Finding high-quality candidates for your open roles is often the most time-consuming and difficult part of the recruiting process. Gem's AI Sourcing function can help expand your sourcing reach using our AI-powered solution to match potential candidates from our database of 800M+ public profiles with your specific job requirements.

In this article, we’ll take you through basic steps to get up and running with AI Sourcing, including:

• Transforming your typical sourcing flow to an AI mindset
• Setting up an AI Sourcing search
• Tips for AI Sourcing searches

Requirements

Gem AI Sourcing must be enabled. If you don't have this feature but are interested in adding it, please contact your Gem Customer Success Manager.

Get into the AI mindset

AI can be a powerful tool in your recruiting workflow, but it requires that you shift your mindset from your traditional sourcing motions. In this section, we'll cover best practices for shifting your approach away from traditional, non-AI search methods to develop a truly AI-first sourcing practice.

Write prompts instead of keyword filters

Before AI, when searching for candidates, you'd filter by keywords and operations (AND, OR, NOT) to limit your results to only candidates that are a literal match for those terms.

For example, you might filter for ("software engineer" OR "programmer" OR "developer") AND (“fullstack” OR “FS”) AND (“tech” OR “startup”) AND ( java OR python) AND NOT ( junior OR entry-level).

This finds only candidates that have this specific combination of title, skills, and experience from your pool.

With AI, you write a detailed description of your ideal candidate in natural language. For example, you could prompt the AI with “Find me software engineers with 3−4 years of full-stack experience and work experience at early-stage tech startups. Ideally, they have Java or Python experience.”

The AI tool will find all the same candidates as your filtered search, plus additional candidates that use other or no titles, list skillsets that are comparable to a full-stack developer but don't describe themselves as such, or who have worked at companies that fit your desired profile, without having to explicitly say they're in tech or consider themselves to be startups.

Given a prompt and a set of ranking criteria, AI Sourcing finds candidates that match your search criteria and extrapolates to surface candidates that your original prompt might have excluded.

Use punctuation to clearly separate thoughts

The best way to think of what a prompt should include is to consider how you'd brief a new colleague on the candidate you're looking for. If you hired a new recruiter to lead this job search, you'd use natural language to describe your ideal candidate fully. All of the same thought and crafting that would go into that debriefing, or that you'd include in your internal job specifications and external job postings, can be used to enrich your AI searches.

It's important to remember that AI understands and depends on punctuation. As you build more complex prompts, use commas to create groups of criteria and periods to separate your thoughts in the same way you would pause in a conversation.

Using our previous example, if you removed the period and submitted “find me software engineers with 3−4 years of full-stack experience and work experience at early-stage tech startups that have Java or Python experience”, the AI wouldn't be able to tell if the Java experience was a criterion for the company or the candidate.

Be detailed and specific

Avoid broad prompts like "find software engineers." Just like humans, AI struggles with interpreting vague, broad requests. If you were describing your ideal candidate to a teammate, this wouldn't give them the level of detail they'd need to narrow down their search.

Instead, add context to the prompt, just like you would for your new recruiter. Include desired traits or skills like:

• The ideal candidate's must-have skills
• The size or industry of their previous employers
• What culture or value attributes would lead someone to be successful in the role

For example, "we’re hiring for a Series C B2B SaaS startup building AI infrastructure. We need engineers familiar with distributed systems. Find senior backend engineers with infra experience at B2B SaaS campaigns, especially those who've worked in scaling systems for high-growth startups."

Understand how AI interprets requests

It's helpful to understand what the AI is doing when it's searching for results that match your prompt.

If you write a simple prompt like "find me a good software engineer", the model has to decide what "good" means. This isn't something candidates will have in their profile, so the model needs to review the pool and decide what differentiating factor might qualify a candidate as "good".

It might decide that "good" engineers have long tenure and prioritize candidates with long work histories. It might also decide "good" means candidates who've worked at large companies, and only prioritize candidates from Fortune 500 companies, even if the candidate only interned there briefly.

The AI has no problem finding "software engineers". However, without more guidance on what skills you specifically want, it has no way to prioritize the available pool other than by weighting candidates who describe themselves as "software engineer" more heavily than those who describe themselves as "developers". Though the AI recognizes that "software engineer" and "developer" are often the same, it weighs your prompt content more heavily than the values it can infer are related.

The "find me a good software engineer" prompt could result in a COBOL engineer with 25 years as a mainframe engineer for Microsoft as the top-ranked candidate, when you're really looking for a full-stack developer with SaaS-based AI experience.

Bad prompt: "Help me find good candidates from top companies for a machine learning role."

Good prompt: "I need a Senior ML Engineer for our autonomous vehicle startup (Series B, 100-300 employees) building perception systems for self-driving trucks. Our ideal candidate has experience with: computer vision in production environments, sensor fusion (camera + LiDAR + radar), real-time inference at edge, and automotive safety standards (ISO 26262). They should come from companies that have deployed ML in safety-critical applications."

Tell the AI what to avoid

What you don't want in a candidate is just as important to specify as the criteria you're looking for. Similar to how you might apply negative filters to a traditional search, include the criteria you want the AI to avoid in your prompt.

For example, you can ask the AI to skip candidates from specific industries ("exclude fintech companies"), candidates who have only worked in academia ("avoid candidates with only academic or research experience"), or phrases in their profile that don't align with your corporate values ("don't use buzzwords like 'rockstar' or 'ninja'").

Keep practicing: Adjust and re-run your search

AI prompts benefit from iteration. After running your first search, try reviewing the first 25-50 candidates in your results to look for patterns in work experience or titles. If you're not getting the results you wanted, or if you didn't see something you expected to see (or the opposite) try adjusting your prompt. Better yet, ask the AI to run a new search for you - you can ask the AI itself to help you craft the perfect prompt. Even small changes to a prompt can greatly impact your results.

Set up an AI Sourcing search

To start a new AI Sourcing search, complete these steps.

1. Select AI Sourcing under the Outreach tab.
2. Select the + New Search button. This is located in the center of the page, under 'Welcome to Gem AI Sourcing,' and in the AI Sourcing section at the top, to the right of the count of AI search bots used.

A pop-up appears, guiding you through a 4-step workflow to set up your search. Each step of this workflow is covered in the sections below.

Step 1. Set up your first search

Fill out these three initial fields to kick off your search:

• What job are you sourcing for?: Select the position you’re sourcing for (Gem will only show active jobs in your ATS)
• Job description: If the job you have selected does not already have a job description, you will then give Gem a description of the role. If you have already created a role intake doc or other job description outside of Gem, you can paste it here. Otherwise, describe your ideal candidate profile when it comes to their current or past titles, location, years of experience, or desired skills.

Utilize the best practices described earlier to make this as rich and detailed as possible. This prompt is the most important piece of the AI Sourcing search. Using our same prompt example from above, you should strive to have something at least as detailed as:

"I need a Senior ML Engineer for our autonomous vehicle startup (Series B, 100-300 employees) building perception systems for self-driving trucks. Our ideal candidate has experience with: computer vision in production environments, sensor fusion (camera + LiDAR + radar), real-time inference at edge, and automotive safety standards (ISO 26262). They should come from companies that have deployed ML in safety-critical applications."

You can also add ideal profiles as part of your prompt. This would be someone that already exists within your Gem CRM that closely matches the type of candidate you are looking for, giving the AI a solid example of the skills, title, and work history you'd like to see in your results pool.

Once you’ve filled out these fields, select Next to proceed.

Step 2. Define talent pool

Use filters like job title, location, and years of experience as hard criteria to shape your talent pool; the set of eligible profiles Gem AI will evaluate, score, and rank. The prompt you've entered in Step 1 will create a large group of potential candidates that match your open role. You will use these filters to help define any must-have qualifications, such as past experience, education, or specific job titles, ensuring only candidates who meet your requirements are considered. You can apply as many or as few filters as needed to refine your search and focus on the most relevant prospects. Filters include:

• Job titles: Filter by “Is/Contains,” “Current title only,” and use Boolean “OR/AND” to include multiple title variations within a single query. Using multiple titles here, such as "ML Engineer - Perception", "Computer Vision Engineer", "Autonomous Systems Engineer", "Robotics Engineer", continuing with our ML Engineer example, will give the AI room to surface more potential candidates.
• Location: Filter by city, metro, state/province, and country, and exclude specific regions using a “NOT” option. This is especially useful for roles that require either hybrid or in-person commitments. Because these will be passive candidates, you may want to avoid reaching out to people who are not within commuting range of your offices.
• Experience & tenure: Months at current company and overall years of experience (custom date range).
• Companies: Select from default presets and team presents, filter by “Current company only,” include companies from specific industries, of a certain size, backed by certain investors, and allow Gem to find similar companies to those you have already selected.
• Education: Filter by school, degree, and field of study; use Boolean “OR/AND” to include multiple degree and field-of-study variations within a single query.
• Keywords: Get even more granular by filtering for specific keywords. This is where you would want to list any terms that indicate the candidate would have the skills that are a hard requirement for the role, such as "computer vision + automotive", "sensor fusion", "real-time ML", or "safety-critical systems" for our example ML engineering role.
• Spoken languages: Use Boolean “OR/AND” to include multiple language variations within a single query
• Last contact: Use the dropdown to filter by contact history (e.g., in the past three months, none in the past three months, never, etc.) for prospects from your Gem CRM or past applicants. This should align with the rules of engagement your team uses to avoid contacting the same candidates too often or too soon after your last outreach to them.
• Exclude Projects: Use the dropdown to filter out candidates who belong to specific Gem Projects.

As your filters are applied, you can see the approximate size of your defined talent pool by selecting the See filter impact button. This will show you how many candidates Gem estimates each filter will remove from the pool of the net-new prospects your search would return, existing prospects within your Gem CRM, and past applicants. If your filter is excluding too many or too few candidates, you can refine it and check the estimated impact.

Note: Gem AI performs best with a well-defined talent pool. If your search includes more than 20,000 candidates, you may encounter less relevant results. Adjust filters strategically to maintain a strong balance between specificity and reach.

For a deeper dive into filtering best practices, see Gem AI Sourcing: Filters Best Practices.

Step 3. Add criteria

Gem AI analyzes your talent pool based on the criteria you set, ranking all of the candidates that matched your prompt and were not excluded by your filters. You can:

• Copy and paste must-have and nice-to-have qualifications directly from your job description
• Manually enter key criteria to refine how candidates are ranked

These criteria should be concise, single-line statements about your ideal candidate similar to what you might include in a job description, such as experience with a specific technology, process, or application.

For tips on optimizing your criteria, see Gem AI Sourcing: Scoring Criteria Best Practices.

Gem will also suggest criteria for you, based on your prompt and filters, and provide a few preset criteria for you to choose from. As you add criteria, the sample pool shown on the right side of the screen will update, providing a preview of the candidates you can expect when you begin the search.

Marking your criteria as priority by selecting the star next to each one will change their weighting in the candidate's overall ranking within your results. You can also select the X to remove any criteria that have an undesired impact on the sample pool.

Step 4. Summary & settings

Before Gem AI begins your search, you’ll be shown a summary of your search criteria. Name your search and select a default Project to associate with the search if you'd like. Then review the details, make any necessary adjustments to the filters or criteria, and select ✦ Activate Search to proceed.

Review AI Sourcing search results

Once Gem AI completes the sourcing process, you can start reviewing candidates. Go to the AI Sourcing page and select Start Reviews next to your search to begin evaluating prospects.

On the left side of the screen, you’ll see a list of prospects, ranked by how well they match your search criteria

In the center, the selected prospect’s profile is displayed, along with their Gem AI match score (e.g., “100% Match,” “65% Match”), and their work experience (with any relevant keywords that match your criteria or prompt highlighted), helping you assess fit at a glance. From a profile, you’ll be able to see other information, including:

• Resumes (if applicable)
• Activity feed: a full history of past interactions with the prospect and a space to add public or private notes
• Scorecards (if applicable)
• Applications (if applicable)
• “Last updated” icon (⟳): shows how recently the profile’s data was refreshed

On the right, the Gem Extension allows you to take actions such as adding candidates to Projects or Sequences, and sending one-off messages. Here’s where you’ll also mark the candidate as Yes, No, or Maybe—making it easy to quickly review and move on to the next prospect. (Use the → arrow beside the prospect’s name to navigate to the next candidate.)

Note: Enabling Feedback mode allows you to specify why you're passing on a candidate. This input helps Gem optimize future search results. It is not used for AI model training.

If your results are providing too many or too few candidates, our results troubleshooting guide can help you improve your sourcing results.

Once you’ve finished reviewing the prospects Gem AI has surfaced, you can take bulk actions on the candidates you’re interested in:

1. Select the Yes tab at the top of the page to view all approved prospects
2. Select Bulk action at the top of the list
3. Choose from:
• Bulk add to Sequence to kick off outreach campaigns
• Bulk add to Project to organize candidates for future engagement

This allows you to efficiently move top prospects into your hiring workflow and start connecting with them right away.

Edit your search

If you need to make changes to your search, you can do so from the AI Sourcing page. Locate your search by name, then select “View Search Details.”

Your filters and criteria will be displayed, just as they were in the original search. Simply the option to add, edit, or remove criteria. When complete, select ‘Apply changes’ to restart the search.

Tips for AI Sourcing searches

The following are some additional tips for improving or expanding your AI Sourcing search results.

• Consider filtering by Metro instead of City. This expands the AI's available pool to include nearby cities and also helps it understand which city you are targeting, in cases where multiple large cities have the same name.
• Unchecking Email finding will also expand the pool, allowing the AI to surface candidates that do not have email addresses already.
• Allowing the AI to Include partial matches will also expand the pool.
• When creating a job title filter, include multiple title variants. Gem will also suggest additional variants based on your prompt and selected filters.
• Use Exclude job titles as well to help make your search results more precise. If you are looking for an engineer with front-end experience, you may want to exclude "back-end developer" from your titles, for example.
• Consider whether the titles you are filtering for much be their current title. Consider if a candidate with a different title who had previously done the role you are searching for would still be a good fit.
• Include your company story in your AI Sourcing prompts. Including information like your stage, mission, product details, or core values will help the AI highlight candidates who look like good fits for your company, not just the role, based on their profile content.
• Review our additional filter tips in this Help Center article. You can also watch a recorded webinar with even more information on making the most of Gem's AI Sourcing offering.

Have any issues or questions on this topic? Please feel free to contact your dedicated Gem Customer Success Manager directly or our Support team at support@gem.com.

• Requirements
• Get into the AI mindset
• Write prompts instead of keyword filters
• Use punctuation to clearly separate thoughts
• Be detailed and specific
• Understand how AI interprets requests
• Tell the AI what to avoid
• Keep practicing: Adjust and re-run your search
• Set up an AI Sourcing search
• Step 1. Set up your first search
• Step 2. Define talent pool
• Step 3. Add criteria
• Step 4. Summary & settings
• Review AI Sourcing search results
• Edit your search
• Tips for AI Sourcing searches