How to Review Whether an X Post Mentions Your Product

Numbered review actions

1. Define your product identity record before examining candidates. List your exact product name, known aliases, official public website, official handle, and specific product terms. Keep these rules reader-created and separate from Actor verification fields.
2. Begin with supplied known candidate status URLs and the current Task input. Replace the saved startUrls list with one intended known candidate status URL for your pilot test.
3. Check the native source ID or URL, text, and status or time. Review whether the text is sufficient. Images, quote bodies, and linked pages may contain context the row does not expose. If required context is missing, inspect the original public source separately or mark the item as needs review. A nonblank text string does not prove complete context.
4. Record the literal match separately from the entity decision. Name the matching word or alias and the evidence passage. An ordinary-word alias such as River can match discussion of waterways. Do not infer that the author means the product from capitalization, a keyword match, a blue badge, or a large follower count.
5. Use a written context rule. An exact checked product URL, official handle, or distinctive product name in the actual passage can support the intended entity under that rule. A name-only or pronoun-only reference can remain ambiguous. Read the surrounding sentence carefully.
6. Assign supported product reference, ordinary word use, or needs review with a concrete reason. A supported reference means the reviewed passage refers to the stated product. It does not mean the statement is true, positive, made by a customer, or approved by the brand. Sentiment and claim truth are separate analyses.
7. Keep source evidence, rule version, reviewer, time, and unresolved context in the small queue. Repeated export rows for the same post do not become separate mentions. Conflicting source copies stay unresolved. A source failure is not proof of no mention. Do not silently replace missing text with an earlier version as current.
8. Hand only the reviewed relevance decisions to your next process. If later building an alert tool, separate the collection, relevance, and delivery receipts. No notification integration was tested here. This guide ends with a source-backed relevance queue and a concrete unknown case, not a promise that alerts cover every post.

OUTPUT and field structure

The X Post Metrics 1.0.4 Actor accepts `startUrls` containing known public status URLs. The exact native fields returned for each post are `inputUrl`, `postUrl`, `postId`, `postType`, `publishedAt`, `text`, `profileUrl`, `profileId`, `profileUsername`, `profileName`, `status`, `statusReason`, and `observedAt`.

Note that `text` can be null, and the `postMetricStatus` array describes metric availability and precision rather than text completeness or semantic relevance. The run summary outputs aggregate counts (`inputCount`, `datasetRows`, `usefulRows`, `billableRows`, `failedRows`, `statusCounts`, and `stopReason`). A single-post control run demonstrates the run summary output structure:

Plaintext

inputCount: 1
datasetRows: 1
usefulRows: 1
billableRows: 1
failedRows: 0
statusCounts: {"ok": 1}
stopReason: "inputs_exhausted"

In this control run, the target returned nonblank text of 200 characters. This confirms text collection support on one known public post, but it does not guarantee text presence across all posts, images, quote bodies, linked pages, or historical original wording. Unavailable text results in an unknown keyword match state, not false.

Worked example queue

All teaching text and examples below are original fiction.

Reader-created product identity record

• Product Name: RiverQueue
• Aliases: River (case-insensitive keyword rule: lowercase "river" matches alias "River" without altering raw source text)
• Official Handle: @RiverQueue
• Official Domain: riverqueue.example
• Checked Premise / Source: Reader-verified brand identity record (Rule v1.0, checked 2026-10-13 by Reviewer)

Candidate relevance queue

Four distinct candidate posts produce one supported reference, one ordinary-word use, and two review cases. Do not call this a whole-X accuracy rate, false-positive rate, or share of voice, because no known ground-truth sample was tested. The fictional public product link is a manually checked identity premise, not an Actor lookup outcome.