Recruiting outreach with AI: 4 steps to test it on one open role
With AI recruiting outreach, tie the candidate's career to the role, keep decisions human, and track to the hire.
Key points
- Tie career to the role in one sentence
- Leave unknowns to ask in the interview
- AI drafts; people decide and send
- Track to the hire, not just replies
Contents
Write “why you” by tying their career to the role
A&A perspective
Recruiting outreach is when a company contacts candidates directly and invites them to apply or talk. The core of the message is “why you”: one sentence on which part of their career you looked at and which job you think it fits.
From the sources
Indeed, the job search site, sends job seekers invitations to apply for jobs that seem to match their resume and profile. In an August 2024 article on its own work, Indeed says it used AI to add an explanation of which part of a person's background or previous work makes them a good fit for the job.
From the sources
In the same article, Indeed's CEO says that however accurate the match between job seekers and jobs, being able to explain why a job is recommended is key to making a recommendation system work.
A&A perspective
Automate & Augment (A&A), the company writing this article, sees that what a small company can copy is not polished wording. It is the form: pick one fact from the candidate's career and one line of work from the job posting, and join them into one sentence that says “given this career, we think you fit this job.”
Hypothetical example
For example, imagine a 20-person company that sells business equipment and is looking for a salesperson to visit its existing customers (a hypothetical example). If a candidate's career says they “regularly visited business customers and took repeat orders,” you can write, “We think that experience overlaps with the work of visiting our customers.”
A&A perspective
If you can't find two lines to join, stop before polishing the message. A&A finds that, more often than not, the job posting doesn't describe the work, or you are contacting the wrong people. If you let AI fill the message with praise that isn't in the career, you end up explaining the job again from the start in the interview.
Leave what the career can't show to ask in the interview
A&A perspective
What the career doesn't show, don't assume in the message; write it as something to ask in the interview. Put what you can tell into the reason and what you can't into a question, and one message carries both what you expect and what you want to confirm.
From the sources
Indeed also has a feature employers use when searching for candidates. It summarizes each resume and suggests why the candidate could fit the role. Indeed says it also points out possible gaps in their experience.
A&A perspective
A&A reads this as a design that puts the reasons for a fit and the possible gaps on the same screen. Putting the two side by side in an outreach message, too, keeps you from overselling the candidate.
Hypothetical example
In the hypothetical example, the career doesn't tell you whether the candidate can win new customers on their own. So you add one line to the message: “In our first conversation, we'd like to hear how far you have handled new customers.”
A&A perspective
On the other hand, don't write a line that asserts something the career doesn't show, such as “We're confident you can take on new business development.” Promising it before the interview confirms it leads to a mismatch about the job after they join.
A&A perspective
The interview can then start from the questions the message raised. Because what the message said connects to what the interview asks, you don't have to explain the job again from the start.
A&A perspective
Limit what you give the AI to the work history the candidate submitted for hiring and the job description your company has decided. Don't use hobbies or private life as material to guess whether someone suits the job.
A&A perspective
If the draft keeps track of which career line was joined to which job line, the owner doesn't have to judge the overall impression of the message. It is enough to check whether the link holds up.
| Task | Who owns it | What a person checks |
|---|---|---|
| Write the work you will hand over in the job posting | The owner | Whether the posting describes the work you will hand over |
| Choose the candidates to contact | The owner or the person in charge of hiring | Whether two lines joining career and job can be found |
| Draft the reason and the interview questions | AI | Whether the two joined lines hold up (the owner checks) |
| Check the recipient and content, then send | The owner or the person in charge of hiring | Whether pay and working conditions match the job posting |
| Assess the interview and make the hiring decision | The owner or the person in charge of hiring | Whether the questions raised in the message were checked in the interview |
Let AI draft only; people decide and send
A&A perspective
When you test on one open role, give AI only the draft of the reason and the questions to ask in the interview. People decide who to contact, the hiring decision, the recipient and sending, and the interview assessment.
From the sources
Anthropic, a company that develops AI, recommends in its published guidance starting with the simplest approach when building with AI and adding complexity only when needed. It also says that for many uses, refining a single request to the AI is usually enough.
From the sources
It says systems that keep deciding and acting on their own bring higher costs and the risk of errors compounding.
A&A perspective
A&A applies this to hiring and stops AI's role at “ask once and get a draft back.” If AI makes the hiring decision and gets it wrong, people can no longer trace which career it read and how. A draft, the owner can fix by looking at the two lines it joined.
A&A perspective
To test it, the next few messages for one open role are enough. Paste the work section of the job posting and the career section of the candidate's profile into an AI such as ChatGPT, and ask: “List two lines that can be joined, and what the career doesn't show.” The owner then checks whether the two lines in the draft hold up.
A&A perspective
The only part that changes per person is where their career overlaps with the job. Copy pay and working conditions for everyone from the same job posting. If the way conditions are written changes from person to person, you get “this isn't what I was told” after they join.
Don't stop at reply counts; track to interview and hire
A&A perspective
Judge the test not by the number of replies alone, but by following each person through to the interview and the hire. Even if replies go up, the owner saves no time if the job has to be explained again in the interview.
From the sources
Indeed says it compared the old invitations with the ones that had AI-written explanations, in several stages over several months. It reports that started applications rose by 20% or more. It also reports gains of 13% or more at stages further along than applying, such as interviews and hires.
From the sources
Indeed's product lead says in the same article that the invitations grew from 1 million to 20 million messages a day.
A&A perspective
What A&A takes from this case is not the numbers but the habit of measuring past the application. The scale is too different to use them to forecast results for a 20-person company. The numbers come from Indeed's own article and have not been checked by anyone outside.
A&A perspective
If you send only a few messages a month, comparing rates won't show a difference. So for each person you contact, write one line: “the reason you joined,” “reply,” “whether the job had to be corrected in the interview,” and “hired.”
A&A perspective
If someone who isn't a fit turns you down early, that is not a failure. Count it as a record of both sides avoiding an interview neither needed.
A&A perspective
Some say that with a small team it's faster for the owner to write each message. Even so, a record lets you look back at which reasons brought people to an interview and where the conversation went off track. Decide from that record whether to keep the AI drafts or fix the job posting first.
Summary
A&A perspective
Write “why you” by joining one fact from the candidate's career with one line of work from the job posting. Don't assume what the career doesn't show; ask it in the interview. Keep AI's role to drafting, and let people make the hiring decision and send. Don't stop at reply counts; record each person through to the interview and the hire.
To write recruiting outreach with AI, first check, before polishing the wording, whether you can write a reason that joins one fact from the candidate's career to one line of work in the job posting. Keep AI's role to drafting the reason and the interview questions, and let people make the hiring decision and send. Don't stop at reply counts: record each person through to the interview and the hire, and use that record to decide whether to keep the AI drafts or fix the job posting first.
FAQ
Can I let AI choose candidates and make the hiring decision?
People make the hiring decision. AI only drafts the reason for candidates a person has chosen. That way, when something goes wrong, a person can see which career was joined to which job and fix it.
Will writing outreach with AI bring more replies?
We can't say it will. Indeed's numbers are a large site's own report on sending more than a million invitations a day. A small company should look less at reply counts and more at whether the interview stayed on the same page.
Compared with writing it myself, what gets easier?
A&A thinks the bigger gain is not less writing but a narrower thing to check. The two reason lines and the interview questions come in the same form every time, so the owner only has to check whether those two lines hold up. The links are kept, so they also give you something to look back on later.
Sources & editorial note
Primary pages read for this article. Publication dates below belong to the sources; access dates record our research.
- How Indeed Uses AI to Provide Better Matching Context for Job Seekers
Indeed · 2024-08-14
Accessed 2026-10-22 - Building effective agents
Anthropic · 2024-12-19
Accessed 2026-10-22