> For the complete documentation index, see [llms.txt](https://help.impact.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.impact.com/brand/what-would-you-like-to-learn-about/platform-features/review-partner-applications/recruitment-agent/write-effective-agent-instructions.md). # Write Effective Agent Instructions Your Recruitment Agent works from a single set of instructions you control. This article explains what the Instructions do and how to write and refine them. To set up the agent itself, see [Recruitment Agent](/brand/what-would-you-like-to-learn-about/platform-features/review-partner-applications/recruitment-agent.md). {% hint style="warning" %} **Beta:** The Recruitment Agent is currently in Beta and is available to a limited number of brands. {% endhint %} ## **Agent Configuration** Agent Configuration is the core context the agent uses to evaluate prospective partners and inbound applications. You write it as plain-text instructions, and it is the single source of truth for how the agent reasons about partners, both for Discovery and for inbound application evaluation. The accuracy of your instructions directly determines how well the agent performs. Vague or incomplete Instructions produce vague or unpredictable results. Specific, detailed Instructions produce consistent, on-brand matches. When the agent evaluates a partner, your instructions are the only context it uses. It doesn't draw on any information you haven't included, so add as much relevant detail as you find useful. When you activate the agent, impact.com uses AI to generate a starting set of instructions based on: * Your brand URL * Your Marketplace Profile, including verticals, business models, service areas, audience, and brand size * Brand and audience intelligence, including audience countries, demographics, and top content categories ## **Write your instructions** You have full control over your instructions. You can: * Edit any part of the generated text. * Delete content you don't need. * Rewrite everything from scratch in your own words. * Add bullet points, paragraphs, or sentences anywhere. * Include any instruction you would give to a human reviewer. You can write up to 13,000 characters, in any format. You can write in prose, use bullet points, or organize the text into sections. Anything you would normally check when manually evaluating a partner is worth including. For example: * Filters you usually apply when browsing the Marketplace, such as verticals, regions, partner size, and business models. * Data points you check on applications, such as audience demographics, content categories, and social engagement rates. * Partners that are not a good fit, such as those whose content, audience, or audience location doesn't align with your brand. Be explicit about exclusions. If there are partner types you would never accept, state them directly rather than relying on the agent to infer them. The more specific your instructions, the more selective the agent is: *Discover* adds fewer prospects, and *Evaluate Applications* accepts fewer partners. Tell the agent what to do when an application does not clearly meet your criteria. For example, you can write rules that leave certain applications pending so the agent takes no action and you review them manually, rather than having it reject them. Write criteria the agent can verify against real partner data. The agent checks each partner's profile against your instructions, so a requirement that is not visible in a partner's data can count as not matched, and a broken exclusion can be enough for the agent to skip or reject a partner.
View an example We are ACME Corporation, an outdoor gear and apparel brand selling hiking, camping, and trail-running equipment across the United States and Canada. We are looking for content creators and review sites focused on hiking, camping, and outdoor adventure, with audiences primarily aged 25 to 45 in North America. Our strongest partners produce gear reviews, trail guides, and how-to content on YouTube, Instagram, and personal blogs. We work best with medium to large partners that have established audience trust and consistent engagement. Do not add or accept: * Partners who promote fast fashion * Partners with audiences primarily outside North America * Partners with no outdoor or adventure focus * Partners who have posted content promoting unsafe wilderness practices
Example instruction lines Use clear, specific lines the agent can match against partner data. For example: * Core niche: lead with one main keyword that best represents your ideal partners, such as "Primary niche: clean beauty" or "Core niche: jeans fashion." * Content focus: "Content focus: denim outfit ideas, jean fit guides, and jeans reviews." * Partner location: "Partners based in the US and UK." * Audience location: "Audience primarily in the US, Canada, and Australia." List partner location and audience location separately, as they are different criteria. * Audience demographics: "Audience primarily women aged 18 to 34." * Channels: "Prefer Instagram Reels and YouTube Shorts creators." * Size and reach: "At least 100,000 Instagram followers" or "50,000 or more YouTube subscribers." * Engagement: "Prefer partners with strong engagement, around 2% or higher." * Business type: "Include creators, review content, and editorial partners." * Exclusions: "Do not accept coupon-only sites" or "Do not accept audiences outside our core market countries." To broaden your results, remove or loosen your most restrictive lines, such as narrow country lists, strict follower thresholds, or very specific channel requirements.
{% hint style="warning" %} **Warning:** Edits to your instructions apply only to future Discover runs and future application decisions. The agent never reopens decisions it has already made. To revisit a past decision, do so manually from the Applications screen. {% endhint %} ## **Test and refine your instructions** Your instructions improve with iteration. Use a small test run to see how the agent interprets them before you commit to full-scale recruitment. * **Test with Discover:** set the Discover sub-agent's **Maximum Run Limit** to a small number, such as 10, then select **Run Discover**. Review the prospects the agent sources, and refine your instructions based on whether they match what you expected. Repeat until the results are consistently on-brand, then raise the limit. See [Track Recruitment Agent Results](/brand/what-would-you-like-to-learn-about/platform-features/review-partner-applications/recruitment-agent/track-recruitment-agent-results.md). * **Test with Evaluate Applications:** run the Evaluate Applications sub-agent in **Autopilot** so the agent acts on applications. As soon as your first applications come in, review the agent's decisions closely and tweak your instructions based on what you see. If you would rather check the agent's decisions before they take effect, run the sub-agent in **Copilot.** The agent will then assign each application to a group as a suggestion, and allow you to make the final decision from the *Applications* page. When the agent's suggestions consistently match your decisions, switch to **Autopilot**. ## Instructions FAQ Common questions about writing and refining your instructions.
How do I improve Discover results fastest? Lead with one clear core niche phrase, then add explicit lines for partner location, audience, partner type, channels, and exclusions.
Should I list partner location and audience location separately? Yes. "Partners based in \[country]" and "audience in \[country]" are different criteria, and both are useful.
Can I use exact numbers? Yes. Numeric ranges, such as a minimum follower count, are very helpful when you have specific goals.
Why didn't a partner I wanted pass automatically? Usually the criteria were too strict, contradictory, or not visible in the partner's data. Loosen or clarify the relevant lines, then run again.
How often should I update my instructions? Update your instructions any time your goals change, for example a new market, product line, channel strategy, or audience focus. Even small edits can shift your results, so update whenever your targeting needs to change.
--- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://help.impact.com/brand/what-would-you-like-to-learn-about/platform-features/review-partner-applications/recruitment-agent/write-effective-agent-instructions.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.