> For the complete documentation index, see [llms.txt](https://doc.autoreach.tech/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.autoreach.tech/finding-leads/lookalike-audiences.md). # Lookalike Audience Discovery Find high-quality prospects by identifying influencers and accounts whose audiences match your ICP. AutoReach uses AI-powered web search to discover seed accounts, then extracts and scores their audience as leads (followers on X and Instagram, post commenters on LinkedIn). ## How It Works Lookalike Audience Discovery follows three steps: ### 1. Discover Seed Accounts AutoReach uses AI-powered web search to find accounts whose audiences overlap with your target market. Based on your offer, it searches for relevant accounts, verifies them against live platform APIs, and returns a curated list of seed accounts. Discovery works on X, LinkedIn, and Instagram from the same interface. **What it searches for:** The AI searches across multiple audience categories: * **Industry publications**- newsletters, media outlets, trade publications * **Professional communities**- groups, forums, community accounts * **Vendor/competitors**- competitor accounts and adjacent vendors * **Thought leaders**- executives, analysts, and domain experts * **Vertical-specific**- niche accounts specific to your industry If your offer specifies preferred locations, the AI prioritizes accounts with audiences in your target regions. ### 2. Extract the Audience Once seed accounts are saved, their audience is extracted through the standard extraction pipelines: * **X seeds:** Followers are extracted and queued for enrichment * **LinkedIn seeds:** The commenters on the seed account's posts are mined (a capped number of posts per run) and queued for enrichment * **Instagram seeds:** Followers (or following) are extracted and queued for enrichment In the **manual flow**, you discover seeds from the Lookalike Audiences page, save them to your library, then separately trigger extraction via the "From a Lookalike" option. Under **Autopilot**, extraction is fully automatic after discovery. ### 3. Score and Add Leads Every extracted profile goes through the standard enrichment and scoring pipeline. All extracted profiles are added as leads and scored by buyer intelligence like any other source. ## Search Options When starting a lookalike search, you configure: * **Offer**- select the offer that provides ICP context, location, and competitors * **Channel**- choose X, LinkedIn, or Instagram * **Platform account**- select your connected X, LinkedIn, or Instagram account * **Profile URL** (optional)- enter a profile URL for profile-based discovery instead of offer-based **Two discovery modes:** * **By Offer:** Select an offer- finds seeds matching your offer's ICP, domain, and audience * **By Profile URL:** Enter a profile URL- finds accounts similar to a specific person ## Available Actions From the Lookalike Audiences page, you can: * **Estimate cost**- preview the estimated cost before running a search * **Search for seeds**- discover seed accounts with real-time progress updates * **Save accounts**- save one or more discovered accounts to your library * **Browse saved accounts**- view your saved accounts with pagination and filtering * **Delete accounts**- remove individual or multiple saved accounts ## Automatic Rotation Under Autopilot When Autopilot is enabled, seed accounts rotate automatically: 1. When a seed is exhausted, Autopilot discovers **new** seed accounts via web search 2. New seeds are saved and extraction begins automatically 3. Previously used accounts are tracked to avoid duplicates 4. If the AI can't find new relevant accounts, it broadens its search to find lesser-known accounts 5. If discovery stalls, you receive a notification This means Autopilot actively discovers fresh seed accounts- it does not just re-extract from existing ones. ## Cost Estimation Before running a search, use the cost estimation feature to preview estimated costs for the web search AI calls. ## Best Practices 1. **Mix account types.** Combining publications, communities, and individual thought leaders produces more diverse and well-rounded audiences. 2. **Prioritize engaged audiences.** Accounts with high engagement rates tend to have more active, reachable followers. 3. **Use multiple platforms.** Run lookalike discovery across X, LinkedIn, and Instagram to maximize coverage. The extraction pipelines differ but all produce scored leads. 4. **Monitor lead quality.** Track response rates from lookalike-sourced leads. If quality drops, your seed accounts may need adjustment. 5. **Combine with other methods.** Lookalike audiences complement content search and people search. Use them together for best coverage. 6. **Let Autopilot handle rotation.** Manual seed management works, but Autopilot's automatic rotation with escalating diversity prompts keeps the pipeline flowing without intervention. ## Example Workflow **Offer:** "API platform for payment processing" **Step 1:** AutoReach web-searches for fintech publications, developer community accounts, API-focused thought leaders, and competitor ecosystems on X. Returns 10 verified seed accounts with 1,000+ followers each. **Step 2:** You save the 6 most relevant seeds. Follower extraction begins- 200 followers per seed, totaling \~1,200 profiles. **Step 3:** Each follower is enriched and scored. CTOs, engineering leaders, and technical founders who match your ICP surface as active leads. The rest land in Monitor or Poor Fit. **Under Autopilot:** When the 6 seeds are exhausted, the scheduler automatically discovers 10 new seed accounts and creates fresh extractions. This repeats continuously. ## Troubleshooting **Not finding relevant seed accounts?** * Make sure your offer description includes relevant domain keywords * Check that your target audience is active on the platform you are searching * Broaden your offer's pain points or audience definition to give the AI more to work with * Try profile-based discovery with a known relevant account **Low ICP match rate on extracted followers?** * Review whether the seed accounts truly align with your ICP * Refine your ICP definition, particularly pain points and target audience * Try targeting more niche, specialized accounts rather than very broad ones **Seed accounts getting exhausted?** * This is expected. Previously extracted followers are not re-extracted * Under Autopilot, new seed accounts are discovered automatically * You can also add seed accounts manually to increase flow **Getting prospects with missing profiles?** * Some followers do not maintain active profiles. These are filtered out during enrichment and are not counted against your lead totals **Autopilot stall notification?** * After 3+ consecutive misses finding new seeds, you'll receive a notification * Review and update your offer description to give the AI fresh context * Consider switching channels (X, LinkedIn, or Instagram) if one platform is exhausted ## Next Steps * [**Continuous Operations**](/autopilot/continuous-operations.md): Automate lookalike discovery with Autopilot's daily expansion * [**Enrichment Pipeline**](/enrichment/pipeline.md): Understand how extracted followers get enriched and scored --- # 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://doc.autoreach.tech/finding-leads/lookalike-audiences.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.