如何將 AI 引擎引用的品類頁面整理成外聯清單
Which pages do AI engines cite for your category? Turn the sources into an outreach to-do list
文章示範如何使用 Apify 上的 AI Brand Monitor，收集 AI 引擎回答所引用的頁面，並將來源整理成外聯待辦清單。流程以 4 條品類問題、Perplexity 和每題 3 次抽樣為例，成本為 $0.60；建議忽略穩定度低於 0.5 的來源。再按來源網頁類型安排上架、參與討論、提案或競品分析，並可在數週後重跑，比較新增和失去的來源。
Disclosure: I built the Apify Actor in this post. Written with AI assistance.
Knowing whether an AI engine mentions you is the thermometer. The useful part is the list underneath: the exact pages the engine read to build its answer. Ask "best CRM for small business" and the sources might be a review site, a Capterra listing, a Reddit thread and two vendor blogs. You cannot edit the model, but you can get onto those pages.
AI Brand Monitor on Apify returns, for every answer, the full source list with URL, page title and the quoted snippet. Here is how to turn it into a to-do list.
Step 1: collect the sources (5 minutes)
- Open the Actor page. The form is prefilled.
- Queries to track: enter 4 category questions your buyers type, such as "best [category] for small business" or "[category] alternatives".
- Engines to query: Perplexity only (remove the prefilled ChatGPT; Perplexity needs no API key).
- Samples per query: 3. The same question is asked three times, and each source gets a stability score.
- Add your domain and 3 competitor domains, then click Start.
Cost: 4 queries x 1 engine x 3 samples = 12 records x $0.05 = $0.60. That also fits the 12 Perplexity calls per run included on a free Apify plan.
Step 2: keep only the sources that hold
AI answers are not deterministic. A domain cited once in five runs is noise; a domain cited every time is the engine's settled answer. When you set more than one sample, the Actor writes a stability record per query and engine, with each domain's occurrences, samples and stability (1.0 means cited every time). The recommended filter: ignore anything below 0.5.
Step 3: build the to-do list
Open the run's dataset and use Export to download a CSV, or read the citations field of each record. Make one spreadsheet row per stable domain, with the page URL, the title and how many of your queries cite it. Then sort by count and assign an action to each row:
- Review site or marketplace: get listed, or collect reviews there.
- Reddit or forum thread: join the conversation with a useful answer.
- Vendor or comparison blog: pitch your inclusion, or publish your own comparison.
- A competitor's own page: study what it covers that yours does not.
Check the competitorsCited field on each record too. It shows which rivals are cited on the same question, so you know who you are being compared with.
Step 4: check whether it worked
Save the input as a task and run it again in a few weeks. Pass the earlier run's dataset ID as previousRunDatasetId and the Actor writes a delta record with the sources you gained and lost per query.
Why this beats guessing
The source list is the best available lever for getting cited. Your first list is a starting map, and the repeat run shows which moves counted.
Get your category's source list: one run, about 60 cents, and a list of pages to get onto.
來源：DEV Community · dev.to