task you run repeatedly, or one your team triggers in Slack\.
AI\-Assisted Application Review
Can see: The candidate’s resume and, optionally, whether they answered selected long\-form application questions, evaluated against criteria you define\. Best for: tructuring and prioritizing high\-volume review against explicit, recruiter\-defined criteria\.
AI auto\-fill in AI Notetaker
Can see: The transcript from a recorded interview you have access to, plus the interview briefing instructions, job description, and any custom auto\-fill instructions your org has set\. Best for: Turning what was said in an interview into structured feedback\.
Ashby MCP Server
Can see: Live Ashby data that you can already access through a connected MCP\-compatible AI tool\. Best for: Combining Ashby data with context that lives outside Ashby\.
💡 Think of a prompt like a recipe\.

The ingredients \(your inputs\) and the instructions \(your steps\) stay the same each time you use it\. However, the dish that comes out can still vary a little depending on the day, the oven, or the cook\. A good prompt doesn't guarantee an identical result every time, but it narrows the range of outcomes and makes them more predictable\.

Continue →

Step 1: Define Your Outcome

Define Your Outcome

Step 1

Before writing your prompt, decide what outcome you want\. Being specific helps the AI tool deliver a stronger and more targeted response\.

Start With the Outcome

Plan Your Prompt

If prompts are like recipes, then context is your ingredients – the background you already have in your head, but that's often missing from the prompt itself\.

When paired with your instructions, context enables the AI tool to produce better, more targeted results\.
Answer these questions as you plan your prompt:
Goal & Outcome: What are you actually trying to accomplish? Name the outcome, not just the task in front of you\.

Success Criteria: What would a good result actually look like? Be specific enough that you'd know it if you saw it\.

Failed Output: What would a bad or unusable result look like? Naming this up front helps you spot it fast\.

Prompting Design Examples

Expand each example where we've identified the outcome, success criteria, and failed output to build the prompt\.

Weekly Hiring Summary

Goal & Outcome Success Criteria Failed Output
You want to produce a weekly hiring update for a Hiring Manager who does not have access to Ashby\. The weekly hiring update would include the current pipeline for all active candidates across the open roles they're working with you on, including: - Active candidate count - Candidates in later stages - Candidates moved forward or rejected this week - Roles where the pipeline has stalled A hiring update that points to Ashby, or uses ATS and recruiting jargon they're unfamiliar with\.

Stronger Prompt

Draft a weekly hiring update for a hiring manager who doesn't have access to Ashby\.

Pull the current pipeline for every active candidate on the roles we're working on together — how many are active, who's furthest along, who moved forward or was rejected this week, and whether a role has stalled\.

Write it as a Slack message under 250 words: a one\-line headline on overall momentum, then 2–3 plain\-English sentences per role, folding any stall into that role's own paragraph, then one “What I need from you this week” section — say “nothing this week” if that's true — then a warm closing line\.

No Ashby or ATS terminology\. Sound like a recruiter giving a verbal update, not a system generating a report\.

Ask me whether to send it or hand it back for me to message the hiring manager\.

Vague Prompt

Run a weekly hiring update for a hiring manager who I'm working with on various roles\.

Prioritized To\-Do List

Goal & Outcome Success Criteria Failed Output
You want a prioritized list of what needs attention across your pipeline today, without reviewing every job or candidate yourself\. The list would include, for active roles you own: - candidates awaiting feedback - applications stalled 5\+ calendar days with no activity - feedback overdue past the SLA your team sets - interviews in the next 7 days with missing confirmations A list that includes closed or inactive roles, buries urgent items under routine ones, or pads action items with background context instead of just what needs doing\.

Stronger Prompt

Review my full pipeline and give me a prioritized to\-do list of anything that needs attention before the end of the day today\.

Include:

candidates awaiting feedback

applications sitting in the current stage for more than \[number\] days with no activity

my outstanding emails or tasks

offers with no update this week

upcoming interviews early next week where scheduling or confirmations look incomplete\.

Format your response as a simple numbered to\-do list, grouped by job\. Keep it brief — just what I need to action, not background context\.

Vague Prompt

Give me a prioritized to\-do list of anything that needs attention before end of day today\.

Test your knowledge

You want to re\-engage strong candidates from a past hiring cycle who were never actually rejected – just went stale\.

Which prompt best states the goal and outcome you're after?
Single Choice Question
Options:

[ ] 'Look through my old candidate pool and see who's still worth reaching out to\.'

[ ] 'Format a list of candidates from the Q1 sourcing pool as a table with name, role, and last contact date\.'

[x] 'Review candidates in the Q1 \[job title\] pool with no activity in 90\+ days who were never explicitly rejected, and flag those whose original screening notes show a specific skill match or an explicit recommendation to move forward — these are the ones worth a re\-engagement outreach\.'

[ ] 'Send a re\-engagement email to candidates in my Q1 pool\.'

Submit

Continue →

← Return

Step 2: Design Your Output

Design Your Output

Step 2

With the outcome and any additional context identified, you're ready to design your output\. Your output takes the task your prompt outlines one step further by defining the format, level of detail, evidence, and shape of the result\.

The clearer you are about how to present the information you want, the less back\-and\-forth it takes to get there\.

Determine the Shape

Structure & Format

Pick the structure and format that matches how it'll be used:

Specify the shape: bullets, table, or a plain paragraph

Set a length or word limit

Give a heading or section order if the output has multiple parts

State how the items should be ordered – by urgency, by stage, by date, or by strength of evidence

For example:
Draft a status update\.

VS\.

Draft a status update as three short bullet points — one for progress, one for blockers, one for next steps — under 50 words total\.

Constraints

Include any constraints:

"Always do\.\.\." \(e\.g\., always cite the source record\)

"Never do\.\.\." \(e\.g\., never share compensation details\)

"Only look at… / Ignore…" to narrow scope

"Confirm with me before…" \(e\.g\., before sending a message or making an update\)

For example:
Draft interview questions\.

VS\.

Generate 4–6 behavioral or situational interview questions focused on the highest\-priority untested areas\. Include a brief rationale for each question and 1–2 follow\-up probes for the top questions\.

Build on areas with strong, consistent evidence rather than repeating them, unless a material concern warrants additional testing\.

Do not reference protected characteristics or make a hiring recommendation\.

Examples

Include any examples:

Show the exact tone, structure, and length you want

Include diverse examples, especially edge cases, not just the easy ones

Format examples as input → output \(e\.g\., a sample request and the ideal response\)

Pair a "do this" example with a "not this" example when the distinction is subtle

For example:
Write feedback in a professional tone

VS\.

Write feedback in this tone:

Example: 'Candidate showed strong ownership in the walkthrough, but struggled to explain trade\-offs when pushed\.'

Match that style: specific, evidence\-based, no vague adjectives like 'good communicator'\.

Ambiguity & Assumptions

Determine how to handle ambiguity and unstated assumptions:

Say what NOT to infer or assume beyond what's explicitly stated

Define ambiguous terms in the criteria itself – don't assume shared understanding \(e\.g\., shorthand you'd use with a colleague\)

Tell the AI tool what to do when information is missing or unclear \(e\.g\., flag it, don't guess\)

Call out edge cases you already know are tricky

For example:
Prioritize candidates with a Big Four or top\-tier consulting background\.

VS\.

Prioritize candidates who have worked at one of: McKinsey, Bain, BCG, Deloitte, EY, KPMG, or PwC\.

Do not include candidates from other consulting firms unless they also meet \[criterion\]\.

If a firm is ambiguous, list it separately rather than deciding for me\.

Step-by-Step Reasoning

For complex tasks, tell the AI tool how to reason:

Ask the model to cite the specific evidence and criteria behind each conclusion, not just the final answer

Break a complex ask into smaller steps when it involves multiple judgments \("first identify X, then compare against Y"\)

Name what evidence or criteria to weigh, and in what order, when a judgment call is required

For example:
Decide if this candidate is a strong fit

VS\.

First, list the specific requirements from the job description\.

Then, for each one, note whether the resume provides explicit evidence\.

Only after that, give an overall fit rating — and show your reasoning\.

Sources & Failures

If your prompt uses a tool or agent, include guidance on tool use and what happens when they fail:

What "good" tool use looks like \(e\.g\., search before answering, don't guess a file path\)

Failure handling: what to do when a tool errors or returns nothing

For example:
List my upcoming debriefs in the next 14 days\.

VS\.

Using the \[debrief agent\], list my upcoming debriefs in the next 14 days\.

For each, include job, date, candidate, and whether the debrief has been scheduled or is awaiting confirmation\.

Do not take an action; just return the list\.

If the agent can't find any debriefs or the lookup fails, tell me directly rather than assuming there are none\.

Components That Shape Output

Define the desired format, add relevant constraints and examples, clarify any assumptions, and identify the sources or criteria that should shape the response\.

Click each tab to review the components that contribute to a stronger, more accurate output\.

💡 Advanced Prompting

These components cover most day\-to\-day prompting\. If you're working on something higher\-stakes, dive into more advanced strategies like:

Break it down first: Ask the AI tool to list the sub\-questions it needs to answer before it answers the main one, then work through them\.

Ask more than once: Run the same question a few different ways and compare\. Where the answers agree, you have somewhere to start\. Where they disagree, you've found the part that needs your judgment\.

Prompting Design Examples

Expand each example where we've identified the outcome and output design to build the prompt\.

Weekly Hiring Summary

Goal & Outcome Success Criteria Failed Output Output Design
You want to produce a weekly hiring update for a Hiring Manager who does not have access to Ashby\. The weekly hiring update would include the current pipeline for all active candidates across the open roles they're working with you on, including: - Active candidate count - Candidates in later stages - Candidates moved forward or rejected this week - Roles where the pipeline has stalled A hiring update that points to Ashby, or uses ATS and recruiting jargon they're unfamiliar with\. A concise, professional Slack message structured as a one\-line headline on overall momentum\. Includes: - A 2–3 sentence plain\-English update per role - A single "What I need from you this week" section, stated plainly if there's nothing pending - A warm closing line Under 250 words\. Offers the choice to return the draft to copy into Slack, or send it directly — with review either way\.

Stronger Prompt

Draft a weekly hiring update for a hiring manager who doesn't have access to Ashby\.

Pull the current pipeline for every active candidate on the roles we're working on together — how many are active, who's furthest along, who moved forward or was rejected this week, and whether a role has stalled\.

Write it as a Slack message under 250 words: a one\-line headline on overall momentum, then 2–3 plain\-English sentences per role, folding any stall into that role's own paragraph, then one “What I need from you this week” section — say “nothing this week” if that's true — then a warm closing line\.

No Ashby or ATS terminology\. Sound like a recruiter giving a verbal update, not a system generating a report\.

Ask me whether to send it or hand it back for me to message the hiring manager\.

Vague Prompt

Run a weekly hiring update for a hiring manager who I'm working with on various roles\.

Prioritized To\-Do List

Goal & Outcome Success Criteria Failed Output Output Design
You want a prioritized list of what needs attention across your pipeline today, without reviewing every job or candidate yourself\. The list would include, for active roles you own: - candidates awaiting feedback - applications stalled 5\+ calendar days with no activity - feedback overdue past the SLA your team sets - interviews in the next 7 days with missing confirmations A list that includes closed or inactive roles, buries urgent items under routine ones, or pads action items with background context instead of just what needs doing\. A numbered to\-do list: - grouped by job - ordered by urgency - capped at 5 actions per job, with a direct link or record reference where available Action items only, no background explanation\.

Stronger Prompt

Review my full pipeline and give me a prioritized to\-do list of anything that needs attention before the end of the day today\.

Include:

candidates awaiting feedback

applications sitting in the current stage for more than \[number\] days with no activity

my outstanding emails or tasks

offers with no update this week

upcoming interviews early next week where scheduling or confirmations look incomplete\.

Format your response as a simple numbered to\-do list, grouped by job\. Keep it brief — just what I need to action, not background context\.

Vague Prompt

Give me a prioritized to\-do list of anything that needs attention before end of day today\.

Test your knowledge

Match each prompt segment to the component it represents in shaping the output design\.
Match the following:
Left side Right side
"If evidence is unclear, flag it rather than coming up with a response\." Ambiguity & Assumptions
"Write it as a Slack message under 250 words: a one\-line headline\.\.\. then a warm closing line\." Structure & Format
"Action items only, no background explanation\." Constraints
"Match this tone – Example: 'Candidate showed strong ownership in the walkthrough, but struggled to explain trade\-offs when pushed\.'" Examples
"First list the requirements, then note whether the resume gives evidence for each – only after that, give a fit rating\." Step\-by\-Step Reasoning
"Using the \[named agent\], pull my upcoming debriefs\. If the lookup fails, tell me it failed – don't report an empty week\." Sources & Failures

Submit

Continue →

← Return

Step 3: Test and Iterate

Test and Iterate

Step 3

Prompts don't always produce the intended outcome the first time\. Even a well\-scoped prompt often needs a second pass, and that's the process working as intended, not a sign you got something wrong\.

It's the same with a recipe: no dish comes out perfect on the first taste – you adjust the seasoning and try again\.

Testing Is Part of the Process

What to Check

Unclear goals, missing context, weak structure, wrong mechanism, or ignored constraints all contribute to a prompt underperforming\.

Test with more than one real input, including a tricky or unusual case\.
Watch for two different kinds of misses:
When a prompt underperforms, check in this order:

Is the goal actually clear?

Is the data actually there, and can this feature see it?

Is there missing context or an unclear scope?

Is the structure or format specific enough \(e\.g\., shape, headings, word limit\)?

Are constraints being ignored \(e\.g\., a "never do," a boundary, a confirm\-before\-acting rule\)?

Is this even the right mechanism \(e\.g\., a prompt, or a feature or workflow you'd set up once\)?

Something that should have been caught and wasn't

Something flagged that shouldn't have been

💡Ashby agents only act on what you explicitly ask for, while tools like Claude or Glean proactively infer your intent and suggest next steps\.
AI Tools & Hallucinations

Outputs from AI tools can be wrong in ways that still read as confident, often referred to as "hallucinations":

Detail that isn’t in the record: It can produce a specific\-sounding fact like a date, a title, or a number that isn’t actually there\.

Inference from adjacent information: A “full stack engineering” title doesn’t tell you which languages someone knows, but the AI tool may treat it as if it does\.

Stale or partial data: If a stage wasn’t updated or feedback was never submitted, the AI tool could work around the gap rather than flagging it\.

An empty result that reads like an answer: “Nothing needs attention this week” can mean nothing needs attention, or that the lookup returned nothing\.

If you suspect the AI tool has produced a hallucination, ask where the answer came from, including which record, resume line, or piece of feedback each conclusion is based on, then spot\-check two against the source\.
💡 Revise One Thing At a Time
For example, if you change three instructions at once and the output improves, you won't know which change actually mattered, which makes it harder to reuse what worked the next time you write a prompt\.

Testing and Iterating Prompt Example

Expand the example to see how one change turned a vague prompt into a more reliable one\.

Summarizing Interview Feedback with Evidence

Version 1

Summarize this candidate’s interview feedback concisely and with evidence\.

Summarize this candidate’s interview feedback in this style –
Example: ‘Handled the system design prompt confidently; asked clarifying questions before diving in\.’

Match that: one sentence, grounded in a specific moment, no general praise\.

What Came Back:

Two sentences of general praise, with nothing tied to anything that happened in the interview:

“Strong communicator and a good culture fit\. The team felt he would ramp quickly and work well across functions\.”
Version 2
What Came Back:

One sentence, tied to a specific moment, in the same shape as the example:

“Walked through the trade\-offs in his caching design unprompted, then flagged the one assumption he was least sure about\.”

Why It Worked:

“Concise and evidence\-based” was a description the AI tooI had to interpret\. Adding the example helped the AI tool more clearly understand what kind of summary to produce\.

Before your draft

Stuck on how to even start? Hand the AI tool your current attempt and ask what's unclear or missing\.

Rewrite based on that feedback before you try the real task\.

After you get an answer

Once you have a draft answer, ask the AI tool to review its own response and point out what's weak or missing before you accept it\.
Note: Use this once or twice per task\. Repeating the request may produce repetitive feedback, unnecessary revisions, or confusion about which version of the prompt to use\.
Ask the AI Tool What's Missing

The AI tool can provide help before you've written anything, and after you've got the first answer\.
💡Save and Re\-Use Strong Prompts
If you produce a strong prompt, save the version that worked somewhere your team can find it\.

For a task you’ll run every week, save the prompt that worked, note what you changed, and re\-check it when your process changes\.

Test your knowledge

You want the AI to flag interview\-scheduling conflicts, but it should never reschedule anything without asking first\. Which prompt sets the clearest boundary?
Single Choice Question
Options:

[ ] Check my calendar for interview conflicts and fix them\.

[ ] Check my calendar for interview conflicts and let me know what you find\.

[x] Check my calendar for interview conflicts\. Never reschedule anything — flag conflicts and ask me before making any change\.

Submit

Test your knowledge

You want the AI to provide candidate summaries that all follow the same tight, evidence\-based style\. Describing it as "concise and evidence\-based" in the prompt hasn't worked reliably\. What's the stronger prompt?
Single Choice Question
Options:

[x] Summarize this candidate's feedback in this style — Example: 'Handled the system design prompt confidently; asked clarifying questions before diving in\.' Match that: one sentence, grounded in a specific moment, no general praise\.

[ ] Summarize this candidate's interview feedback in 1–2 sentences\.

[ ] Summarize this candidate's interview feedback concisely and with evidence\.

Submit

Continue →

← Return

Ready-to-Use Prompts

Ready\-to\-Use Prompts

Use the three\-step framework as you build your prompting skills\.

Apply the Prompting Framework

Define your outcome

Design your output

Test and iterate

Ready\-to\-Use Prompts

Each prompt lists a recommended Ashby AI feature, but you can customize and use it in the AI tool of your choice\.

As a best practice, always review AI\-generated outputs for accuracy, relevance, and compliance with your organization’s policies before relying on or sharing them\.

Click into each option to review the goal, recommended use, and copy\-read prompt\.

Prioritized To\-Do List
Goal: Populate a prioritized to\-do list based on tasks in your candidate pipeline\.
Recommended Use:

At the start of the day to see what needs action across every role you own\.

Ashby Agents or Ashby MCP

Customize the days overdue and actions per job based on your own internal SLAs and preferences

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

Review my full pipeline and give me a prioritized to-do list of anything that needs attention before the end of the day today. Include: -candidates awaiting feedback -applications sitting in the current stage for more than [number] days with no activity -my outstanding emails or tasks -offers with no update this week -upcoming interviews early next week where scheduling or confirmations look incomplete Format your response as a simple numbered to-do list, grouped by job. Keep it brief — just what I need to action, not background context.

Weekly Hiring Update \(Hiring Manager\)
Goal: Produce a weekly hiring update for a hiring manager who doesn't have access to Ashby\.
Recommended Use:

Use this for a recurring update to someone without Ashby access\.

Ashby Agents or Ashby MCP

Customize the hiring manager and structure and format based on your own preferences

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

I need to send a weekly hiring update to a hiring manager who does not have access to Ashby. Pull current pipeline status for all active candidates across our open roles. For each role, include: -active candidate count -furthest-progressed candidate -candidates moved forward or rejected this week -roles where the pipeline has stalled Then draft a concise, professional Slack message with the following structure: A one-line headline summarizing overall hiring momentum this week (e.g. "Good progress across the board" or "A couple of roles need attention") A short update per role in plain English. No ATS jargon. Write it as if you are a recruiter giving a verbal update, not a system generating a report. Each role should be 2–3 sentences maximum. Work any stalled-pipeline notes into that role's own paragraph — don't create a separate list for them. Example: "Senior Backend Engineer: Good momentum — two candidates moved to onsite this week, and we have an offer out with a decision expected by Friday. No action needed from you yet." If a role has stalled, the same paragraph style covers it: "Product Marketing Manager: This one's stalled — no strong applicants in about two weeks. Worth 15 minutes to revisit sourcing if you have time." A single "What I need from you this week" section listing any actions the hiring manager needs to take: interview feedback outstanding, upcoming interviews to prepare for, or decisions pending. If there's nothing pending, say so briefly (e.g., "Nothing needed from you this week") rather than leaving the section blank or padding it out. A brief closing line keeping the tone warm and collaborative. Keep the whole message under 250 words. The hiring manager is busy, non-technical, and does not care about pipeline stages or ATS terminology. Write it in plain English and make it feel human, not automated. Ask me whether to return the message here for me to copy into Slack myself, or send it directly to the hiring manager on my behalf. Either way, I'll review it before sending.

Weekly Hiring Update \(Executive\)
Goal: Produce a weekly hiring update for an Executive who does have access to Ashby\.
Recommended Use:

Use this for a recurring update to an Executive with Ashby access\.

Ashby Agents or Ashby MCP

Customize the Executive, department, job title, or team, and number of days based on your own preferences

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

I need to send a weekly hiring update to [executive] for [department/org]. They have access to Ashby but don't work in it day-to-day. Pull current hiring activity for [job title(s)/team] using active applications updated in the last [number] days. For this period, include: -number of open roles -candidates in final stages -hires or offers made this period -any roles with a significant risk or delay (not routine day-to-day movement) Then draft a concise, executive-ready Slack message with the following structure: A one-line headline on overall hiring velocity this period (e.g. "Hiring is on pace across the team" or "A couple of roles need attention") A short velocity summary in plain English. No ATS jargon. Write it as if you are a recruiting leader briefing an executive, not a system generating a report. Keep this to one short paragraph covering the whole team or org — not a role-by-role rundown. Fold any risks or delays into this same paragraph, unless there are more than two — then a short bulleted callout is fine. If there's nothing worth flagging, say so briefly rather than manufacturing a concern. Example: "Engineering and Product hiring stayed on pace this week — 3 roles closed, 12 candidates in final stages. One flag: the Staff Data Scientist search has stalled for three weeks and may need revised sourcing." A single "Where to look deeper" section with a link to the Executive Dashboard in Ashby, for anyone who wants to review specifics themselves. If nothing needs deeper attention this period, keep this line short rather than padding it. A brief closing line keeping the tone confident and collaborative. Keep the whole message under 150 words. [executive] is busy, doesn't work in the ATS day-to-day, and doesn't need individual candidate detail — just the headline and somewhere to go for more. Write it in plain English and make it feel like a real update, not an automated report. Ask me whether to return the message here for me to copy into Slack myself, or send it directly to the executive on my behalf. Either way, I'll review it before sending.

Targeted Interview Questions
Goal: Produce interview questions tailored to a candidate, job, and interview stage based on job details\.
Recommended Use:

Use this before a later\-stage interview to test what earlier rounds didn’t cover\.

Ashby Agents or Ashby MCP

Customize the candidate, job, and interview stage, as well as the number of questions and follow\-ups based on your own preferences

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

For [candidate] interviewing for [job title] at the [interview stage] stage, review the job description, interview plan, and prior feedback. Identify: competencies already validated concerns or gaps 3.competencies not yet assessed. Generate 8–12 behavioral or situational questions focused only on gaps and untested areas, with a one-sentence rationale and 1–2 follow-up probes for the highest-priority questions. Do not reference protected characteristics or repeat areas with strong, consistent positive evidence.

Debrief Lookup
Goal: Produce a summary of upcoming debriefs in a specific time period that have been scheduled or need to be scheduled\.
Recommended Use:

Use this to see which debriefs are coming up\.

Ashby Agents or Ashby MCP

Customize the count of upcoming days and structure and format based on your own preferences

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

List my upcoming debriefs in the next 14 days. For each, include job, date, candidate, and whether the debrief has been scheduled or is awaiting confirmation. Do not take an action; just return the list.

Referral Checker
Goal: Build an Ashby Agent to map existing candidates to existing employees to identify employees who may know the candidate\.
Recommended Use:

Use this on a sourced candidate before outreach to find someone internal who may know them\.

Ashby Agents

Customize the structure and format based on your preferences

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

You are an expert recruiting referral-mapping agent for [Company name]. Your task is to compare a sourced candidate’s resume/work history against [company name] current employees and hire history to identify employees who may know the candidate or have relevant overlap for outreach. For each candidate, analyze: -Prior companies -Job titles/functions -Employment dates, if available -Teams, departments, or product areas, if available -Education or network overlap, only if clearly relevant Return only likely referral connections where there is meaningful overlap. For each match, include: -Employee name -Candidate company overlap -Employee’s relevant company history -Why they may know the candidate -Strength of match: High / Medium / Low -Recommended outreach angle Prioritize: -Same company during overlapping dates -Same or adjacent function/team -Similar product area, customer segment, or department -Close timing at the same company, even if exact dates do not overlap. Do not use location as a matching factor. Do not assume people know each other solely because they worked at the same large company years apart. Flag uncertain matches clearly. Output format: -Candidate: [Name] -Potential [Company name] -Referral Matches: [Employee Name] -Overlap: [Company / team / dates] -Match strength: [High/Medium/Low] -Reasoning: [Brief explanation] -Suggested outreach note: [1-2 sentence angle]

Feedback Formatting
Goal: Determine the format of feedback in AI Notetaker Auto\-Fill\.
Recommended Use:

Use this to make AI Notetaker Auto\-Fill consistent across your team\.

AI auto\-fill \(AI Notetaker\)

Customize the structure and format based on your preferences

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

Return no more than five bullets. Start each bullet with the evidence or observation, followed by the implication for this interview question. Do not repeat the candidate's name, use paragraphs, or include a hiring recommendation. Keep the total response under 100 words unless the question requires more detail.

Pipeline Health Analysis
Goal: Analyze pipeline health for a specific job\(s\) associated with a specific hiring manager\.
Recommended Use:

Use this when a hiring manager asks, “How’s the search going?” and you want a quick answer\.

Ashby Agents or Ashby MCP

Tailor to specific job title\(s\) and hiring manager\(s\) based on your preferences

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

Analyze pipeline health for the open [job title] role under [hiring manager]. Review active candidates only and report current stage, days since last activity, feedback gaps, stalled candidates, and upcoming actions. Return a short verdict—on track, at risk, or stalled—followed by the evidence and the top three recommended next actions. Do not evaluate candidate quality or make hiring recommendations.

Interview Questions & Responses
Goal: Summarize candidate's response based on interviewer questions\.
Recommended Use:

Use this when you want to review the interview question by question to see how the candidate responded, identify supporting evidence, and flag any unanswered or unclear questions\.

AI Notetaker → Interview Recording → AI Assistant

✍️ Try it out

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

Using only evidence from the interview transcript, identify each question the interviewer asked and summarize the candidate’s response. For each response, include: -Question: A concise version of the interviewer’s question -Response: A concise summary of the candidate’s answer -Evidence: A specific example or direct quote from the transcript -Open questions: Any unanswered question, ambiguity, or useful area to explore further Do not infer information, evaluate the candidate, or make a hiring recommendation. If the transcript does not provide enough evidence, state “Insufficient evidence.”

Custom Job AI Criteria
Goal: Generate custom criteria to add to AI\-Assisted Application Review Criteria\.
Recommended Use:

Use this when the AI\-suggested criteria based on the job description are too broad or are not surfacing the candidates your team is looking for\.

Ashby Agents or Ashby MCP

Tailor to a specific job title and skills based on your preferences

✍️ Try it out
*

Click the copy icon to copy the prompt, then adjust it to suit your workflow\.

For the [job title] job in Ashby, review the job description.

The job description is intentionally broad to support candidate reach. 


Based on the additional hiring context below, generate 3–5 AI job criteria that will help identify candidates with the specific experience the team wants to prioritize.


Additional hiring context:
[Describe the specific experience, skill, or qualification you want to identify. For example: The team wants candidates who have directly led implementations for B2B SaaS customers, rather than candidates with general customer success or account management experience.]


For each criterion, return:
-Short title: a concise name for the criterion
-Criterion prompt: one standalone requirement for the AI to evaluate
-Include the resume evidence that should count and the clearly adjacent experience that should not count.


Keep each criterion focused on one requirement. Use neutral, objective language that a junior recruiter could apply consistently. 


Make each criterion specific and verifiable from a resume. Do not infer experience that is not stated. Do not use protected characteristics, rank candidates against one another, or make a hiring recommendation. Do not rewrite the job description or make changes in Ashby.


Format the recommendations so a recruiter can add the selected Short title and Criterion prompt as custom AI job criteria in Ashby.

Back to All Playbooks

← Return

Left side	Right side
"If evidence is unclear, flag it rather than coming up with a response\."	Ambiguity & Assumptions
"Write it as a Slack message under 250 words: a one\-line headline\.\.\. then a warm closing line\."	Structure & Format
"Action items only, no background explanation\."	Constraints
"Match this tone – Example: 'Candidate showed strong ownership in the walkthrough, but struggled to explain trade\-offs when pushed\.'"	Examples
"First list the requirements, then note whether the resume gives evidence for each – only after that, give a fit rating\."	Step\-by\-Step Reasoning
"Using the \[named agent\], pull my upcoming debriefs\. If the lookup fails, tell me it failed – don't report an empty week\."	Sources & Failures